MIT reported in August 2025 that 95% of corporate GenAI pilots are failing. The cause is not the quality of the models. It is that the tools are being pointed at workflows the systems were never built to read.

The media industry is about to make that number worse. In a typical media supply chain, the work that controls, integrates, and orchestrates the actual workflow lives in the heads, runbooks, and habits of the people executing it. That is the control layer this piece is about. It has been holding the supply chain together for two decades because the foundation that should have replaced it never got built. I have been calling that pattern Human Automation. Years of it are not replaced by an Agent in a day, and this piece is about why that matters now.

Years ago I was on a team replacing an asset management system that had been in place for about fifteen years. I had also been on the team that built the original, which gave me one piece of context most don’t get: I knew what the system was for the day it shipped. We started discovery the way you start most of these, asking the people who used the system every day to walk us through their work.

The conversations went like this.

“So you ingest these images. Why?”
“Because that’s my job.”
“Right, but these less obvious metadata fields. Why these specific values?”
“Because the runbook says to.”

People were surprised when that work they’d been doing often had no meaningful relevance to the then current workflows.

The people answering those questions had inherited a workflow and lost the reasons for it years ago. But, they kept the “machine” running while the organization above and around them spent its time, effort and capital on finance systems and back-office platforms and told itself, year after year, that media operations was fine or would be addressed eventually.

This exemplifies the concept of Human Automation. It demonstrates why the current industry conversation about Agents and GenAI replacing operations work is heading toward the wrong conclusion. The human layer in a media supply chain is not a deficiency, it is the layer that controls, integrates, and orchestrates the work. The Agents we are about to point at it do not know that, because the data and process logic they would need does not yet exist in any system the Agents can read.

What I mean by Human Automation

Human Automation is what happens when an organization decides, explicitly or by default, that building the technology to connect two systems costs more than putting a person in between them. A fulfillment coordinator reading an order out of a sales system and typing it into a media order system because no integration exists. A mastering PM reading a release calendar and creating work orders by hand because there is insufficient demand tooling. People are used as the integration layer itself. It is not a failure of the operators. It is a default position the industry has chosen, repeatedly, when it had other options.

The phrase collides with three adjacent terms that mean different things. Human-in-the-loop is an automated system that intentionally inserts a person at decision points. Robotic process automation (RPA) is software bots that imitate what a human does inside a user interface. What Andy calls human-enhanced is the human contribution that turns generative output into a legally protectable work. Human Automation is none of those. The other three describe specific patterns of human-machine collaboration. Human Automation describes what fills the gap when the machine was never built at all.

The human is the control layer, and does not know it

In a Human-Automated supply chain, the people doing the work are often not executing a documented process. They are absorbing variation the system was never built to handle. The absorbing is invisible to whatever tool comes next. The system reports green because the human effort kept it green.

The control layer is also not uniform. There are at least two tiers of it inside any real operation, and they fail differently when an Agent shows up.

At one tier are the people who execute. What looks like a documented process is mostly habit, with a runbook capturing the visible part. Context across the team varies. Some have built real understanding over years and know when something looks wrong. Others know the actions but not the reasoning behind them. The asset ingest conversation at the top of this piece is from someone closer to that end. They have been running this habit long enough that it looks indistinguishable from a process.

At another tier are the senior staff that carry the context the systems were never built to hold. The Director of Mastering tracking talent relationships and preferences in a leather-bound notebook. The Mastering or Localization PM who knows from one phone call this morning that a vendor is over capacity this week or where to route that job for that recipient. None of that knowledge is in a system. It is the reason the operation keeps running. It is also the reason the next person in the role takes eighteen months to be effective.

The people inside Human Automation have not been passive. They have innovated continuously inside the space they were given, building spreadsheets that hold rules that were never formalized, writing runbooks for processes nobody at the system level would document, absorbing turnover and tool changes and merger integrations. The complaint about Human Automation is not that the humans were not capable. It is that they were capable enough to make the underlying problem invisible to the organizations that should have been solving those problems.

The three ways we did not build the necessary foundation

For decades, the industry has made the same decision in three different ways: do not build the underlying technology ourselves. The three patterns have never been sequential. Human Automation persists while outsourcing continues at scale. Outsourcing persists as DTC investment ramps. GenAI and Agents are being layered on top now, while the previous two are still doing the load-bearing work. All three are active, and all three have left the foundation work unfulfilled.

The first choice was Human Automation. Major media companies spent most of the licensing era avoiding the conclusion that they needed to effectively operate as technology companies. My own analysis of technology capital allocation during this period at a major media company across a fifteen-year window found less than ten percent of the spend going to content systems. The rest went to finance and back-office. Headcount filled the gap. The phrase that justified those decisions was “we’re content creators, not a technology company.”

Underinvestment was only part of the problem. Where money did show up, often around the direct-to-consumer (DTC) rollouts of the last five or six years, the partnership between operations and technology did not. Technology organizations sat at arm’s length from operations, treated by accounting governance and capital committees as project shops that delivered something with a beginning, a middle, and an end. I clearly remember fighting battles a decade ago with finance departments who insisted IT projects had to have a discrete close-out or they could not approve or track the investments. The shift from an “IT organization” to a “Technology Delivery organization” capable of co-owning living systems with operations came late, was ill-understood, and is often still incomplete. Operations teams that had spent decades being let down by IT had no reason to trust even these new tech teams now. The phrase that justified that resistance was “we don’t have time to change.”

The pattern that produces successful technology transformations in other industries is well known. Operations and technology co-own the work. The technology is treated as a living system rather than a discrete project. Finance funds capability rather than feature delivery. None of those preconditions has held in media operations long enough or consistently enough to produce a durable foundation. Until they do, the next wave of technology will land on the same hollow ground regardless of what the wave is.

The second was outsourcing. Media companies that did not want to build internal technology had another option, and they’ve used it for decades. They handed operational complexity to specialty vendors. Technicolor, Deluxe, Vubiquity, and many others absorbed work the media companies chose not to own. It let them delay the underlying investment by paying someone else to live with the consequence. The perceived ability to hold the vendor accountable was a fallacy. A failed delivery still failed, and legal recourse was monetary rather than operational. Then, in early 2025, Technicolor itself was liquidated globally after more than a hundred and ten years in business. Thousands of jobs gone. In-flight releases scrambling for substitutes. The remaining vendors struggled to fill the gap and the media companies finally started to look at in-sourcing. The industry’s most visible escape hatch partially collapsed, and the foundation it had been propping up was suddenly responsible for bearing the load.

The third is the one we are in now. Leadership that did not invest in or partner with internal technology, and that handed operational complexity to outside partners, is either reaching for GenAI and Agents as the next way to avoid the foundational work, or pretending those tools are not the future. GenAI and Agents are clearly the future of operational automation. Decades of underinvestment, broken partnership, and outsourced capability created the gap. GenAI and Agents will not close it on their own. The same leaders are making the same bet for the third time. While investment has definitely showed up in some organizations, the partnership and the trust deficit have not caught up. The phrase that describes this is “we have to keep the plane in the air while we change the engine,” which sounds responsible until you notice it has been the justification for not changing the engine for the entire life of the airline.

What the Agents are actually inheriting

At the execution tier, the Agent inherits the runbook but none of the reasoning behind it. It reads the runbook the same way the human did. Confidently, but without the comprehension built over years of habit. The difference is the human had an accumulation of feedback from years in the operation that occasionally surfaced the gap between the runbook and reality. A rejected asset, stalled workflow or a partner complaint. The Agent does not have that accumulation of feedback to build upon. It then executes that imperfect understanding at speed, and the throughput problem shows up at scale before anyone notices.

At the senior tier, the Agent inherits nothing. The leather notebook is not in any system. The Tuesday phone call about vendor capacity is not in any system. The routing knowledge in one PM’s head for ten years is not in any system. When that PM is asked to do something else with her time because an Agent now handles routing, the routing decisions stop being good.

The Agent is also inheriting workflows that suffer from the ongoing failure of two of their three former crutches. Internal Human Automation continues to be cut. The outsourcing market has just demonstrated that even the largest specialty vendor in it can disappear. Both buffers are weaker than they have been in recent memory. That is the operational risk nobody senior is talking about.

Andy framed the creative-side resistance around four questions about who benefits, who is retained, who decides, who owns. The operational parallel is similar: who carries the context, who notices when the Agent is wrong, and who is accountable when the green status stops being green.

The teams being asked to live through this transition are legitimately concerned. The fear of job loss is real. The distrust of the technology is earned, because the same leadership now pointing at Agents has been promising better tools for decades and not always delivering them. The commonly held instinct that the data has to be perfect before the tools are useful is technically wrong but emotionally honest, because the price of the tools being wrong has historically been paid by the operator, not by the organization that deployed them. Treating that resistance as irrational misunderstands the problem.

Step back, then tool it

The way out of this is not faster automation. It is the work the industry has historically declined to do. Step back and define what the process should be, not what it has accidentally become. Then build the foundational technology that lets demand, task, and state live in systems rather than in people. That foundation is concrete: demand management systems that make incoming work legible to automation, task management that holds in-flight state outside someone’s head, process orchestration that runs without the person who understands it in the room, and operational telemetry that measures what is actually happening rather than what the dashboard claims. Build the Agents on that foundation, where they can actually be useful, rather than on top of a fifteen-year-old runbook and systems nobody can defend.

The most credible Agent deployments I have seen recently look less like one Agent doing one job and more like what technologists have started calling an Agentic mesh: specialist Agents, loosely coupled but tightly integrated, orchestrated into a workflow. The pattern works for the same reason microservices worked. Smaller units fail in isolation, can be tested and replaced without rebuilding the whole, and surface their foundation requirements explicitly rather than hiding them inside one large black box. The mesh approach requires more from that foundational tech and process, not less. Each specialist Agent needs a system-legible view of demand, task, and state to do its job. The mesh does not solve the problem the runbook created. It makes the problem easier to see.

There is a rule I have repeated for years, and it gets sharper in this moment: you should never tool a bad process. Agents do not make that rule less true. They make it more expensive to ignore.

Start with the diagnostic

The diagnostic is clear and approachable. Inventory where your in-flight work state actually lives. Not where you wish it lived, not what the dashboard reports. Where would the next person taking the role have to look? If the honest answer involves a spreadsheet, an email thread, or a senior PM’s memory, that is the foundation gap, and no Agent will close it for you. For vendors and investors, the same question applied to the companies you work with. Their answers will tell you more about their readiness than any AI roadmap they share.

The wrong answer in 2026

Andy’s Human Layer series asks whether the person inside the work survives the transformation. Whose decisions are in the output. Whose face is in the frame.

The operational Human Layer asks a quieter version of the same question. Whose judgment was holding the process together. Whose context was the reason the dashboard reported green last quarter. Whose absence will be revealed by the next bad status report.

Both questions have the same wrong answer in 2026, and the industry is converging on it at speed. Pretend the human layer was not there. Pretend the runbook was the process and the leather notebook was inefficiency. Pretend the variation humans absorbed every day for decades was somehow being absorbed by the systems all along.

