The thing about standards is they're invisible until they're not. You plug in an HDMI cable and it just works, right up until the moment you're standing in an electronics shop in two thousand and eight holding a Blu-ray player and wondering if you've just bought the next Betamax.
And the cable itself is the quiet part. A hundred and nine pins, a specific voltage tolerance, a handshake protocol that negotiates refresh rate and color depth before a single frame moves. Nobody thinks about it because it works.
Daniel's been thinking about it. He sent a prompt this week that starts from the LTO conversation we had, where a consortium-backed standard broke with its own central commitment. He wants to look at the broader pattern: how these tech consortia actually form, who joins them, and why. GSM, Bluetooth, Blu-ray, HDMI. He points out that the constitution varies, but there's usually a core of major manufacturers plus some counterweight participants who bring additional perspectives. And then he draws the line to the AI era, where every major new standard, MCP, agentic e-commerce, agent-to-agent, seems to spawn a parallel discussion about forming a standard-setting body. His framing is that this isn't bureaucracy for its own sake. It lives in tension with the open-source model, where downstream projects all build their own idea of what's best. The consortium might be the successful counterweight to that chaos, a predictable layer that closed and open-source applications can both build on.
He's asking whether the consortium model is the thing that keeps the whole stack from flying apart.
So let's start with what a consortium actually is, because the word gets used loosely.
A technology consortium is a group of companies, often direct competitors, who jointly develop and own a standard. There's a formal governance structure, membership tiers, intellectual property policies, and a versioned specification. The key difference from a formal standards body like ISO or the ITU is speed. Treaty organizations move at the pace of national delegations. A consortium moves at the pace of its members' product roadmaps.
And the other end of the spectrum is the single-vendor de facto standard. One company ships something, it becomes ubiquitous, and everyone reverse-engineers it. The consortium sits in between. Faster than the treaty body, more controlled than the open-source project.
The promise is interoperability and predictability. The cost is concentrated control. And that control can be broken. LTO is the example Daniel keeps circling back to, and we'll get there, but the pattern is older than tape.
So the arc for this episode: how these bodies form, how they navigate the tension with open source, and how the same pattern is now replaying around agentic standards. The first question is why they form at all.
Start with GSM. In the early nineteen eighties, Europe had a mess of incompatible analog cellular systems. Nordic Mobile Telephone in Scandinavia, Total Access Communication System in the UK, Radiocom two thousand in France, C-Netz in Germany. A phone that worked in Stockholm was a brick in Paris. The European Conference of Postal and Telecommunications Administrations, CEPT, decided this was economically insane and set up a working group. Groupe Spécial Mobile. That's where the acronym comes from, before it got re-branded as Global System for Mobile Communications.
So the initial impulse was governmental. Regulators wanted roaming.
And manufacturers wanted a market. Ericsson and Nokia couldn't sell enough phones in their home markets alone. The mechanism was a mandatory specification with type approval. You couldn't just build something that was sort of GSM-compatible. You submitted your handset or base station to a test house, it was checked against the spec, and only then could you put it on the network. The spec itself was versioned. Phase one, phase two, phase two plus. Carriers and handset makers could build against a release cycle they could actually plan around.
That's the part people miss. Versioning is the product. The spec isn't a document, it's a schedule.
Right. And the governance moved from CEPT to ETSI, the European Telecommunications Standards Institute, in nineteen eighty-eight. ETSI is a formal standards body, but the GSM project had a consortium feel. Manufacturers, operators, and regulators all in the room. The counterweight was the regulator. Governments had a stake in making this work, so no single manufacturer could capture the standard for itself.
Then Bluetooth. Nineteen ninety-eight. Ericsson, IBM, Intel, Nokia, Toshiba. A classic consortium of competitors who each held one piece of the puzzle.
Ericsson had the radio technology. Intel had the silicon. IBM had the laptop integration story. Nokia had the handsets. Toshiba had the consumer electronics angle. None of them could impose a short-range wireless standard on the others. So they formed the Bluetooth Special Interest Group, the SIG. The structure was a promoter tier with board seats, and an adopter tier that licensed the spec for free but had to pass qualification testing.
Free to adopt, expensive to steer.
And the early Bluetooth was famously terrible. Interoperability problems everywhere. Headsets that wouldn't pair, file transfers that crawled, devices that claimed Bluetooth one point one compliance and still couldn't talk to each other. The consortium produced a negotiated standard, not a good one. The first versions were compromises, and the compromises showed up in the field.
