The next article initially appeared on the Asimov’s Addendum Substack and is being reposted right here with the creator’s permission.
Invoice Gurley has a superb article on what he calls open supply technique, which we suggest studying. There’s a lot to debate about his concluding argument specifically: that open-weight fashions are central to preserving the AI market rent-free. The bounds of open-weight AI as the first open supply technique are absolutely appreciable although, if it nonetheless requires costly {hardware} to run on, and if the structure finally stays monolithic—reasonably than composable and protocol-centric.
A associated consideration comes from Anthropic’s current acquisition of Stainless—a startup that generates SDKs, command-line instruments, and MCP servers from API specs. This illustrates that open protocols like MCP, even when publicly ruled,1 stay uncovered at their complementary layers to personal actors capturing rents. (Protocol openness doesn’t get rid of this and as an alternative in all probability allows it, by enabling market progress).
We requested Claude to investigate this acquisition, going past the press releases. Its first move overstated elements of the competitive-denial story; what follows is what survived it taking a more in-depth look:
- Complement seize, not protocol seize. MCP—the usual that lets AI brokers discuss to different software program—stays open, and its governance has been handed to an unbiased basis. What Anthropic purchased is the corporate that turned that normal into one thing most builders might really use. Stainless was the dominant software for taking an strange enterprise API (say, a resort reserving system or a buyer database) and changing it into one thing an AI agent might name by MCP. The open normal remains to be open. The trail most builders walked to make use of it has now been purchased.
- This isn’t a one-off—the entire layer is consolidating. Stainless wasn’t alone on this market. Its most important competitor, Fern, was purchased by Postman in January 2026. Anthropic purchased Stainless 4 months later, in Might 2026. That leaves Speakeasy as the one main unbiased participant, plus an open-source fallback known as OpenAPI Generator that the majority builders think about too tough for manufacturing use with out vital handbook work. In underneath 5 months, two of the three critical firms on this a part of the market have been absorbed into bigger platforms. The Stainless deal is extra seen due to who purchased it and why, however the broader sample issues extra: a complete layer of AI infrastructure is being pulled inside platform homeowners.
- Moat migration. The hole in uncooked mannequin functionality between Anthropic, OpenAI, and Google has narrowed significantly and continues to shut, and the implication is that mannequin high quality alone is unlikely to be the principal foundation of aggressive benefit over the following two years. What could distinguish the main companies as an alternative is the standard of the developer expertise round their fashions: how simply a enterprise or an engineer can construct one thing helpful on prime of a given mannequin, how cleanly the tooling integrates with current techniques, and the way dependable the connectors are over time.
- Pricing logic, with caveats on denial. Stainless was final valued at $150M in December 2025; at >$300M 5 months later, it is a roughly 2x strategic markup, not acqui-hire arithmetic. Eradicating a critical-path exterior dependency on Anthropic’s personal SDKs, whereas denying it to a good set of rivals, is rational at that worth—however the denial logic is partial. Speakeasy is a viable substitute, and OpenAI was reportedly already migrating off Stainless. The friction tax falls hardest on smaller gamers who lack the engineering bench to soak up migration value.
Stainless was based by Alex Rattray, previously of Stripe. Stripe constructed its market place largely on unusually well-designed developer instruments, and Stainless was, in impact, an try to use the identical strategy to the layer between AI APIs and the remainder of the software program economic system. Anthropic has acquired the group that is aware of how to do that.
…The press launch calls it “extending attain”; the InfoWorld learn—“last-mile developer expertise”—is nearer, however the complement-capture part, even when partial, is actual.
-*-
Now, whereas Claude is perhaps overstating a number of the market dangers related to this acquisition (you inform us?), it exhibits that open supply’s impacts are extremely conditional on its dependencies and may by no means be analyzed in isolation from the market’s software program stack and structure. That is equally true for open weight fashions—being depending on information, compute, and distribution—as it’s for open protocols like MCP, depending on fixed API translations and entry. Monitoring these interdependencies is what a full ecosystem view entails and is useful to undertake in an effort to think about the place chokepoints would possibly come up, and in flip, the place open supply technique would possibly finally fail or be captured.