It wasn’t. It isn’t now. The Agents are arriving to try to fill a gap that should have been closed by foundational work the industry chose not to do. The people who held that gap closed for decades are the people the industry is calling inefficient and trying to move past. When the Agent is wrong, no one will be left to say why. When it needs direction, no one will be left to give it.

Years of Human Automation in the media supply chain are not replaced by an Agent in a day. They are replaced by the work the industry chose to gloss over, ignore, or not fund. Fix that and the Agent is the part that comes after.

Sources and further reading:

• Andy Beach, Human-Enhanced is the New Automation, Identity as Infrastructure, Observability as Default.

• MIT NANDA, “The GenAI Divide: State of AI in Business 2025,” reported by Healthcare IT News. Finding: 95% of enterprise GenAI pilots fail to deliver measurable ROI. Methodology: 150 leader interviews, 350 employee survey responses, 300 public deployment analyses.

• Lisanne Bainbridge, “Ironies of Automation,” Automatica, 1983.

• Bloomberg, “The Collapse of Technicolor.”

• Variety, “Technicolor Bankruptcy: How The Mill’s U.S. Team Launched ARC Creative.”

• McKinsey QuantumBlack, “How We Enabled Agents at Scale in the Enterprise with the Agentic AI Mesh.” For the multi-agent / agentic mesh pattern referenced in the “Step back, then tool it” section.

The Upfronts said the same thing

Netflix’s 2026 Upfront, delivered May 13 by President of Advertising Amy Reinhard, included the announcement that the company is testing AI agents that can autonomously manage, optimize, and purchase ads on the platform. The framing matters. These are not media-planning tools that recommend buys for human approval. They are agents that execute. Netflix described them as already in test with brands like DoorDash, Target, and TurboTax. The ad-supported tier itself crossed 250 million monthly active viewers, up from 190 million in November, a 32% jump in roughly six months.

Netflix was not alone. Fox unveiled AdStudio with agentic media planning and buying capability rolling out in the coming months. Disney and YouTube announced parallel moves. Industry coverage from Adweek and Hollywood Reporter described “agentic” as the buzzword that won upfront-week bingo this year. The pattern at the Upfronts is the pattern Google demonstrated at I/O, transposed from the consumer surface to the advertiser surface. The buyer states intent. The agent executes. The infrastructure stays inside the platform.

For media operators, the read is the same as Google I/O. The interface layer that mediated between buyer and supply, meaning the planner, the buying desk, the manual optimization workflow, is being absorbed into a model layer the platform itself controls. Brands hand over creative and objectives. The platform decides where, when, and at what price. Whatever you thought your demand-side leverage looked like, the assumptions just shifted.

Then there’s Google I/O

Shoreline Amphitheater sits outdoors. The main keynote happens inside the amphitheater itself, but most of the conference unfolds in temporary buildings and tents arrayed around it. The developer crowd was out in force this year, lined up at demo stations and side-stage sessions, genuinely excited. That excitement read as real, not staged. The marketing, on the other hand, was relentless. The word agents came up roughly every ten seconds across the keynote and the side stages. By the end of the morning it had stopped functioning as a word and started functioning as a chant.

This was my first I/O. I have been to Microsoft Build and to Apple’s WWDC, so the hyperscaler developer-conference circuit closed on my card this week. Of the three, I/O is the most consumer-pitched and the most dominated by a single product line. This year that line was Gemini.

For a developer in that crowd, the picture was straightforward. Google rolled out new components, APIs, model variants, a redesigned development platform, and agentic primitives that fit cleanly into a Google-centric stack. The toolkit got richer. The excitement was appropriate.

For a media operator, the more useful read is different: these were components, not operational solutions. They are parts a media business may eventually build with. They are not products that plug into a broadcaster, streaming platform, or content studio next week.

Underneath the marketing, three things did happen at Google I/O 2026 that a senior media operator should be paying attention to. They matter individually, and they matter more together. Read them together and you can see where the world’s largest video platform is restructuring, and where Google Search is making the same move on the open web. That is not a feature update. That is a re-platforming, and the dependency math just changed.

YouTube changed at the creation layer

The announcement most consequential to the media business was the integration of Gemini Omni into YouTube Shorts Remix and the YouTube Create app. Gemini Omni is Google’s new multimodal model, pitched as a system that takes any input and generates or edits video through conversation. Inside YouTube, that means a user can take an eligible Short, add a prompt and an image, and regenerate scenes. The model handles the technical work behind the scenes. The user states intent. The platform executes.

That is the creation layer of YouTube being absorbed into the model layer. The remix culture that already drives much of Shorts behavior has been there for years. What is new is that the editing, the visual work, and the audio adjustment have been moved from a human in a tool to a model in a pipeline. The barrier to making derivative content just dropped by a meaningful amount.

For an operator, two practical implications land immediately. First, if your content travels on YouTube as a distribution channel, the surface you optimize for changed. Derivative content created on top of original work will arrive faster and look more competent. Second, talent contracts written before this week did not anticipate Gemini-powered remix of likeness, voice, or performance. Those agreements need to be revisited before the volume of AI-edited remixes becomes a problem to resolve in retrospect.

Pair this with the other YouTube announcement made the same day. YouTube expanded its likeness-detection tool to creators aged 18 and up, allowing them to flag and request removal of AI-generated uses of their image. The defensive tool launched on the same day as the generative one. That is not a coincidence and it is not balance for its own sake. Google’s legal team is signaling where it thinks the lawsuits are going to land. For a media business holding rights in a content library full of identifiable people, that signal is forward information.

YouTube changed at the discovery layer

The second announcement that matters is Ask YouTube, a conversational discovery layer that shows where YouTube search is headed. Instead of typing a query and receiving a list of videos, the user describes what they actually want and receives an interactive structured response that mixes long-form, Shorts, and in some cases clip excerpts pulled from within videos. Follow-up questions extend the session. Search becomes a session. At launch the feature is available to U.S. Premium members aged 18 and up, through youtube.com/new, with broader rollout planned.

For more than a decade, YouTube discovery has rewarded creators who optimized for legible signals. Titles. Descriptions. Tags. Thumbnails. Watch time. Subscriber relationships. The whole creator-economy SEO playbook was built on the assumption that the platform’s discovery engine read certain inputs and weighted them in known ways. Ask YouTube introduces a layer where the model interprets intent and selects content, and the criteria are by design less transparent. Optimization in the new regime targets what the model considers responsive to a conversational query. That is a different game.

The Premium gate at launch matters for a specific reason. Google is testing this on the audience least likely to react badly to it, which is paying subscribers who are already opted into a lower-friction experience. When the feature rolls out beyond Premium, the people who feel the disruption will be the creators and media businesses that optimized for the previous regime. The window between now and that broader rollout is the window in which media operators can think about what their content looks like to a discovery engine that compiles rather than ranks.

Search is making the same move

The third piece, and the one that turns this from a YouTube story into a pattern, is what happened to Search. Google announced what it described as the largest overhaul of its search box in twenty-five years. The bar accepts longer queries, multimodal inputs, and contextual references. Results increasingly arrive as AI-compiled responses, sometimes as custom widgets generated on the fly. The widely circulated example from the keynote was a black-hole explanation rendered as an interactive visualization built specifically for the query, rather than a list of links to articles about black holes.

During The Verge’s I/O wrap-up podcast, one of the chat comments to the hosts captured the structural read more cleanly than the keynote framing did. The line was “this is how the web dies.” The Verge’s hosts then said out loud what the comment was pointing at. Google is taking the current web and remaking it on google.com. The user states a question and the platform composes the answer in-place. The link out becomes optional. For the user, that is convenience. For the link economy that supports nearly every advertising-funded media business on the open web, it is structural.

Open-web acquisition has been softening for publishers for years already. AI Overviews accelerated it. What Google announced this week extends the same logic across more of the search experience and adds generative interactive responses on top. A media business that depends on search-driven traffic to its own properties is now operating in an environment where the platform itself prefers not to send the user out. That preference is built into the product.

Read all four together

Across the Upfronts, YouTube creation, YouTube discovery, and Google Search, the same operator decision shows up four times in one week. The platform interprets intent and generates or compiles the response. The user, or the buyer, remains inside the platform. The link out, the click out, the route to another business, all become less frequent by design. This is consistent. This is deliberate. And it is happening across the most-used surfaces in media at the same time.

The framing that this is a feature roadmap misreads it. A feature roadmap adds capability inside an existing structure. What Google announced is a structural shift in how the platforms work. The interface layer that mediated between user and content for the last two decades, meaning the search box, the thumbnail grid, the video editor, is being absorbed into a model layer that does the composition itself. The user input goes from query to intent. The platform output goes from list to artifact. For a media operator, that means the surface you have spent years optimizing for is being replaced underneath you, on the platform’s schedule.

The dependency question

The keynote closed with Demis Hassabis, the CEO of Google DeepMind and the executive most directly responsible for what gets shipped under the Gemini name. Normally Sundar Pichai bookends the event. This year Hassabis got the closer, and he used it to talk about standing in the foothills of the singularity and ushering in a new golden age of scientific discovery. It was the kind of language that lands well in a developer room and lands strangely in a media boardroom.

The interesting question for a media operator is not whether Hassabis is right about the singularity. The interesting question is about dependencies. Every operator reading this has lived through platform restructurings on platforms they did not own. Facebook’s video pivot. The Twitter API. Snap’s developer story. YouTube monetization changes. Search algorithm updates. The pattern is consistent across companies. Platforms restructure when their economics require it. The businesses that built dependencies on those platforms get no vote in the timing.

Google has a famous service graveyard, and that is the easy joke. It is also the wrong worry. The real worry is that every layer Google absorbs into the Gemini stack, meaning creation, discovery, advertising, identity verification, is a layer where the media operator’s leverage compresses. Taking a Gemini Omni dependency at the creation layer, or an Ask YouTube dependency at the discovery layer, means accepting that future restructurings will happen on Google’s economic calendar, not yours.

That is the operator translation of Google I/O 2026. The demos were exciting, and the developer enthusiasm was earned. But for media companies, the signal is less about what can be built next week and more about where leverage is moving. Creation, discovery, and search are being pulled into a model layer Google controls. The dependency math just got more expensive. The useful work now is not panic. It is dependency design: know what you are adopting, what you can swap out, and what it would cost to leave. The dependency itself is the design choice.

Around the clip, an argument forms. One side surfaces the manifest as evidence. Look. It has a signature. It came from a camera. It hasn’t been edited. The other side responds that the manifest does not address whether what is depicted in the footage happened the way the surrounding caption claims. Both sides are now demanding the wrong thing of the technology. The audit trail is intact. The truth claim is unresolved. The two get conflated, and the public conversation collapses into noise that the architecture was never designed to settle.

This is the observability question playing out at the worst possible scale, in front of an audience that does not yet have the vocabulary to distinguish chain-of-custody attestation from truth validation. The infrastructure is doing its job. It was never built to do the job people are now assigning to it.

This is the third and final piece in the Human Layer series, and the closing of the larger arc that began with Context Is the New Integrity Layer.

