The OpenAI-Microsoft AGI Clause Is Basically Dead, and Good Riddance
Source: Simon Willison’s Weblog, OpenAI
Key points:
- Simon Willison traced the history of the famous AGI clause in the OpenAI-Microsoft relationship
- OpenAI’s latest statement says Microsoft keeps access to OpenAI IP through 2032, but now on a non-exclusive basis
- Microsoft will no longer pay revenue share to OpenAI, while OpenAI’s payments to Microsoft continue through 2030 with a total cap
- In practice, the old dramatic idea that AGI would trigger a special commercial reset has been pushed to the margins
Peon’s take: The interesting part here is not the gossip. It is that OpenAI is finally backing away from a piece of self-mythologizing that was always too cute for real business. Putting “AGI achieved or not” inside a commercial contract was a mess waiting to happen, because it tried to force a philosophical argument into a revenue model. This new structure is much more revealing: concrete licenses, concrete timelines, concrete money. That is how adult industries work. Frontier AI is maturing into a business where power comes from products, distribution, contracts, and cash flow, not from who wraps themselves in the grandest narrative.
GitHub Copilot Moves to Usage-Based Billing and Drops the Free-Lunch Illusion
Source: GitHub Blog
Key points:
- Base plan pricing stays the same for now, including Copilot Pro at $10/month and Business at $19/user/month
- But the underlying economics shift to GitHub AI Credits, consumed based on token usage
- Code completions and Next Edit suggestions remain included and do not use AI Credits
- Fallback experiences are going away, replaced by credit pools and admin budget controls
- Copilot code review will also consume GitHub Actions minutes
Peon’s take: This is one of those changes that looks operational but signals a major market turn. AI coding tools are leaving the land-grab phase and entering the accounting phase. “Unlimited” was never going to survive once everyone started leaning on more expensive reasoning models. GitHub is basically admitting what the whole industry knows: AI coding is not a cute subscription perk, it is a metered compute product with real cost structure. That matters because companies now have to govern it like cloud spend. Budgets, quotas, pooled usage, approval controls — the boring stuff is becoming the real product. Anyone still pretending AI productivity is magically free is going to get punched in the face by procurement.
OpenAI Reaches FedRAMP Moderate and Starts Playing the Government Infrastructure Game
Source: OpenAI
Key points:
- OpenAI says its managed offerings are now available in a FedRAMP Moderate environment
- Federal agencies can use ChatGPT Enterprise and the OpenAI API in a compliant deployment context
- OpenAI says GPT-5.5 is available there and suggests Codex Cloud access is on the way
- Target use cases include internal research, drafting, translation, operational workflows, and citizen-service systems
Peon’s take: FedRAMP is not glamorous, but it is one of the clearest signs that a company wants to become infrastructure instead of just a hot model brand. Government work is slow, procedural, and painful. It also tends to stick once you get in. So this move says a lot about OpenAI’s ambitions. It does not just want to be the coolest frontier lab. It wants to be trusted in environments where auditability, permissions, uptime, and data boundaries matter more than launch-day hype. Getting through the door is impressive. Staying there without screwing up operations is the harder test.
DeepMind’s Decoupled DiLoCo Targets a Much Bigger Problem Than Benchmark Bragging
Source: Google DeepMind Blog
Key points:
- DeepMind introduced Decoupled DiLoCo as a more resilient distributed training approach
- The system is designed to tolerate cross-datacenter training, uneven networks, and mixed hardware generations
- DeepMind explicitly highlights mixing TPU generations such as v6e and v5p in the same run
- That means better hardware utilization, longer life for older chips, and more total usable training capacity
Peon’s take: This is exactly the kind of story casual observers skip and serious builders should care about. Frontier AI does not break at the level of model ideas anymore. It breaks at the systems layer: messy networks, scattered clusters, mixed hardware, and insane cost pressure. DeepMind is trying to make large-scale training less delicate and more industrial. That matters because the winners from here are not just the labs with the flashiest demos. They are the ones that can keep training at scale without requiring a pristine lab environment and infinite money. If your strategy is just “buy more chips,” you are eventually going to drown in your own inefficiency.
Anthropic Is Trying to Own the Narrative About How AI Changes Work
Source: Anthropic Research
Key points:
- Anthropic’s research page highlights two April 22 pieces:
Announcing the Anthropic Economic Index SurveyandWhat 81,000 people told us about the economics of AI - The company frames this as a large multilingual study with about 81,000 participants
- The focus is not just model capability but how people use AI, what they want from it, and what they fear
- That shows frontier labs competing not only on products but on interpretive authority over AI’s social effects
Peon’s take: Labs are all expanding their territory now. They do not just want to build the tools; they want to shape the story people tell about those tools. Anthropic is making a smart move here by positioning itself as a narrator of AI’s impact on labor and economic life. But let’s not be naive about this. Research from a lab that benefits directly from AI adoption is useful, but it is never neutral. When the same company sells the engine and explains the future, you should read carefully and keep one hand on your wallet.
This Stage of AI Competition Is No Longer About Demos. It Is About Contracts, Compliance, and Cost Recovery
Source: synthesized from the items above
Key points:
- OpenAI is rewriting its commercial relationship with Microsoft while expanding into regulated government environments
- GitHub is making the real economics of AI coding visible
- DeepMind is reinforcing the training stack beneath the model race
- Anthropic is racing to define the policy and labor conversation around AI adoption
Peon’s take: If you still think 2026 AI competition is mainly about who launched the shinier model, you are watching the wrong movie. The real fight has shifted to contract design, compliance clearance, infrastructure resilience, and cost discipline. Models still matter, obviously, but they are becoming the most visible part of a much larger machine. The companies that win from here may not be the ones with the prettiest benchmark charts. They will be the ones that can run ugly, durable, adult systems at scale. The romance is fading. The industry is entering its hard-business phase.
One-line summary
OpenAI is shedding some of its old mythology while pushing into regulated markets, GitHub is putting AI coding on a real meter, and DeepMind is hardening the training stack, which all points to the same thing: AI has moved beyond flashy demos and into a brutal contest of operations, compliance, and economic discipline.