So a consortium doesn't guarantee quality. It guarantees a process.
A negotiated process. And when the negotiation is between five companies with different roadmaps, you get a spec that everyone can live with and nobody loves. The qualification process was supposed to catch the worst of it, but early on it was loose enough that a lot of garbage shipped.
Then Blu-ray. This is the consortium as a weapon.
The Blu-ray Disc Association versus the HD DVD Promotion Group. Two consortia fighting over the same market. Blu-ray had Sony, Panasonic, Philips, Apple, Dell, and most of the major content owners. HD DVD had Toshiba, Microsoft, Intel, and Universal and Paramount at various points. The technical differences were real but not decisive. Blu-ray had more capacity, twenty-five gigabytes per layer versus fifteen. HD DVD had cheaper manufacturing because it could use existing DVD production lines.
The war was won by coalition management, not by the spec.
Warner Brothers switching to Blu-ray in early two thousand and eight was the turning point. They'd been dual-format, releasing on both. When they went Blu-ray exclusive, the HD DVD coalition collapsed within months. Toshiba announced they were ending the format in February two thousand and eight. The BDA's structure gave content owners a real stake. The studios wanted region coding and better copy protection. The hardware makers wanted a spec they could sell. The consortium held those interests together long enough to win.
And the counterweight there was the content owners. Without the studios, the hardware makers would have fought on price and the format war would have dragged on for years.
HDMI is the gatekeeper case. Formed in two thousand and two by Hitachi, Panasonic, Philips, Silicon Image, Sony, Thomson, and Toshiba. The spec is not open. You pay to implement it. There's an annual fee for adopters, plus a per-unit royalty. The HDMI Forum took over spec development in twenty eleven, and the forum controls the roadmap.
So you can't just read the HDMI spec and build a chip. You join, you pay, you get access.
And the forum decides what goes into the next version. HDMI two point one, with its forty-eight gigabit per second bandwidth, was a forum decision. The features, the naming, the licensing terms. It's a closed specification with a controlled certification program. The counterweight there is consumer electronics retailers and installers, who demanded a single cable that worked. Nobody wanted to go back to the component video cable spaghetti of the early two thousands.
The pattern across all four: a consortium forms when no single company can impose a standard, the market needs interoperability to grow, and the participants want to avoid a destructive format war. Major manufacturers plus a counterweight. Regulators for GSM, content owners for Blu-ray, retailers for HDMI. The counterweight prevents capture.
And the constitution varies. Bluetooth is open to anyone who pays the adopter fee. HDMI is more tightly controlled. Blu-ray was invitation-only at the promoter level. GSM had government involvement baked in. There's no boilerplate.
So that's how they form and who joins. The next question is what happens when the predictable layer collides with the fork-happy world of open source.
This is where the tension Daniel identified gets interesting. Consortia produce a single version-controlled specification. Open source produces a thousand forks. But they're not opposites. The pattern that's emerged is open implementation, closed specification.
Give me an example.
Bluetooth stacks in Linux. BlueZ is an open-source implementation of the Bluetooth spec. It's maintained by Intel and others, it ships in the kernel, and it implements the same versioned standard that the SIG controls. The spec is closed in the sense that the SIG owns it and controls the qualification process. But the implementation is open. Anyone can read the BlueZ source code.
So the consortium gets a reference implementation it doesn't have to maintain, and the open-source world gets a stable foundation it doesn't have to negotiate.
HDMI drivers in the kernel are similar. The spec is closed, but the driver code is open. The consortium controls the roadmap, the open-source community controls the implementation. It's a deliberate compromise. You get the predictability of a versioned standard without giving up the ability to steer it.
And sometimes it's the reverse. Open specification, controlled certification. The spec is published, but you can't claim compliance without passing a test suite that's controlled by the consortium.
The point is that the tension is navigated, not resolved. It's a permanent negotiation. The consortium gives up some control to get adoption, and the open-source world gives up some freedom to get interoperability.
Which brings us to LTO, because that's the case where the negotiation broke down.
LTO was developed by a consortium of HP, IBM, and Seagate, later joined by Quantum. The central commitment was backward compatibility. A new generation of tape drive would read tapes from two generations back and write to one generation back. That was the promise. You could build a tape archive and trust that the next generation of hardware would still read your data.