The category mistake

Olga Kornienko, on the Future Frames podcast, named the distinction in a single line: “C2PA is not gonna argue if I say that my wall is pink.” She is the COO of EZDRM, which has been doing the operational work of making content provenance actually function at the asset level for several years.

The pink wall framework is the load-bearing distinction for everything else in this piece. C2PA — the Coalition for Content Provenance and Authenticity — validates that a thing was made in a particular way, by particular tools, with particular edits. It does not, and structurally cannot, validate whether the claim attached to that thing is correct.

The audit trail is not the argument. The audit trail is the precondition for the argument.

Read more

That same week, the company is still tagging AI-generated music, demonetizing fraudulent streams, and backing the new DDEX standard for AI-use credits. The signaling is loud: AI music is a problem to be governed. AI audio in the next breath is a feature to be shipped.

These are not contradictions. They are the same rule applied to two different economic surfaces. AI is welcome wherever it does not break the existing model. AI is contested wherever it does. The fight is not about authorship, taste, or even consent. It is about which contractual surface absorbs the cost.

The fight that already moved

For two years, the AI music conversation pretended to be about audiences. Would listeners accept it? Would they tell the difference? Would human artistry survive the flood? The data settled most of those questions quietly. Luminate’s most recent report shows U.S. consumer sentiment toward AI in music creation moving from -13% in May 2025 to -20% by November, with the sharpest decline among Gen Z and Gen Alpha. Deezer is now receiving around 75,000 fully AI-generated tracks every day, roughly 44% of daily uploads, while consumption sits at 1 to 3% of total streams and 85% of those streams get flagged as fraudulent.

Listeners did not vote for this. They are net-negative on it. The supply curve and the licensing curve kept moving anyway.

Because the actual fight had already migrated. Universal settled with Udio in October 2025. Warner settled with Udio a month later, then with Suno a week after that, picking up Songkick in the bargain. Suno raised $250 million at a $2.45 billion valuation with NVIDIA participating. Merlin signed in January 2026. Kobalt followed in March. Sony is the lone holdout from the original lawsuit; everyone else is already inside the licensing economy they spent eighteen months claiming was illegitimate.

The piece of this most coverage misses is that the labels did not win the listener fight. They abandoned it. They stopped trying to convince anyone that AI music was bad and started building the contractual surface that would let it exist on terms they could meter. Audience preference was the rhetorical shield. The real prize was the metering rights.

Two doors, same building

Spotify’s split posture this week is the cleanest possible expression of where this lands. The royalty pool is a defended door. AI tracks that try to walk through it get tagged, demoted, and demonetized, because the economics of pro-rata payouts mean fraudulent streams pull money out of real artists’ pockets at industrial scale. Music had to fight its way to a deal because it was inside the royalty engine. The rules are about defending what is already metered.

The Personal Podcast door is wide open. AI-generated audio walks straight in, gets stored in a private library, and never enters search, recommendations, or the royalty graph at all. There is no metering surface to defend because there is no shared revenue pool to dilute. It is your library, your prompt, your file. Spotify is the playback layer, not the marketplace.

Same company. Same week. Same coherent rule: AI is welcome wherever the existing economic model is not at risk.

That is not hypocrisy. It is the new operating principle for any platform with a royalty obligation. Once you see it, the rest of the industry’s behavior becomes legible. Why labels licensed Suno but Sony has not. Why Deezer can transparently tag AI music as a commercial feature. Why YouTube can tolerate enormous volumes of AI content while Spotify aggressively filters it: their economic geometries are different, so their tolerances are different. The question every platform is now answering, whether they say so out loud or not, is which surfaces of their stack can host AI without breaking the contracts that pay creators.

The clearest contrast is Bandcamp, which banned AI-generated music outright in January. Bandcamp can do that because its revenue model is direct-to-fan purchase, not streaming royalties. There is no pro-rata pool to dilute, no metering surface that AI uploads can game. Banning becomes a clean editorial choice rather than an economic one. Spotify does not have that option. The ban is available to one and not the other.

Where the leverage moved

This is the part that is not yet settled, and where the next fight is forming.

If the listener was the wrong battlefield, and if the licensing rooms have mostly rationalized into deals, the open question is who controls the surface where the audio actually gets made. Spotify’s CLI is a tell. The integration target is the agent, not the consumer. A developer writes a prompt. An agent generates the file. Spotify ingests it and renders it. The user is downstream of three steps that used to be invisible.

That changes the economics in a direction the music industry has not had to think about yet. In the royalty model, the unit of value is the stream and the contractual counterparty is the rights holder. In the agent model, the unit of value is the generated artifact and the counterparty is the agent vendor. Different metering surface. Different leverage. Different fight.

And this is not a Spotify story. The CLI is essentially browser compatibility for the agent era. Every platform with a metered surface is about to face the same compatibility question, because so much consumption is moving through agents that fetch and assemble on behalf of the user. Whoever ships first becomes a default integration target. Whoever waits gets routed around. The agent is the new browser, and platforms are about to spend the next eighteen months deciding which of their surfaces are addressable through one.

Right now Spotify is positioning itself as the listening-surface gravity well. Every agent that can drop audio into Spotify becomes a feature for Spotify. But that bargain runs both ways. If agent-generated audio becomes a meaningful share of listening hours, and if agents become the way users compose their day, then the agent vendor sits between the platform and the user. That is the exact place where Apple sat between developers and iOS users for fifteen years, and where YouTube sat between creators and audiences for the streaming era. It is the place that turns into the rent-collecting layer once volume arrives.

Where this actually ends

The AI music story stopped being a story about creativity a while ago. It is a story about which surfaces of the media stack can absorb generative content without breaking the economics underneath them. Listeners showed up in the prep doc and the trade press as the moral center of the conversation, but the data shows they were never the binding constraint. The binding constraint was always the contract.

The equilibrium probably arrives from the listening side, not the supply side. Platforms have already started shipping the primitives. YouTube recently let users set Shorts duration to zero and effectively turn the format off. The AI-content version of that switch is coming, and once it ships, the economic incentive that drove the flood collapses. Spam works because it is cheap to make and the platform delivers the audience. Take away the audience and the math stops penciling out, regardless of how cheap generation gets.

The same agent layer that lets Spotify ship Personal Podcasts is what eventually arms the consumer to refuse the parts they did not ask for. An agent that can generate audio on demand can also filter what shows up in the feed by whatever criteria the user sets. Sounds too AI. Mimics an artist I like. Lacks credited human contribution. The filter does not need to be perfect. It needs to be cheap, and it will be.

That is the loop that closes this. Platforms get to ship AI audio without breaking their royalty contracts because the new audio lives outside the metered surface. Listeners get to keep AI out of the parts of their listening they care about because the agent does the sorting. The labels keep the licensing economy they just rebuilt. Everyone gets a version of what they wanted.

The fight that is left is the one nobody has named yet. When the agent is the integration layer for ingestion and the filter for consumption, the agent vendor sits at both ends of the pipe. The labels know how the meter-the-supply game ends because they have played it before. The meter-the-listener game is new, and the contracts that would govern it do not exist.

The performer might be a verified, consenting subject. They might be a stylistic echo trained on someone whose name is in the model’s training set but whose name is nowhere in the platform’s output. The system has no resolvable answer because the system was not built to have one. The contract that covered the production sits in someone’s email. The likeness rights, if they exist, sit in a separate database belonging to a different entity. The performer’s identifier, assuming they have one, was not requested, was not stored, and would not have been resolvable from inside the workflow even if it had been.

Once agency is observable, the next question is whose agency. That was the bridge from the last piece. It is also the question most of the industry’s rights, contract, and credit frameworks quietly assume has already been answered. It hasn’t. There is no resolvable identity layer for the people inside generative media, in any form a workflow can query and trust. The asset has an identifier. The performer in the asset usually does not.

This gap does not exist because the industry hasn’t thought about it. It exists because the industry built identity infrastructure for assets fifteen years ago and never built the equivalent for people.

The industry has done this before, for assets

The Entertainment Identifier Registry has been operating since 2010. Over two million records. Founded by MovieLabs, CableLabs, Comcast, and TiVo. Widely adopted across major studios, streamers, and broadcasters, and used by PBS at the network level. It is built on the same Digital Object Identifier system that runs academic journals and scientific data, governed by the same DOI Foundation, resolvable through the same handle infrastructure originally designed by Bob Kahn.

What makes EIDR work is what it deliberately doesn’t do. The registry is intentionally thin. It holds the minimum metadata required to disambiguate one piece of content from another: title, language, release date, country, length, type. It does not hold ownership. It does not hold rights. It does not aggregate commercially valuable metadata about the asset. It is purely functional, and persistent enough to survive ownership changes. When an asset is sold, transferred, recombined, or re-edited, the identifier travels with it.

That design philosophy is the load-bearing point. EIDR works because it solved one problem cleanly and let the value layers sit on top. Disambiguation became an industry-wide service that nobody owns and everybody benefits from. The studios accepted, fifteen years ago, that some kinds of identity infrastructure cannot be a competitive advantage without breaking the system that produces the advantage. Make the registry neutral. Make it persistent. Make it governed by a body whose only job is to run the registry. Build the commercial systems on top.

That principle, accepted for assets, has not been accepted for people. The asset has had a resolvable identifier for fifteen years. The performer in the asset has not. The voice in the dub, the writer of the script, the director of the cut, the choreographer of the movement, the musician on the underscore: none of them have. Their identity has lived in contracts, payroll systems, IMDb pages, agent databases, union rolls, and credit slates. None of those are resolvable. None of those persist through ownership changes. None of those federate.

Read more

What’s moving isn’t the technology. It’s the economic primitive underneath it. Per-seat pricing, the model that has organized white-collar work, enterprise software, and broadcast funding for decades, is structurally incompatible with per-token costs. The institutions built on the old logic are compressing in real time, and the receipts are coming in from every layer of the stack at once.

The buy-side can’t track its own spending

Source: The Information, May 6, 2026

A partner at a large venture firm gave five staffers enterprise Claude accounts. Within weeks, those accounts were costing the firm upwards of $1,000 per day, per account. Power users were defaulting to top-tier models for trivial tasks like email replies. At that pace, five people would have cost the firm over $100,000 per month.

The partner surveyed his portfolio companies on whether they knew what their AI was costing them. Fewer than 10% did. Uber blew through its full-year 2026 AI budget in a few months. Shopify, reporting earnings the same week, told the market that AI costs had offset the efficiency gains in its software business.

Why it matters

The smart money has been operating without instrumentation. The pricing model changed underneath them and the dashboards haven’t caught up. Model routing startups, dormant since 2023, are starting to come back. Open source eats the simple work. The tokenmaxxing tide is turning, not because anyone made a strategic call but because the bills started arriving.

The sell-side is being re-rated in public

Source: Business Insider, May 6, 2026

Pat Walravens, an analyst at Citizens, told Business Insider that two-thirds of today’s top SaaS companies won’t emerge from the AI era intact. His framing splits software into two layers. Infrastructure, the pipes and communications and data, gets more essential as AI agents need connections to the real world. Twilio posted its fastest growth rate in three years last week. Bandwidth’s CEO put it directly: AI agent intelligence is only as good as its connection to the real world.

