Anthropic's First Acquisition Is Stainless, the SDK Startup Used by OpenAI and Google

Stainless has a quiet job that almost no one outside developer circles has heard of. It writes the code that developers use to talk to AI services.

Not the AI models themselves. Not the infrastructure that runs them. The SDKs: the Python libraries, the TypeScript packages, the Go and Kotlin clients that developers import when they want to build with GPT-4 or Claude or Llama. When a developer installs the official Anthropic Python package or the OpenAI JavaScript library, they're importing code that Stainless generated.

OpenAI uses Stainless. Google uses it. Meta uses it. Anthropic, until very recently, used it as a paying customer.

Now Anthropic wants to own it.

The Information reported that Anthropic is in advanced talks to acquire Stainless for at least $300 million. If the deal closes, it would be Anthropic's first acquisition ever. A company that has spent three years building safety-focused AI models would be buying the infrastructure layer that powers developer access to the entire AI industry, including developers building on its competitors.

What Stainless Actually Does for Developers

Most developers never think about who built the SDK they're using. They install the package and get to work. That invisibility is exactly what Stainless is designed for.

Stainless takes an API specification, specifically an OpenAPI spec, and automatically generates client libraries in multiple programming languages: Python, TypeScript, Go, Kotlin, Java. The generated code handles authentication, request formatting, error handling, rate limit management, and response parsing. It exposes the right types so development environments can autocomplete correctly and catch errors before they become runtime failures.

Alex Rattray, Stainless's founder, spent years building exactly this kind of system at Stripe. Stripe's SDKs are widely cited as among the best developer experiences in the industry: consistent across languages, always up-to-date, easy to understand. Rattray built the code generation system behind them.

He left Stripe to found Stainless because he saw the same problem appearing everywhere. Every API company was either building custom SDKs by hand, which is slow, expensive, and inevitably inconsistent, or generating mediocre boilerplate that created more friction than it solved. Neither approach scaled in an era where APIs change constantly and teams can't dedicate engineers exclusively to SDK maintenance.

The quality of an SDK matters more than most developers consciously register. At Stripe, Rattray saw that developer experience wasn't a polish problem; it was a product problem. When a library handles pagination automatically, retries failed requests with exponential backoff, and surfaces error messages that tell developers exactly what went wrong, developers move faster and file support tickets less often. Stripe's low support volume, relative to its API complexity, was partly a function of SDK quality.

Stainless applies that philosophy to the full AI industry. When an API adds a new endpoint, Stainless regenerates the SDK and publishes the update. When a parameter name changes, the old name appears as deprecated in the generated code rather than simply breaking. That automatic lifecycle management is what allows AI companies to ship new API features quickly without creating a maintenance crisis for developer experience.

Stainless launched out of stealth in 2023 and quickly found product-market fit with the companies building AI's most-used APIs. OpenAI was an early customer. Anthropic signed on. Groq, Cerebras, Runway, LangChain, and Meta's Llama Stack all followed. In December 2024, Stainless raised a \$25 million Series A from Andreessen Horowitz, led by Jennifer Li, with Sequoia and The General Partnership participating. The company was valued at roughly $100 million at that round.

At $300 million, Anthropic would be paying three times that valuation about 18 months later.

Why Anthropic Is Spending $300 Million on SDK Infrastructure

The number sounds high for developer tooling. But consider what Anthropic is actually buying.

First, control of its own developer experience. Anthropic currently uses Stainless to generate the official Claude Python and TypeScript SDKs, the primary way developers interact with Claude's API. That dependency on an independent third party is manageable when the third party is a neutral vendor. It becomes a different situation if a competitor acquired Stainless.

That competitive pressure almost certainly factors into Anthropic's thinking. OpenAI, Google, and Meta all depend on Stainless for their official SDKs. If OpenAI or Google were to acquire Stainless, Anthropic would face the awkward choice of continuing to rely on competitor-owned infrastructure or rebuilding everything in-house. Acquiring Stainless first eliminates that option for rivals.

Second, Anthropic is buying a capability that becomes more important as AI shifts from direct API calls to agentic workflows.

Stainless has been building what it calls SDK Code Mode MCP servers: Model Context Protocol servers that expose TypeScript SDK execution in a sandbox rather than individual API endpoints as separate tools. In Stainless's own benchmarking, this approach achieves 98% task completeness with 95% efficiency when an AI agent needs to interact with an API.

That matters enormously. When a Claude agent needs to call an external service, the quality of the SDK determines whether the call succeeds. Generic HTTP calls with JSON payloads are prone to type errors, missing parameters, and ambiguous error handling. A typed SDK with specific error messages and correct type definitions dramatically improves an agent's ability to call APIs correctly on the first try. Stainless generates that typed, agent-ready SDK infrastructure automatically from an API spec.

The company also generates AGENTS.md files: project-level context documents that tell AI coding tools how an API works. As tools like Claude Code and other AI coding agents become the primary interface through which developers write API integrations, the quality of AGENTS.md documentation becomes a competitive factor in developer adoption.