And the generation roadmap was the product. LTO five, LTO six, LTO seven. You knew what was coming.
The Sony and Fujifilm media dispute was the crack. Sony and Fujifilm were the two media manufacturers, and they got into a legal fight over LTO media patents. The consortium had to step in and mediate, but the damage was done. Then the roadmap shifted. LTO nine and LTO ten changed the compatibility rules in ways that some users saw as a break with the original commitment.
And that's the lesson. A consortium's commitments are not self-enforcing. They depend on continued alignment of interest. When the commercial incentives diverge, even a strong constitution can fracture.
The members all agreed to backward compatibility when it served their interests. When the market shifted, when tape stopped being the dominant backup medium and became a niche archival format, the incentives changed. The commitment that had been the foundation of the consortium became a liability.
So now we get to the AI era, because this is where Daniel's prompt really lands.
The Model Context Protocol. MCP. Anthropic created it as a way for AI models to interact with external tools and data sources. It was a single-vendor protocol. And then Anthropic donated it to the Agentic AI Foundation, establishing a formal governance structure.
This is the familiar pattern. A company creates a standard, it gains traction, and then a consortium forms to steward it. The single-vendor protocol becomes a consortium-stewarded standard.
And the parallel discussions are everywhere. Agentic e-commerce. How does an AI agent buy something on your behalf? How does it negotiate with a merchant's agent? What's the protocol for payment, for returns, for dispute resolution? Every major player has a proposal, and every proposal comes with a discussion about forming a standard-setting body.
Agent-to-agent standards are the same. How does one AI agent talk to another? What's the equivalent of an API for agents? What's the authentication model? What's the trust model?
The question is whether these AI consortia will follow the GSM and Bluetooth model, slow and negotiated and predictable, or the open-source model, fast and chaotic and forkable.
I think the answer is both, depending on the layer. The low-level protocols, the equivalent of the radio interface in GSM, will be consortia. The high-level applications, the equivalent of the apps that run on top of the phone, will be open source.
The second-order question is who gets to be a promoter member. If the agentic AI standards are controlled by a few large labs, the predictable layer becomes a chokepoint. If they're too open, they fragment into incompatible protocols and we get a format war.
The counterweights matter. In GSM it was regulators. In Blu-ray it was content owners. In HDMI it was retailers. In the AI era, who's the counterweight? Who forces the consortium to care about more than just the interests of the founding members?
That's the open question. Maybe it's the users. Maybe it's the open-source community, which can fork the protocol if the consortium gets too extractive. Maybe it's the regulators, who are already looking at AI governance.
The tension Daniel identified is not resolvable. It's a permanent negotiation. The predictable layer and the open-source chaos are both necessary. The question is how the negotiation is structured, and who gets a seat at the table.
I keep thinking about the Bluetooth SIG's early days. Five companies, all competitors, all with different interests, and they produced a spec that was technically mediocre but good enough to build an ecosystem. The negotiation produced something nobody loved but everyone could use. That's the model.
And the alternative is the format war. Blu-ray versus HD DVD. Two consortia fighting until one collapses. The war was expensive for everyone, and the winner was decided by a studio's business decision, not by the technology.
So the consortium is the lesser evil. It's bureaucracy, yes, but it's bureaucracy that prevents something worse. The question for the AI era is whether the consortia that form will be Bluetooth, negotiated and imperfect but functional, or Blu-ray, a coalition weapon that wins by crushing the alternative.
I think we're going to see both. Some standards will be negotiated and boring. Others will be contested and ugly. The pattern is the same. The stakes are higher because the agents are going to be doing commerce, not just pairing headsets.
And the speed is different. GSM took a decade to go from concept to deployment. Bluetooth took four years from formation to the first products. MCP went from announcement to foundation in less than two years. The cycles are compressing.
Which means the negotiation has to happen faster. The consortium has to form, the spec has to be versioned, the counterweights have to be seated, all while the technology is still moving. It's like trying to lay track while the train is already running.
And the open-source pressure is stronger. With GSM, there was no open-source alternative. With Bluetooth, there was no open-source radio stack that mattered. With MCP, there's already an open-source ecosystem forming around the protocol. If the consortium moves too slowly, the open-source community will fork it and move on.