The application layer is the layer being eaten. Workday, ServiceNow, SAP, Atlassian, and Adobe stocks have declined sharply over the past year. Walravens’s litmus test for survival: if your product can be vibe-coded with Claude or Cursor over a weekend, you’re in trouble. He named Asana and HubSpot specifically. Both companies’ responses to Business Insider, emphasizing governance, integrations, multiplayer collaboration, read as the defensive posture of products being asked, on the record, to justify their per-seat margins.

Why it matters

A consumption-priced model can’t be defended by a seat-priced product whose function can be reproduced at the prompt layer. The infrastructure layer hardens. The application layer compresses. The line Walravens draws is the line the market is already pricing.

The labor receipt arrives, four times in the same week

Sources: CNBC, May 5, 2026 · Benzinga, May 5, 2026 · Bloomberg, May 5, 2026 · Variety, May 5, 2026

Coinbase announced 700 job cuts, about 14% of staff, with CEO Brian Armstrong stating directly that AI is changing how we work. Freshworks announced 11% cuts citing industrywide AI disruptions. PayPal disclosed plans for a 20% workforce reduction over two to three years targeting roughly $1.5 billion in savings. Disney’s Josh D’Amaro confirmed 1,000 layoffs underway, framed as operational streamlining.

Four firms across financial services, software, payments, and entertainment, announcing within a five-day window, with AI named on the record in three of the four cases. None of them are early-stage companies experimenting at the edge. They are the firms whose seat counts defined the white-collar middle class.

Why it matters

The headcount reductions read less like cost discipline than like a write-down of how much human labor the per-seat model was actually paying for. When AI compresses the value of a seat, the seat itself eventually moves. The first round is happening at companies with the financial clarity to do the math out loud. The second round will happen at companies that didn’t.

The mismatch arrives at a publicly-funded institution

Source: Deadline, April 16, 2026

The BBC, announcing detailed plans for £500 million in cuts and up to 2,000 layoffs concentrated in its News division, did not blame AI directly. Interim director general Rhodri Talfan Davies said something more revealing in his all-staff briefing: “If we had a funding model that mirrored our consumption, all of this would go away. We have a funding model at the moment that is unsustainable and is reaching the end of its sell-by date.”

The license fee is the original per-seat product. It charges every household a flat rate to fund a fixed institution. The cost structure of AI deployment, consumption-based and scaling with use, is incompatible with that funding logic. Talfan Davies is the first major institutional figure to name the mismatch on the record. The handoff lands in a few weeks. Incoming director general Matt Brittin, an ex-Google executive who has spoken about AI giving people “special powers,” takes over May 18.

Why it matters

The per-seat-to-per-token mismatch is no longer confined to enterprise software. It is showing up in the funding logic of public-service broadcasting. The BBC’s challenge is the same as Workday’s, denominated differently. A flat input is being asked to support a variable output, and the math doesn’t work.

Closing Note

The supply side has been moving with conviction because it knows what the units cost. Anthropic took all of Colossus 1’s capacity in a single deal this week. Nvidia took warrants in Corning. SpaceX proposed a $119 billion chip fab. Microsoft is reportedly considering walking back its 2030 renewable target to feed the build-out. Every actor at the top of the stack is reaching down to lock in physical capacity.

The demand side has been operating on instinct. Buyers default to top-tier models for trivial tasks. SaaS vendors charge per seat for functions that can be vibe-coded in an afternoon. Employers staffed white-collar work for a productivity baseline that AI is now resetting. Public broadcasters fund themselves through a consumption-blind license fee while their cost structure becomes consumption-defined.

The compression isn’t coming. It’s already here. What changed this week is who started saying so on the record. A VC firm, a public-markets analyst, four CEOs, and a public broadcaster’s interim director general, in the same five days, all describing different surfaces of the same underlying break. The story of the next eighteen months won’t be about the AI infrastructure being built. It will be about which institutions on top of it can survive the rewiring of how they’re paid.

The funding model and the cost model used to match. They don’t anymore.

Coda

Source: CNN, May 6, 2026

Ted Turner died Wednesday at 87, and being from Georgia myself, I felt obliged to mention him. In 1976 he beamed an Atlanta UHF station up to a satellite and turned it into TBS, the first superstation. Four years later he built CNN, the first network that ran on the assumption the signal should never stop. He didn’t disrupt the media business of his day. He fundamentally rewired how the signal got to the home and set the path consumption would take for forty years. The funding model that paid for it ran on the assumption the bill came once a month. They held just as long.

Read more

The next moon landing may become the largest live streaming event in internet history, and the streaming industry is not yet fully organized around what that means.

That was my biggest takeaway from NAB, after listening to a session opened by NAB’s John Clark with NASA’s Kevin Coggins, Rebecca Sirmons, and Lee Erickson. The conversation had the emotional force you would expect from a discussion about human spaceflight, but what stayed with me was how practical it became. Artemis is a story about exploration, but it is also a story about communications, video, latency, certification, bandwidth, distribution, and the work required to let the world experience spaceflight in real time.

That is where our industry comes in.

The audience is already there

Rebecca shared numbers from Artemis that should make everyone in streaming pay attention. NASA saw roughly 19.1 million viewers on broadcast, while streaming partners collectively reached around 50 million. She noted that YouTube alone drew approximately 47 million views across a 10-day live broadcast.

Third-party reporting points in the same direction. The Planetary Society reported that more than 18 million people watched the Artemis II launch across major U.S. cable networks, while splashdown drew 27 million on those same channels. It compared that audience to Game 7 of the World Series and noted that Artemis II reached “Taylor Swift levels” of public awareness, with polling suggesting roughly 70–80% of Americans had at least some awareness of the mission while it was happening.

Those numbers matter because Artemis II was the preview.

The next lunar landing will be distributed across broadcast, NASA+, YouTube, streaming platforms, connected TVs, social platforms, FAST channels, news outlets, apps, and viewing experiences that may not exist yet. Demand will be fragmented, global, emotional, and simultaneous.

That scale should sound familiar to anyone who has worked on large live events. The difference is that the origin camera is on or near the Moon.

Communication is the mission

One of the most striking parts of the panel came from Kevin Coggins, who described NASA’s shift from traditional RF communications to optical communications. In plain English: lasers.

Streaming people are used to thinking about bandwidth as a scaling problem. Add capacity. Add CDN footprint. Add cloud resources. Tune the ABR ladder. In space, bandwidth starts as a physics problem.

Kevin described the precision required to hold an optical link between a spacecraft moving thousands of miles per hour and a receiving point on Earth. The spacecraft is moving. Earth is rotating. The atmosphere is shifting. Clouds interfere. The link depends on extraordinary pointing accuracy across a system that is constantly in motion.

And yet, during Artemis testing, optical communications worked well enough that the difference was visible. When the optical link came on, the video got noticeably better.

That success changes the trajectory. Optical communications become foundational to the future media path from deep space. Higher bandwidth makes better video possible: higher resolution, higher fidelity, better frame rates, more camera angles, and eventually immersive experiences. It also raises expectations. Once the audience sees what is possible, grainy “good enough” mission video becomes harder to justify.

Kevin also made clear that optical introduces its own realities. Clouds still matter. Atmosphere still matters. Ground stations still matter. He described optical operating around 1550 nanometers and the challenge of getting through the atmosphere. The long-term answer may include optical links from lunar systems to Earth-orbiting constellations, then down through more conventional terrestrial paths. That is where space communications and modern distribution architecture begin to meet.

Space hardware is a hostile production environment

Lee and Rebecca grounded the conversation in the reality of making media technology survive space.

In our world, “production-grade” usually means reliable under operational pressure: a live event, a bad network path, a flaky encoder, a CDN incident, a player bug, or a control-room failure. NASA has to deal with all of that, plus launch forces, radiation, thermal extremes, vacuum conditions, oxygen constraints, human safety requirements, and failure scenarios where nobody can walk into a rack room and swap a box.

Rebecca and Lee talked about what that means for encoders and video equipment. Devices may be dropped to simulate G-forces. They may be exposed to radiation-like conditions. They may be thermally stressed. Every component has to be evaluated for environments where failure may be unrecoverable.

That reality creates one of the hardest gaps between NASA and industry. The cameras and related systems used on Artemis were based on technology that was effectively a decade old. NASA did not arrive there because it prefers old technology. It got there because requirements, testing, certification, and mission assurance take time.

The streaming industry reinvents itself every year.

That gap is one of the most important diagnoses from the panel. NASA needs modular pathways that allow modern encoders, codecs, cameras, and related systems to be evaluated and upgraded without forcing the entire media stack back through a decade-long cycle.

This is an architecture problem as much as a hardware problem. The industry can help define systems where components are isolated, tested, certified, swapped, and improved while preserving mission assurance.

Bandwidth is a prioritization problem

Another detail from the session stayed with me: even when NASA has a link, every bit has a job.

Kevin referenced a 6 Mbps RF Deep Space Network data rate to Orion. For streaming professionals, that number immediately triggers practical questions: resolution, frame rate, codec efficiency, error resilience, latency, quality ladders, and compression tradeoffs. But on a spacecraft, video shares the path with mission-critical telemetry, spacecraft health, astronaut safety data, and operational communications.

Video quality does not win every argument. Sometimes it should not.

During Artemis, that meant practical tradeoffs: lowering frame rate, changing compression strategy, and allocating bandwidth around higher-priority mission data. A streaming engineer may see an image-quality challenge. A mission operator sees a safety and assurance problem with video as one of several payloads competing for constrained transport.

The hard requirement is to make both worlds coexist.

If future lunar missions carry more cameras, more live feeds, more immersive systems, and more public-facing video, the prioritization layer becomes increasingly important. The question is no longer only how to produce the best image. It is how to produce the best image inside a system where astronaut safety, spacecraft operations, latency, resiliency, and public experience all share the same constrained path.

Latency changes the product

The Moon is close enough that live interaction feels possible, but physics still applies. Mars is a different category altogether. Kevin noted that Mars communications can involve delays measured in minutes, depending on orbital position. Voyager pushes the idea even further, with signal travel times measured in many hours.

That matters because latency changes the media experience.

A lunar stream can still feel live. A Mars experience will need different expectations, different workflows, and likely a different user experience. Troubleshooting, control, interactivity, live commentary, social participation, and mission operations all change when the delay becomes minutes instead of seconds.

This is where the streaming industry’s experience with live sports, remote production, low-latency streaming, multi-CDN architectures, and cloud production becomes relevant, but only as a starting point. Deep-space media will require new patterns for delay-aware storytelling, autonomous capture, store-and-forward publishing, asynchronous interaction, and quality-of-experience measurement.

The Moon is the proving ground.

From mission broadcast to continuous presence

The biggest structural shift is NASA’s move from episodic mission communications to continuous lunar presence.

Rebecca described NASA+ as a kind of “Field of Dreams”: build the place where the story can live, make the data accessible, and let audiences and partners come to it. Kevin described a future where cameras are everywhere: on landers, rovers, relay satellites, astronaut-following systems, and eventually lunar infrastructure. Lee talked about 360 video, immersive experiences, and the creative possibilities of opening NASA’s media assets to partners who can build new experiences on top of them.

That is the moment the story turns from broadcast to platform.

A future lunar base becomes a persistent media environment. There may be live feeds, VOD archives, multi-angle views, AI-indexed moments, immersive streams, educational experiences, accessibility workflows, and public datasets that can be remixed in ways NASA will not invent alone.

The archive itself becomes part of the mission. Continuous lunar media will need indexing, search, provenance, rights metadata, descriptive metadata, accessibility metadata, and tooling that lets educators, broadcasters, scientists, journalists, creators, and platforms find the moments that matter.

That ecosystem has to be designed.

The actual gaps industry should help solve

The call to action should be specific.

First, NASA and industry need modular certification pathways for media technology. If every camera, encoder, or codec improvement is trapped behind a multi-year validation cycle, the media stack will always trail the commercial world. The industry can help define architectures where components can be tested, isolated, swapped, and upgraded without compromising mission assurance.

Second, we need shared distribution choreography for an event that could draw hundreds of millions of simultaneous viewers across fragmented platforms. This involves origin strategy, redundancy, partner access, stream packaging, live operations, traffic forecasting, failover, observability, CDN coordination, content protection where appropriate, accessibility workflows, and clear rules for who gets what signal under what conditions.

Third, we need standards and practices for AI indexing and metadata around always-on lunar media. If NASA is sending back continuous video from landers, rovers, relay satellites, and astronauts, AI can help identify events, objects, anomalies, educational moments, accessibility cues, and reusable clips. That only works if the media, metadata, provenance, rights, and access patterns are designed intentionally.

Fourth, we need a shared operating model for public-private contribution. NASA can build the systems that only NASA can build. It can get the spacecraft there. It can operate the mission. It can manage the risks of human spaceflight. The industry around NAB knows how to move live video at scale, recover from failures, distribute across platforms, build media workflows, measure QoE, and turn raw signals into experiences people can actually use.

Lee said it clearly: no idea is too crazy.

Kevin framed it another way: NASA wants to enable others to plug in.

That is the door opening.

The last moon landing was a broadcast. The next one will be a global streaming event. The one after that may be a persistent media platform from another world.

The camera is no longer just documenting the mission.

It is part of the mission.

Personal note: I’m grateful to the Streaming Video Technology Alliance for being at the forefront of this conversation and for giving me the opportunity to help serve as a liaison to NASA. I’m also grateful to Megan Cruz, Nilufar Ramji, Sarah Volkman, and the broader NASA and industry teams doing the hard work of connecting space exploration with real-world media delivery.

None of these moves came from a product launch. None came from a lab. They came from institutions with leverage that the product-builders had treated as background. The Disney FCC review, the Musk OpenAI ruling, the Eve hallucination filing, and the Warner-Paramount shareholder vote are different events with different stakes. Read together, they describe the same shift. The framing privilege is contracting. The permission structure is reasserting itself.

The FCC remembers it has leverage

Source: Financial Times, April 30, 2026

The FCC plans to review Disney’s broadcast licenses because Jimmy Kimmel told a joke that offended the President. The mechanism is straightforward. ABC’s local affiliates hold government-issued spectrum licenses. The FCC controls renewal. Brendan Carr is using that control. By the way, if you haven’t seen the most famous podcast inside of a podcast “Brendan Carr is a Dummy” I highly recommend giving it a try.

The harder part is structural. Disney’s linear networks business is shrinking, with revenue down more than ten percent in 2025 and another seven percent expected this year. By the end of 2026 it will be less than ten percent of company revenue. But it still produces nearly a fifth of group operating profit. Disney+ produced about a billion dollars of profit last year on twenty-five billion in revenue, against eight billion in operating losses over the previous five years. The streaming business is not yet profitable enough to fund an exit from the broadcast business. Disney is structurally trapped on the surface the FCC controls.

Why it matters

For five years, the legacy media transition has been narrated as a managed decline. Investors were asked to look past linear and toward streaming. The FCC just demonstrated why that framing was always provisional. Regulatory leverage attaches to whichever surface the company cannot afford to abandon. As long as linear pays the bills, the regulator who licenses linear has a seat at every editorial conversation. The same logic applies to Paramount, whose CBS is even more dependent on live sports than Disney. The transition story protected the platforms from the regulator only as long as the platforms could credibly threaten to leave. They cannot.

A federal judge narrows the AI risk story

Source: The New York Times, April 30, 2026

Judge Yvonne Gonzalez Rogers, presiding over Elon Musk’s lawsuit against OpenAI in federal court in Oakland, ruled on Thursday that the trial would not address AI’s existential threat to humanity. “We are not going to get into issues of catastrophe and extinction,” she told Musk’s lead counsel. The ruling was procedural, made during a dispute about admissible testimony. Its effect is broader.

For most of the last decade, the existential risk frame has been the AI industry’s preferred public narrative. It justified the founding of OpenAI. It justified the spending. It justified the access. It is the story Musk’s lawyers want to tell the jury, because it casts their client as protecting the world rather than competing with it. The judge cut it from the record on the grounds that it was speculative and would, in her words, cause “this whole thing to explode for the world to view it.” A federal judge just decided which AI narrative is allowed in the official transcript of the most consequential AI lawsuit in progress.

Why it matters

Court records set precedent for what counts as relevant. Other plaintiffs and defendants will read this ruling and adjust their framings accordingly. The existential narrative is not banned, but its admissibility just got harder to argue, and the labs lose a layer of rhetorical protection that has shaped AI policy discussion since 2015. The control over how AI gets described in adversarial settings is no longer fully theirs.

Vertical AI gets named in the court record

Source: Business Insider, April 30, 2026

A Louisiana personal injury lawyer apologized in a private letter to a federal judge for filing two briefs containing fabricated case quotations. The lawyer named the software he had used. It was Eve, a venture-backed legal AI startup valued at one billion dollars, which sold itself to law firms on the explicit claim that it was more reliable than general-purpose chatbots. Eve denied that its software produced the fabricated quotes. The denial is now part of the public record alongside the apology. Opposing counsel in an unrelated case has already pulled the letter into its own sanctions request.

A French researcher who tracks AI hallucinations in court filings estimates that fewer than ten percent of cases identify the software used. Most lawyers stay quiet because they were using tools they were not authorized to use. The naming dynamic is new, and it surfaces a risk profile that vertical AI vendors had successfully kept abstract. When a lawyer says “I used ChatGPT,” the AI industry has plausible deniability. When a lawyer says “I used Eve, which is built for plaintiff-side law firms,” the company gets pulled into the discovery layer of a malpractice-adjacent dispute.

Why it matters

Vertical AI vendors raised on the premise that domain specialization equals reliability. That premise is now testable in court records, by opposing counsel with an incentive to find the failures, in industries where the cost of being named is reputational rather than just commercial. Eve will not be the last vendor in this position. Every billion-dollar vertical AI company in legal, medical, financial, and accounting workflows is one filing away from the same disclosure. The marketing layer met the institutional layer this week. The institutional layer keeps a transcript.

The defensive consolidation closes

Source: CNN, April 23, 2026

Warner Bros. Discovery shareholders approved Paramount Skydance’s eighty-one billion dollar acquisition by a wide margin. The combined entity will absorb HBO Max into Paramount+, control both CBS and CNN, and become one of three remaining major Hollywood studios. The deal carries a ticking fee that escalates the price if it does not close by September 30. State attorneys general are preparing antitrust challenges. Thousands of industry professionals signed a letter opposing the merger. None of that materially slowed the vote.

This is the institutional response from the other direction. The legacy media companies cannot wait for AI to finish rewriting the cost structure of production and distribution. The combined Paramount-Warner entity is a bet that scale at the asset layer can offset whatever the substrate does next. It is also a bet placed under the same FCC dynamic that just hit Disney. CBS and CNN both depend on broadcast distribution. The new owner will be Trump-aligned David Ellison. The regulator and the regulated are getting closer, not further apart.

Why it matters

Consolidation under uncertainty is not a strategy. It is a posture. The companies executing it know the ground is moving and are trying to lock position before it stops. Whether the bet pays off depends entirely on how the next three institutional rounds go. The shareholder vote was the easy one. Regulatory approval, antitrust review, and integration without the FCC weaponizing licensing leverage are all harder. The deal closing in September is not the end of the story. It is the beginning of the harder version.

Pattern Synthesis

Four rooms. A federal courthouse in Oakland. A regulatory review at the FCC. A court filing in Louisiana. A shareholder meeting in New York. None of them are a product launch. None of them are a lab announcement. None of them happened on a stage the AI or media industries control.

The framing privilege that the AI labs and the legacy media platforms have enjoyed depended on the assumption that the institutions around them were deferential. The institutions are no longer deferential. They are using the leverage they have, on the timelines that suit them, with the agendas that motivate them. A judge cuts a narrative from the record. A regulator notices the licensee cannot afford to leave. Opposing counsel turns a vendor name into evidence. Shareholders ratify the defensive move. None of these institutions coordinated. They did not need to. The pressure was building in each of them independently, and it surfaced this week because the underlying conditions finally made the moves cheap enough to execute.

What gets harder from here is not the technology. The technology will continue to advance. What gets harder is the explanation. Every AI company now has to defend its claims in adversarial settings it cannot script. Every legacy media company now has to explain its transition to a regulator who can revoke the license while it waits. The companies that win the next phase will be the ones whose claims survive the institutional translation. The ones that lose will be the ones still running the old playbook, in which the audience was assumed to be patient.

Closing Note

The labs spent three years explaining what AI is. The studios spent five years explaining the streaming transition. Both stories required the audience to wait. The institutions might have stopped waiting.

I spent the week at NAB 2026 expecting to hear operators admit they’re tearing down what they built. What I heard was more interesting. Most of them aren’t tearing anything down. They’re abstracting over it, hoping the layer above will do the work the layer below was supposed to.

The diagnosis is widely shared. An operator I sat with on Saturday, who’s been inside the supply-chain rebuild conversation across multiple tier-one media companies, walked me through it without flinching. The top media companies spent the last five-to-eight years building out internal supply chains that worked when the underlying technology was stable. AI made the target volatile. The original engineers have moved on. What’s left is Frankensteined. He’d just come from a meeting where a sports league said the quiet part out loud, we’ve got to tear it down. It doesn’t work at all.

The deeper claim he made stuck with me. Best-of-breed is dying not for snobbish reasons but because AI metabolizes glue. If your components can’t share metadata throughout the pipeline, AI has nothing to act on. A welded-together stack is the only way to get actionable intelligence end-to-end. He pointed at Akta, the Google spinout now powering CBS, Nexstar, Fox TV Stations, Discovery Latam, and TelevisaUnivision’s ViX, as the example of what that looks like when it ships.