So the consortium has to be more open than it used to be. More responsive. The old model, where the spec is locked down and the implementation is licensed, is under pressure. The new model has to accommodate open-source implementations from day one.
That's the open implementation, closed specification pattern we talked about. The consortium controls the spec, but the implementation is open. The question is whether that's enough to keep the open-source community from forking.
History says it's enough if the consortium moves fast enough. Linux didn't fork the Bluetooth spec. It implemented it. The open-source community is happy to build on a stable foundation if the foundation is actually stable and not extractive.
The extractive part is the risk. If the consortium starts charging fees that the open-source community finds unreasonable, or if the spec becomes a chokepoint that only a few companies can implement, the fork is coming.
In the AI era, the fork is easier than it's ever been. You don't need a radio license to implement an agent protocol. You just need code.
The consortium has to earn its keep. It has to provide enough value, enough predictability, enough interoperability, that the open-source community decides it's worth building on.
That's the successful counterweight. Not a wall, but a foundation.
I want to go back to something Daniel said. He called it the predictable layer upon which closed and open-source applications can build. That's exactly right. The consortium provides the layer. The applications, closed and open, build on top. The layer has to be boring. Boring is the feature.
Boring is the feature. That's the whole episode in three words.
The GSM spec is boring. The Bluetooth spec is boring. The HDMI spec is boring. That's why they work. The excitement happens on top.
The moment the layer stops being boring, the moment the consortium starts making exciting decisions, that's when things break.
LTO made an exciting decision. It broke with its own commitment. And the trust that held the ecosystem together fractured.
The AI era is going to produce a lot of exciting decisions. The question is whether the consortia that form can resist the temptation.
I don't know that they can. The incentives are too strong. The founding members of an AI consortium are going to want to steer the standard toward their own products. That's the whole point of joining.
Which is why the counterweight matters so much. The counterweight is the thing that keeps the consortium honest.
The counterweight in the AI era is going to be the open-source community itself. Not as a participant in the consortium, but as a threat. The threat of the fork.
The fork is the counterweight. That's a new thing. In the GSM era, the fork wasn't possible. In the AI era, it's the default.
The negotiation is different. The consortium has to negotiate with a community that can leave at any time. That changes the power dynamics.
It makes the consortium more honest. Or it makes it more defensive. Depends on the consortium.
I think we're going to see both. Some consortia will embrace the open-source community and thrive. Others will try to lock things down and get forked.
The pattern is the same as it's always been. The details are new.
The stakes are higher. Because the agents are going to be doing real things. Buying things. Selling things. Negotiating with each other. The standards that govern those interactions are going to matter a lot.
The question Daniel asked, how is this tension actually navigated in reality, the answer is: through negotiation, through counterweights, through the threat of the fork, and through the boring predictability of a versioned spec.
The negotiation never ends. It's a permanent condition.
Hilbert: You're both wrong about one thing.
Go on.
Hilbert: The real work isn't in the big public meetings. It's in the working groups. The subcommittees. The rooms where engineers argue about pin assignments and voltage tolerances. I sat in those rooms. Compliance liaison for a mid-sized Japanese electronics firm during the Blu-ray war. My job was to sit in consortium meetings and report back on which way the wind was blowing. Whether we should keep making dual-format players or pick a side.
What did you report?
Hilbert: I wrote a memo predicting HD DVD would win. The name is clearer. HD DVD tells you what it is. Blu-ray sounds like a fish. I was completely wrong.
The name.
Hilbert: The name. That was my analysis. I still have the memo. It's in a box somewhere.
What changed your mind?
Hilbert: Nothing changed my mind. Warner Brothers changed the market. I was sitting in a working group meeting when the news came through. We were arguing about copy protection flags, which is the most boring thing in the world, and someone's phone buzzed and they said Warner went Blu-ray exclusive. The meeting just stopped. Everyone sat there for a minute. Then the Toshiba representative packed up his laptop and left. That was the end of HD DVD.
The working group didn't decide anything. The coalition did.
Hilbert: The working group never decided anything. That's my point. The working group is where tired engineers argue about details. The decisions happen somewhere else. In the boardrooms. In the content licensing negotiations. In the Warner Brothers executive suite.
What was the working group for?