The vendors told the same story in public language. AWS used its NAB platform to push agentic AI as the bedrock of modern media operations, with multi-step workflows the existing supply chain was never designed to feed. The chain was built for human consumers. AI exposed the seams. Avid and Google Cloud announced a multi-year partnership during the show to embed Vertex and Gemini directly into Media Composer with a unified data layer underneath. Adobe is moving in the same direction. The bet from the AI-native camp is that the seams are now structural and the only honest fix is a tightly coupled stack where metadata flows end-to-end because it never had to leave the building.

The most architecturally serious answer is the one getting the least press. The EBU, AMWA, BBC, and Linux Foundation are quietly building the open-standards equivalent: MXL for in-memory exchange, DMF for facility orchestration, TAMS for time-addressable storage. Frame-as-URL, but as a public utility rather than a product. It’s running in production at the BBC. Almost nobody on the floor was talking about it.

The proprietary version of the same idea showed up in a fundraising meeting on Sunday morning. Peter Bruggink and the Media-Anywhere team demoed AirFrame, scrubbing through ProRes media stored in S3 in Ohio from a laptop on conference Wi-Fi, with frame-level caching, forensic watermarks burned in per logged-in user, and no proxy. Premiere thinks the files are local. The player asks for thirty frames at a time over async HTTP/2 and the server reads them in parallel from object storage. They’ve already signed a top-three US sports broadcaster. The most interesting customer in their pipeline is a financial institution, because the bank’s compliance posture demands a zero-trust media handling that media itself has never enforced. Their architecture aligns with eight of the ten MovieLabs 2030 principles, which is the studios’ own published roadmap for where the production stack needs to go. The shift they’re betting on is the right one, and it’s the one TAMS is making in the open-source direction at the same time. The file is no longer the unit of work. The recipe is. An immutable description of what plays, when, from where, and how, that humans and AI can both author and that every output port consumes in its native format. The recipe travels for free.

The third answer is what most operators are actually doing, and it’s the least romantic. Backlight’s iconik data shows their customer deployments holding steady at 65% cloud and 35% on-prem. Net Insight is now selling Nimbra Live Intelligence with the explicit promise that it works without rip-and-replace. Wowza and Cloudflare demoed Media over QUIC framed as a way to apply intelligence to streams without rearchitecting the pipeline. The market is rewarding the abstraction story.

Sam Phillips at AMC said it on stage at AWS Theater. Spending three to six months evaluating vendors and choosing one, only to watch the methods go stale before the contract gets signed, isn’t a place his company can afford to be. His answer was PRISM, an overlay on AWS Media Lake plus TwelveLabs Marengo plus Bedrock, abstracted sufficiently that he didn’t have to wait for the rest of the supply chain to get organized. That sentence will be quoted by every CTO with a budget meeting next quarter.

What nobody at NAB wanted to call a re-architecture is the one actually happening to live. I spent a long evening with a technologist who’s worked inside more than one of the major streamers and broadcasters over the last several years. His diagnosis was that live had been bolted onto VOD-shaped organizations and it doesn’t work, because live is operationally and culturally different. He gave me a small example that I keep coming back to. At one company, you’d call a colleague, say you needed to jump on a video call, and they’d send you a calendar invite. At another, you’d call and they’d just join. The first is a VOD mentality. The second is a live mentality. The difference shows up in the operations bridge during an event. He had specific stories about how it shows up, like vendor scoring practices that exist on the VOD side but don’t on the live side, or regional rebalancing for a major event that used data on what people watched as VOD as the proxy for live demand, with predictably uneven results. You can build the world’s best CDN backbone and still ship a bad show because the cultural muscle for live is something you grow, not something you provision.

Netflix’s own Tech Blog has been making the same argument out loud. The April 2026 post described how engineers who built Netflix’s live pipeline were also the ones operating it, with no operational layer to absorb the load, and how Netflix had to stand up a Live Command Center, dedicated Site Reliability and Ops teams, and a Technical Live Management group to run the shows. It’s the most candid disclosure any major streamer has made about what live actually requires, and almost no one in the trade press has picked it up. The cultural rebuild is real. It just isn’t being announced.

The new generative AI studios are running the same play in their corner. Secret Level, where I’m part of the team, is explicitly refusing SaaS in favor of co-production. The economics only pencil if the studio owns its tooling and its IP. The capital problem is real. We aren’t a tech company, we aren’t a film fund, we aren’t an agency, and the institutional VCs that should fund us have theses that can’t accommodate the hybrid shape. So we’re inventing new financing structures, celebrity syndicates, strategic anchor checks, SAFEs instead of priced rounds, that look more like 1920s Hollywood financing than a 2020s software round. The moat isn’t the model. It’s the production pipeline, the rights chain, the governance layer that lets a studio defend reps and warranties on generative output, and the relationships with brand clients who are pulling us up the value chain because the agencies can’t help with narrative content. Autodesk’s 2024 acquisition of Wonder Dynamics was the same pattern from the other direction. Best-of-breed AI tools are collapsing into platform suites because horizontal tools don’t survive contact with production reality.

The thread that connects all of this isn’t AI. It’s the gap between the people closest to the operational reality and the public conversation about what AI is. The operators have already accepted that this is a re-architecture. They’re choosing their answer, rebuild, open standards, or abstract, based on what their balance sheet will allow. The public conversation is still treating AI as a feature you bolt onto a stack you already own. The longer that gap persists, the more expensive the eventual reconciliation gets, because the cost of getting it wrong compounds in metadata you didn’t capture, contracts you didn’t write, and rights chains you can’t reconstruct.

I called the show floor a lagging indicator before I came out here. A week in, I’d stand by it harder. Every conversation that mattered in this piece happened after the badges came off. The Saturday operator. The Sunday morning demo. The long evening with the technologist. The studio founder’s table. The quietest rooms generated the loudest signal, and I think that’s true every year. It’s just easier to notice when the gap between what gets announced on stage and what gets decided in private is this wide. The vendors who paid for the booths weren’t wrong about what they were showing. They were just one cycle behind the people I was drinking with.

The other thing the night does, which doesn’t show up in any thesis, is remind you why this work is still worth doing. My friend Kreig comes home from NAB every year having forgotten how recharged it makes him. Thirty years in and the show still does it for me, not because of what’s on the floor, but because of who’s in the rooms after. People who’ve been solving the same impossible problems for long enough that they’ve stopped expecting solutions and started enjoying the company. Walking out of one of those late conversations and realizing you’re still curious, still useful, still part of something. That’s the part of NAB the show floor cannot manufacture and cannot replace. It’s also the part that makes the architectural fight worth caring about in the first place. The infrastructure matters because the people building it matter, and the only way to keep both honest is to keep showing up to the rooms where the badges don’t.

The most consequential architectural decisions of the next five years aren’t being made on the NAB show floor. They’re being made in sandboxes being stood up to validate emerging ideas. They’re being made in tier-three sports automation pilots where the economics force what tier-one still affords humans to handle, with operators running ten matches at a time. They’re being made in closed-door operator forums whose attendance lists the show floor never sees. And they’re being made in hotel bars at one in the morning by people who already know each other’s tells.

Three answers are competing. The companies betting on one will be wrong about which one wins. The companies betting on none will be gone. But the people in those rooms will still be in those rooms next year, and the year after that, working on the same problems from new angles. That’s the actual continuity in this industry. Not the platforms. Not the architectures. The people who keep showing up.

AI isn’t a feature. It’s a re-architecture. The companies pretending otherwise aren’t saving money. They’re financing the next teardown. The people who saw this coming were already in the room.

They can reconstruct some of it. The prompt that started the session. A few of the iterations that got discarded. The moment somebody said to try it again with a different composition. They can also remember the part of the project where they bailed out of one platform halfway through, ran the next pass through a different system to get a look the first one couldn’t produce, then brought the result back into the original workflow to finish. The decision to switch was creative. It happened in real time. It made the final piece better. It also broke whatever continuous record either platform was capable of keeping, because neither one was designed to know about the other. The handoff lived in the team’s heads, in a Slack thread, and in nobody’s metadata.

The output exists. The authorship does not, at least not in any form you could put in front of a rights lawyer, an awards jury, or a distribution partner and have it hold up.

This is the shape of the current AI adoption conversation in the creative industries, and it is almost never discussed at this level. The tools work. The outputs are credible. The economics look, at first glance, extraordinary. What keeps slipping is the one thing that makes the output a work in the first place. Under current copyright law in the United States and most comparable jurisdictions, the human contribution is the only element that makes generative output legally protectable. The model produced the frames. A person made it a work. When the workflow obscures the person, it doesn’t just create a documentation problem. It creates an asset that cannot be owned, defended, licensed, or compounded.

The Inheritance Shift described how machine context has to survive transformation. The human layer has its own version of the same problem. Agency has to survive too, and right now most of the systems we have built for it are designed to erase it.

The resistance is not about the tool

For two years, the dominant industry read on creative resistance to AI has been that it is some combination of protectionism, fear of technological change, and failure to understand what the tools can actually do. All three readings assume the resistance is about the tool itself. It is not. The tool is the least interesting variable in the room.

What is actually being resisted is a system deploying the tool without answering four questions. Who benefits from the productivity gain? Who is replaced and who is retained? Who decides how the tool gets used? Who owns the output, creatively, legally, and commercially? Those four questions have very little to do with whether a diffusion model can render a convincing ocean, and almost everything to do with how the industry surrounding the model chooses to operate. When the questions go unanswered, resistance isn’t irrational. It is the only rational response available.

This is not the first time a workforce has been told that its concerns about a new machine were really concerns about progress. The pattern is older than cinema, older than broadcast, older than the union contracts that shape the industry this essay is about. The pattern holds because the useful version of it has always been true: the machine is rarely the problem. The arrangement around the machine is.

Read more

For three years the operative question in AI was whether the technology could do a given thing. This week, across every meaningful move, the question was different. Who owns the pieces required to do it at scale. Who’s picking which lane. What it costs to hold the position once you’ve claimed it. The moves stopped looking like experiments and started looking like industrial allocation.

OpenAI ships ChatGPT Images 2.0

Source: The Verge, April 21, 2026

OpenAI launched Images 2.0 Monday with reasoning built into the generation step. The model ships with 2K resolution, multiple aspect ratios, a separate thinking variant for higher-complexity work, and materially improved rendering of non-Latin scripts. Figma flagged the rollout inside…

Read more

The first three parts of this series traced an infrastructure problem. Context gets lost at every stage of the pipeline. It can persist instead. When it does, the legibility of a catalog becomes its primary asset. That is the capability story. This part is about what the capability produces when it runs without governance: a transformation velocity the industry has no mechanism to account for, and a rights crisis it is not prepared to name.

Transformation Is Now Continuous

Derivatives have always existed. Studios made sequels. Distributors licensed regional cuts. Broadcasters created promotional variants. This was understood as occasional work: a discrete decision, a negotiated agreement, a bounded production event. The human workflow could absorb it because the volume was legible. You could count the decisions.

What changes when context persists is not the existence of derivatives. It is the rate. The finished work is no longer the output. It is the input.

A pipeline operating on structured semantic data does not produce one regional variant. It produces a hundred, automatically, continuously, triggered by parameters the system already understands. Language, platform, device, audience segment, market clearance, content rating, placement window. Each variable the system can read becomes an axis along which the asset can be recomposed without a human decision at every step. The catalog is not a library of things to distribute. It is a library of things to transform.

FAST channels already operate this way, assembling programming from catalog assets in real time based on audience and advertiser signals. Streaming platforms generate platform-specific encodes, alternate audio tracks, and localized metadata at scale. The personalization layer that once applied only to recommendation now reaches into the asset itself. Which cut, which version, which framing, which runtime. The infrastructure for continuous derivative output is not being planned. It is running.

The volume does not feel dramatic because most of it is invisible. Each individual transformation is unremarkable. The aggregate is not.

Read more

Last month something happened to me that hadn’t happened to me in years. It hadn’t happened to me in, dare I say…a decade.

I realized there were four shows I was looking forward to every single week. Not four seasons waiting for me to binge over a weekend, but four different shows, on four different platforms, dropping episodes weekly. (HINT: one of them was The Pitt). And I was so eager to watch these shows as soon as they were released that I actually mapped out their drop schedule so I could keep track of which show was available on which day. I’ll tell you, dear reader, I was just as shocked as you are.

But this experience also brought into focus for me two major shifts that are happening in the media space.

The Return of Anticipation

This first shift is what I’m going to call Days of Yore. It’s similar to the viewer experience during the prime TV era of the 90s and early aughts, but with a bit more technology and awareness. The second shift, because we’re in the media business and because it’s insanely relevant, let’s call Back to the Future, where many of the innovations we’ve long talked about are finally coming to fruition.

To be clear, in the Days of Yore shift, I’m not talking about the story reboot era that has taken over in recent years, but rather how and what we are consuming for entertainment. I’m talking about that brief magical era when appointment TV and streaming were still figuring out how to work together, when bingeing wasn’t yet the default, but you could still watch your show on the broadcaster’s app right after it aired without having to worry about setting your TiVo. See why my mind was so blown when I was trying to figure out the schedule for The Pitt? Or remember all of those nail biting standalone live events? Like when in 2013 we anxiously watched and waited for a spectacular, one-of-a-kind moment, like the night Nik Wallenda crossed the Grand Canyon on a wire and we genuinely didn’t know if the wind was going to knock him off his wire or he was going to be victorious (spoiler: he lived!!).

From Experiment to Strategy

Which brings me to the Netflix Skyscraper Live special, which, in full transparency, I watched on replay. Even so, it still made me think about how, over the years, we’ve watched Netflix, along with other streamers (hi, Apple!), dabble in live programming. But Skyscraper Live felt different. It suggested that Netflix is no longer simply experimenting with live content, but is thinking more strategically about how event-based programming can draw in audiences and keep them engaged over time. Yes, they already have major partnerships and agreements in place for large-scale sports and live entertainment, but their willingness to lean further into live events and test different formats suggests this is no longer a side project. It is becoming a real commitment to the audience and a meaningful part of Netflix’s long-term strategy.

Now, while it’s tempting to say this Days of Yore shift is simply about what audiences want in their heart of hearts, a more realistic take is that content providers are returning to tried-and-true models in order to stretch out programming, maintain conversation around a title for longer, and reduce subscriber churn. In other words, they are slowly walking back the binge era that dominated for so long. But whether that shift is driven by strategy or nostalgia, the result is the same: viewers are once again being asked to anticipate, to wait, and to return.

Back to the Future, For Real This Time

Now let’s get to the second shift, Back to the Future.

In some ways, it is the more nuanced shift, one that is still playing out and shaping itself. But as a technologist and a TV lover, it is the one I’m especially eager to watch. Ever since Back to the Future Part II debuted in 1989, many of us have wondered how closely our own media future would match the one depicted in the film. From multiview television to video calls, art on our TVs, wearable devices, and yes, even the ever-present fax machine, the movie offered a version of tomorrow that felt both outrageous and oddly plausible. And while it’s safe to say that fax machines did not remain central to our day-to-day lives, I would argue that much of the broader vision aligned surprisingly well with the way mobile devices and television evolved to complement one another over the years.

But now with technology advancing around us, we are getting even closer to that actual imagined reality. Maybe we missed the film’s 2015 deadline, but in the next few years we may very well surpass the film’s original vision (although I can’t commit a timeline for flying cars).

The multiview experience though is a great example of something that has been promised for a long time, but the technology and user behavior just haven’t quite lined up enough to make it feel seamless or essential. This year, however, YouTube seems intent on changing that. Similar to how Netflix has dabbled in broadcast, YouTube has been experimenting with multiview since 2023, but now appears to be placing renewed emphasis on that feature. This will be particularly interesting to see how impactful it is alongside the FIFA partnership later this year.

But it’s not just multiview that is making a play for our attention. What feels more significant is that several long-discussed layers of the media future are starting to mature at the same time. Distribution is becoming smarter. Devices are looking to create a better viewing experience (personally I’m hoping the Samsung Vision AI can actually cut down on content discovery time). Apple’s recent talk of smart glasses along with Meta’s Raybans, makes it seem like wearables might finally start shifting from novelty to interface. And while I’m not particularly excited about living in an episode of Black Mirror, I do appreciate that our wearables look a lot sleeker than the ones in Back to Future II. And of course, AI. AI everything: companionship, personalization, localization, sports analytics, shoppable content. There is so much being touted, but we’ll have to see if it can be meaningfully adopted.

The Convergence Layer

For years, many of these developments felt adjacent rather than connected. Now due to rapid-changing technology, low-latency connectivity and the advancement of compute power, media is converging in an incredibly smart way. For better or worse, we are inching closer to the ambient, always-on, screen-rich future that Marty McFly once stumbled into, only now paired with our own AI-powered reality.

And with these two shifts happening simultaneously, we’re entering a world where experiences are being shaped to engage audiences in more intentional and more immersive ways at the same time. But what fascinates me the most is that these shifts are not actually in conflict. They are complementary. One is about reclaiming the emotional rhythms that made television feel communal and eventful in the first place. The other is about using new technology to expand where, how, and with what depth those experiences can happen.

The companies best positioned for what comes next will be the ones that understand that technology alone is not the advantage; the advantage is knowing how to use it to connect to the audience, how to connect to the humans.

So while we may not be completely returning to Days of Yore or going Back to the Future, we are absolutely incorporating both as we create our New Old West. Doc would be the first one to remind us to be thoughtful about this future we are creating because, “Your future is whatever you make it. So make it a good one.”

So here’s how I’m walking into Las Vegas this year, what I expect to find, and why I’m keeping my working thesis loose on purpose.

The Infrastructure Signal

Last year at NAB the big energy was application-layer AI. Tools you could touch, demos you could watch, workflows you could imagine someone in a real organization actually using. That was appropriate for where the market was. This year I think the signal has moved one level down.

What I’m watching for is AI that doesn’t have a UI. The kind that lives inside a workflow and touches automation without announcing itself. The distinction matters because it changes who holds the leverage. Application-layer AI is a product decision. Infrastructure-layer AI is an architectural one. Products get replaced. Architecture gets locked in.

NAB itself is helping make this argument for me, almost accidentally. The West Hall AI Innovation Pavilions, there are two of them this year, up from one; are still organized around content creation, workflow automation, and audience engagement. That’s the loud surface story. What I’m interested in is the layer underneath it, where the architectural decisions are actually being made.

swXtch.io: The Router Is the Story

One of the things I’ll be watching closely is how swXtch.io’s framing lands this week. They’re debuting groundSwXtch, their multicast networking platform, and the broader product architecture they’ve been building around includes what they describe as an AI-native network — a system where intelligence sits inside the signal path and makes decisions about where content goes and what happens to it in transit, rather than operating as a tool downstream of delivery.

What I’m specifically watching is whether that framing starts to land as an architectural shift rather than just a networking one. The capability that interests me is the ability to simultaneously multicast a live source across multiple workflows and AI-driven processing layers in real time — things like generating optimized outputs for different distribution surfaces from a single production feed, driven by the routing logic itself rather than by a human operator making a downstream edit decision. When the router is doing the reasoning and not just moving the bits, the question of who controls the routing logic becomes a very different conversation than it used to be. I want to see whether operators on the floor are hearing it that way yet, or whether it’s still being received as a sophisticated networking story.

V-Nova: Two Different Wedges, Both Worth Watching

V-Nova has two distinct stories at NAB this year and it’s worth keeping them separate, because they land with different buyers and serve different arguments.

The first is LCEVC. This is the efficiency wedge operators already understand. Better quality at lower bitrates, meaningful bandwidth savings at scale, real deployments across broadcast and streaming. That story has been building for a while and the NAB conversations this year include live TV 3.0 deployments in Brazil and continued work across ATSC 3.0.

The second is VC-6, and this is the more interesting one to me right now because the frame around it is shifting. VC-6’s AI Blueprint approach: decode once, reuse the output across multiple AI models, selectively access only the regions and quality levels each model actually needs. This changes the economics of running visual AI on video at scale in a way that LCEVC doesn’t directly address. The real value proposition isn’t bandwidth. It’s reduced decode cost, lower memory bandwidth, less data movement through the inference pipeline, and better throughput when you’re running multiple models against the same content. That’s an AI infrastructure cost argument, not a network cost argument, and it expands the buyer conversation well beyond the engineering team responsible for delivery. I’ll be at their booth to understand how that argument is actually landing, because the gap between a compelling reframe and a changed purchase decision is where the real story lives.

The Displacement Nobody’s Headlining

The C-band satellite story has been grinding for the better part of a decade. It won’t be the flashy headline at NAB this year, but it might be the most consequential infrastructure conversation on the floor. The FCC has committed to competitive bidding for at least 100 MHz of Upper C-band spectrum by July 2027. That’s the auction deadline, not a clean broadcaster move-out date, and the operational displacement timeline is still being fought through policy and protection proceedings. But the direction is not ambiguous. LTN is at NAB explicitly pitching satellite-to-IP migration as spectrum auction pressure intensifies, saying media organizations are already moving hundreds of channels and live sports workflows to IP alternatives. Zixi is making nearly the same argument with a nearly identical show-floor posture.

What’s actually interesting to me is not the migration mechanics. It’s what changes at the relationship layer when the satellite dish is gone. C-band distribution wasn’t just a delivery technology. It was an organizational logic. Affiliates got content one way. Rights flowed through one channel. When that changes, the question of who manages the new delivery path, who sets the SLAs, who holds the affiliate relationship, who controls the failover is genuinely open. The companies winning that answer are building something more durable than a product.

Folded into this is the provenance thread. C2PA authentication is showing up across streaming and broadcast workflows at the same moment MoQ and low-latency security stacks are becoming more concrete. EZDRM is sitting in the middle of both conversations this week. Their DRMaaS now supports MoQ transport, and they’re separately showing C2PA provenance signing. Which makes that intersection worth watching even if the transport story is still settling. Whether MoQ itself is ready for production at scale is a debate the industry is still having honestly, and the benchmark data from real deployments is only now starting to accumulate. The provenance layer threading through all of it is the signal worth tracking regardless of which transport ultimately carries it.