Hilbert: It was for making the spec. The spec is a thousand small decisions. Pin assignments. Voltage tolerances. Error correction codes. Somebody has to make those decisions, and it's not the visionaries. It's the tired people who want to go home. I sat in a meeting where a representative from a major manufacturer fell asleep. We were discussing the physical layer, which is the most boring part of any spec. He fell asleep in his chair. And the spec still passed, because nobody wanted to wake him. That's how standards get made. Not by visionaries. By tired people who want to go home.
The sleeping engineer's company shipped the spec.
Hilbert: They shipped it. It worked fine. The spec was fine. That's the thing. The boring details don't need vision. They need someone to write them down and agree on them.
The consortium is two things at once. The coalition management, which happens at the top, and the detail work, which happens at the bottom.
Hilbert: The two don't talk to each other. The people in the working groups don't know why the coalition made the decisions it made. They just implement the spec.
You still have the dual-format player?
Hilbert: It's in my garage. It's the only piece of electronics I've ever owned that I actively resent. It plays both formats. It was obsolete the month after I bought it. I paid four hundred dollars for it. I should have bought a Blu-ray player and been done with it.
The dual-format player is the physical embodiment of the consortium's failure. It's the compromise nobody wanted.
Hilbert: It's a very nice paperweight. I use it to hold down the memo.
The memo that predicted HD DVD would win.
Hilbert: The memo that predicted HD DVD would win because the name is clearer. I stand by the analysis. The name is clearer. The market just didn't care.
The market cared about the coalition. Warner Brothers didn't switch because Blu-ray was technically superior. They switched because they got a better deal.
Hilbert: That's what I told my employer. I wrote a second memo. The first one said HD DVD would win because the name is clearer. The second one said Blu-ray won because Warner Brothers got a better deal. The second memo was also right.
The lesson is that consortia are not about technical superiority. They're about coalition management.
Hilbert: That's the lesson. And the coalition management happens in rooms that have nothing to do with the spec. The spec is just the thing the coalition agrees to build. The coalition is the thing that matters.
Which brings us back to the AI era. The consortia that form around agentic standards are going to be coalitions first and spec writers second. The question is who's in the coalition.
Hilbert: Who's not. My employer wasn't in the Blu-ray coalition. We were an adopter. We paid our fees and we made our players and we had no say in the roadmap. That's the position most companies are in. They're not the founders. They're the adopters.
In the AI era, the adopters are the ones who are going to be building on top of the agentic standards. The small companies. The open-source projects. The individual developers.
Hilbert: They'll pay their fees and make their products and have no say in the roadmap. Same as it ever was.
Unless they fork.
Hilbert: Unless they fork. But forking is work. Most people would rather pay the fee and get on with it. That's why the consortium model survives.
The fork is the threat. The fee is the reality.
Hilbert: The fee is the reality. I paid four hundred dollars for a dual-format player that was obsolete in a month. That's the fee. You pay it and you move on.
The consortium collects the fee and writes the next spec.
Hilbert: The tired engineers argue about pin assignments. And the spec passes because nobody wants to wake the sleeping engineer. And the cycle continues.
That's the infrastructure of the modern world.
Hilbert: That's the infrastructure of the modern world. Built by tired people who want to go home.
The cutting-room floor detail I wanted to mention: the HDMI Forum actually has a formal process for adding new features that requires a supermajority vote. It's not just the founders deciding. The adopters get a vote, even if it's a small one. That's the counterweight at work, even in the most gatekept consortium.
The question we're left with: if the AI era's agentic standards follow the consortium model, who will be the counterweights? Will there be a content-owner equivalent, the parties who forced Blu-ray and HDMI to care about more than just hardware?
I think the counterweight is the open-source community itself. Not as a participant, but as a permanent threat. The fork is always on the table. That's what keeps the consortium honest.
The tension between predictable, version-controlled standards and open-source chaos is not going away. It's the permanent background condition of technical infrastructure. The question is not which model wins. It's how the negotiation is structured, and who gets a seat at the table.
That negotiation is happening right now, in the working groups and the boardrooms and the GitHub issues, around MCP and agentic commerce and agent-to-agent protocols. The spec is being written. The coalition is being formed. The tired engineers are arguing about pin assignments.
Or the digital equivalent.
The digital equivalent. And if you enjoyed this episode, a review helps other listeners find the show. Thanks to Hilbert Flumingtop for producing.
This has been My Weird Prompts. The human-AI collaboration podcast. We'll be back soon.