The North Hall Problem

One of the conversations in my Mate Talk group last week that’s been sitting with me is something Sean McCarthy observed about the North Hall. The broadcast-to-IP convergence that everyone was talking about three or four years ago has actually been happening. It’s just not on any headline panels. Traditional satellite and playout operators are going software-defined, not because they chose to on their own timeline but because the economics forced the question. The interesting work is happening there, and most of the streaming-world people I know spend ninety-five percent of their NAB in the West Hall.

I’m making a point this year of crossing halls and staying longer when I do. The companies bridging those two worlds, who understand legacy broadcast operations deeply enough to translate them into software-defined workflows without breaking what works, are in a better position than they’re getting credit for right now. That gap between what the North Hall is working on and what the West Hall is celebrating is itself a kind of market inefficiency, and market inefficiencies have a way of attracting capital once someone names them clearly enough.

What the After-Hours Is Actually For

Part of what I do at NAB is watch where capital is paying attention, and this year that’s an explicit part of my agenda. I’ll be working alongside the Hallstone Ventures team and spending time with the MonteVIDEO Tech Ventures crew, who bring a genuinely different operator lens to what gets funded and why. Both are doing the same fundamental thing from different vantage points: watching which early conversations over drinks become real in eighteen months, and which ones evaporate before Sunday morning.

Learning to tell the difference and asking the right questions before you can tell, is part of what the show is actually for if you use it right. The ideas that are just being floated, not pitched, not presented, just surfaced by someone who’s been sitting on a thought for two years and hasn’t said it out loud in a room full of the right people yet — that’s where I pay the closest attention. Part of my job is to find those conversations, take them seriously, and figure out whether they’re worth pulling forward. Some aren’t ready. Some are already late. Occasionally one of them turns out to be the thing everybody’s writing about by the time IBC rolls around.

The Working Thesis, Held Loosely

Going into the show, my read is that 2026 is the year the infrastructure layer at NAB gets more interesting than the application layer for the first time in a few cycles. AI burrowing into the routing fabric rather than wearing a UI. Compression economics being reframed around inference pipeline cost rather than bandwidth. Distribution architectures rebuilding from the pipe up as satellite displacement moves from theoretical to unavoidable. Capital watching the early conversations for what comes next.

But I’ve been to thirty of these things. The version of NAB I’m walking into on Saturday and the version I’m writing about next weekend are never quite the same show. Something always surfaces that wasn’t on the agenda. A conversation in a hallway, a demo that reframes something I thought I understood, a question somebody asks at 11pm that turns out to be the most important thing said all week.

That’s the part I can’t plan for. It’s also the part I trust most.

I’ll be posting from the floor in Substack Notes all week. The post-show summary lands the following weekend.

That logic still operates. It just no longer explains where the value is going.

The constraint has shifted, and it is not ownership or distribution, both of which still matter. It is whether the system understands what it owns. Catalogs that cannot be reasoned about, recomposed, or extended at scale are not underutilized. They are structurally inert. The size of the library stops being the asset. The legibility of the library becomes the asset.

The first two parts of this series traced how media pipelines were built around one assumption: humans carry context, machines carry files. Part III showed what changes when context persists instead of being reconstructed at every stage.

Archives Were Never Designed to Be Used

Nobody set out to build a storage system that couldn’t be used. It happened one handoff at a time.

Studios are sitting on decades of accumulated material: original camera footage, alternate cuts, production stems, localization masters, promotional variants, performance captures, behind-camera documentation that never made it into a final deliverable. The volume is real. The problem is not volume.

Rights are distributed across filing systems, contracts, and the institutional memory of people who left years ago. Lineage has been flattened at every stage, because the pipeline was built to move the finished artifact forward, not to carry the decisions that shaped it. The relationship between an original performance and its derivatives still exists, but in most catalogs it survives only as fragments spread across systems, contracts, and memory.

This is the same reconstruction problem that broke the production pipeline in Part I. It did not stop at the pipeline boundary. It extends backward through the entire catalog.

Most media libraries are not underutilized. They are structurally unreadable.

The asset exists. The context that would make it computable does not. And without that context, the archive cannot tell you what you own well enough to use it.

Read more

In 1869, Jay Gould cornered the gold market by telling President Grant’s brother-in-law one thing and his own trading partners another. When the scheme collapsed on what became known as Black Friday, Gould didn’t apologize. He reframed. The broken promises weren’t deception, he argued. They were reasonable adaptations to a rapidly changing environment.

I thought about Gould this morning while reading Ronan Farrow and Andrew Marantz’s investigation of Sam Altman in The New Yorker. The details are different. The mechanics are not. The investigation is built on interviews with more than a hundred insiders and two previously undisclosed document sets. The first is a 70-page memo compiled by former Chief Scientist Ilya Sutskever. The second is over 200 pages of internal notes written by Dario Amodei during his time at OpenAI. Both paint a picture of a leader who tells different stories to different rooms and treats the gap between them as operational friction rather than a character problem.

When asked to acknowledge a pattern of deception after being fired, Altman reportedly told the board, “I can’t change my nature.” One director interpreted that as an admission, not a defense. Multiple interviewees independently used the phrase “antisocial personality.” A Microsoft executive reportedly said there’s a real, if small, chance Altman will be remembered alongside Bernie Madoff and Sam Bankman-Fried.

That comparison is probably wrong, but it’s wrong in an interesting way. The Farrow piece doesn’t document fraud. It documents something more structurally dangerous.

The Cycle, in Fast-Forward

Tim Wu’s The Master Switch argues that every major communications technology follows the same arc. Open invention, commercial exploitation, consolidation under a single corporate entity, then control. It happened with telephony under AT&T, with radio under RCA, with television under the networks. The pattern is so consistent that the only variable is how long it takes and whether anyone intervenes before the window closes.

OpenAI is running Wu’s cycle at startup speed. Founded as a nonprofit to prevent AI from being monopolized. Converted to a capped-profit entity. Then uncapped. Then lobbying against the very regulations its CEO publicly endorsed before Congress. The name itself, “Open” AI, is the residue of a structural promise that has been quietly dissolved at every stage.

The stakes here dwarf the earlier versions of this pattern. Theodore Vail consolidated telephony but delivered universal service in the process. David Sarnoff locked up radio and built the country a broadcast infrastructure along the way. Those were real tradeoffs with real outcomes on both sides of the ledger. Altman hasn’t delivered AGI, and the economics of the attempt are increasingly shaky. The Information reported this week that CFO Sarah Friar has told colleagues she doesn’t believe the company is ready for a public offering. When the person managing the books isn’t sure the numbers work, the trust question stops being abstract. It becomes a capital structure question.

The Factory Owner Who Says He’s a Worker

Brian Merchant’s Blood in the Machine draws a line from the English factory owners of 1811 to today’s tech moguls. The move is always the same. Present the adoption of automated machinery as an inevitability driven by progress rather than a choice driven by the desire to capture value. The factory owners told their workers that the machines were coming regardless, and that it was better to have reasonable men in charge of the transition. Altman tells the public the same thing at civilizational scale.

The Farrow investigation adds a layer that Merchant would recognize immediately. It’s the safety theater. The alignment team, publicly promised 20% of OpenAI’s compute, was apparently operating on roughly 1-2% of it, mostly on the company’s oldest hardware. When journalists asked to speak with someone working on existential safety, an OpenAI spokesperson responded with a question of their own. “What do you mean by ‘existential safety’? That’s not a thing.”

A company navigating genuine tension between progress and safety doesn’t say that. A company that made the choice a long time ago and is now managing the optics does.

Why This Matters to Infrastructure

The temptation is to treat the Altman story as personality-driven noise on top of a compute buildout that proceeds regardless of who’s running it. The pipes don’t care about character flaws. The models get trained. The data centers get built. The inference endpoints come online.

But that framing misses the structural point. Concentrated control over a communications platform creates a single point of failure for the entire ecosystem built on top of it. Every content company, every distribution platform, every media workflow that touches AI is building dependency on a layer whose governance is, according to the people closest to it, unstable. Wu’s cycle matters precisely because the monopolist’s personality is irrelevant. The structure is the problem.

The investigation also surfaced a detail that should land differently with this audience. Around 2018, OpenAI’s leadership seriously discussed something they called the “National Plan.” The idea was to have major governments, including China and Russia, bid competitively for access to AI technology. The internal framing was blunt. The goal was to create a prisoner’s dilemma where every country would have to fund OpenAI. The plan was shelved only because employees threatened to quit.

That reveals an organization whose leadership views geopolitical leverage as a business strategy, held in check not by governance or principle but by the willingness of individual employees to walk away.

The Window

Gould’s contemporaries understood what he was doing. Newspapers wrote about it. Competitors complained. Regulators investigated. None of it mattered until the system itself broke badly enough to force structural change. The investigations were documents of a window closing, not interventions that kept it open.

The Farrow piece reads the same way. It is meticulous, sourced to an extraordinary degree, and likely to change nothing in the short term. Altman will remain CEO. The IPO preparations will continue. The compute buildout will proceed. The piece will be discussed for a week and then absorbed into the background noise of an industry that has already decided the technology matters more than the person controlling it.

And Wu’s history tells us that the moments where that felt most obviously true were exactly the moments where the person’s control became permanent.

The window closes when nobody’s watching. Right now, almost everybody is watching the technology instead.

Further reading: Gary Marcus’s response to the investigation. Peter Kafka’s analysis at Business Insider.

If you’ve worked in business over the last decade you’ve heard data compared to things like oil or electricity. You may have heard that our data needs to be moved to a warehouse or a lake. But data does not inherently create value when it sits in place; it creates value when it moves through the organization to inform and shape decisions. Too many leadership teams in legacy companies still treat data as a storage problem. I hear discussions about “centralizing” the data, when the real challenge is building a decision system that gets the right information to the right person at the right time. In a healthy company, data should function less like inventory on a shelf and more like the body’s circulatory system, moving continuously through t…

Read more

Not whether AI replaces human presence. But what human presence was actually doing.

It was doing two things at once: performing a function, and conferring legitimacy. A communications director handled press relations and signaled that someone was accountable to questions. A film rating reflected curatorial judgment and told parents they could extend trust. An independent media show covered an industry honestly because it answered to no one in it. These two things moved together so reliably that most institutions stopped noticing they were separate.

This week’s signals each show a different attempt to hold onto the second thing after the first has been removed, automated, or never built.

Oracle Cuts 30,000 to…

Read more

For most of the industry’s history that limitation was manageable. Media traveled through the pipeline as relatively stable objects. A film was edited. A show was localized. A marketing campaign produced a handful of variants. The number of transformations remained small enough that teams could reconstruct the story of an asset when necessary.

AI changes the shape of that workflow.

Media is no longer moving through pipelines as static objects. It is evolving inside systems that continuously reinterpret it. Scenes are recomposed. Dialogue is translated and re-performed. Visual elements are expanded, cropped, or restaged. Marketing assets multiply across formats and platforms. Personalized versions appear for different audiences.

In that environment, the file is no longer the primary artifact. The system that understands how the media evolved becomes the primary artifact.

Media begins to behave less like a collection of files and more like a context-aware system.

Read more