Third, Anthropic is buying a team with deep expertise in a narrow but increasingly valuable domain. Rattray and his colleagues built their careers around making APIs easy for developers to consume. That institutional knowledge doesn't come from large acqui-hires; it comes from years of focused work on a specific class of developer experience problem.

The timing of the acquisition makes more sense when you understand where Stainless's product roadmap was heading before this deal.

Over the past year, Stainless has been building what it describes as agent experience infrastructure. The shift reflects a fundamental change in how APIs get used. A few years ago, most API calls were made by human developers writing code that ran at a known time for a known purpose. Today, a growing share of API calls come from AI agents operating autonomously, making decisions about what to call and when without a human in the loop.

That changes what a good SDK needs to do.

For human developers, a good SDK provides idiomatic, typed interfaces that make common tasks easy and catch mistakes early. Error messages need to be readable. Documentation needs to explain not just how but when to use each endpoint. Types need to be rich enough to support autocomplete in editors.

For AI agents, the requirements overlap but add new constraints. Agents make decisions about what to call based on available information, which means the SDK's structure becomes a signal to the model. Clear types, specific error messages, and accurate documentation aren't just nice to have; they directly affect whether the agent calls the right endpoint with the right parameters on the first attempt. A poorly formed SDK can lead to failed API calls, type mismatches, and broken automation chains.

Stainless's MCP server approach addresses this directly. Instead of exposing individual API endpoints as separate tools in an agent's context, which bloats the tool list and creates unnecessary decision complexity, Stainless generates an MCP server that exposes a code execution sandbox. The agent writes TypeScript to call the SDK, just as a human developer would, rather than navigating a long list of individually exposed endpoints.

The benchmark results are significant. 98% task completeness means agents nearly always complete what they're trying to do. 95% efficiency means they're doing it without unnecessary redundancy. As AI agents become the primary way that APIs get called in production systems, the infrastructure for generating high-quality, agent-optimized SDKs becomes critical infrastructure for every company running an API. That's what Anthropic is acquiring.

The Fallout for Rivals, Developers, and Anthropic's Strategy

The awkward part of this deal is the customer list.

OpenAI's official Python library, installed hundreds of millions of times by developers worldwide, was built with Stainless. Google uses it for several of its API client libraries. Meta's Llama Stack SDK was generated by Stainless. If Anthropic completes this acquisition, three of its primary AI competitors will have their official developer SDKs generated by Anthropic-owned infrastructure.

There's no obvious precedent for this in the technology industry. A company owning the SDK generation layer for its direct rivals would be unusual in any sector. In AI, where developer mindshare is a critical competitive resource and where the difference between winning and losing a developer relationship often comes down to tooling quality, it's an aggressive positioning move.

In the short term, Anthropic has every incentive to keep serving OpenAI and Google as customers. The revenue is real, and the reputational risk of disrupting competitors' developer tooling would undercut Anthropic's positioning as a trustworthy AI company. The more likely near-term outcome is Stainless continuing to operate as a neutral vendor under Anthropic ownership, at least for some period.

But the strategic dynamics shift the moment the deal closes. OpenAI's engineering teams will need to decide whether to continue depending on Anthropic-owned infrastructure for their SDK generation, or invest in building alternatives in-house. The same calculation applies at Google. The $300 million price tag is partly an option on forcing that decision sooner rather than later, before competitors can build or acquire equivalent tooling.

There are also legitimate questions about future feature development. Today, Stainless's product roadmap is driven by what all of its customers need. Post-acquisition, the incentives are different. If Stainless develops a capability that gives Claude's SDK a specific advantage for agentic use cases, how quickly does that capability get extended to OpenAI's SDK? These questions don't have obvious answers before the deal closes, but they're the ones engineering leaders at competing AI companies will be asking this week.

This dimension makes the Stainless acquisition more aggressive than most Anthropic moves to date. Anthropic has generally competed on model quality and safety research. Acquiring neutral developer infrastructure and positioning it within one competitor's orbit is a different kind of strategic choice, one that's harder to walk back once made.

Stainless would be Anthropic's first acquisition, which makes the choice revealing about where the company's competitive thinking has evolved.

Anthropic has operated for three years without making any acquisitions. OpenAI, by comparison, made at least five acquisitions worth over $10 billion between 2024 and 2026. Anthropic has been building core model capabilities in-house, investing in safety research, and partnering with cloud providers rather than buying its way into adjacent markets. The company has been deliberate about staying focused.

The Stainless deal, if it closes, suggests Anthropic has decided that controlling developer infrastructure is important enough to pursue aggressively, even if it means acquiring a company that currently serves competitors. That's a meaningful shift in posture for a company that has largely competed on the quality of Claude and the clarity of its safety mission.

It also signals something about Anthropic's theory of developer growth. Developer adoption is a competitive battleground where Anthropic faces structural disadvantages. OpenAI's ecosystem advantage, built partly through ChatGPT's consumer reach and partly through deep integration with Microsoft's developer toolchain, is real. Google has Gemini embedded across Android, Google Workspace, and Firebase. Anthropic's recent expansion into enterprise infrastructure, including the Claude Platform integration with AWS, represents one vector of the same effort to build developer reach that doesn't depend solely on model quality.

Owning the SDK generation layer gives Anthropic a different kind of presence: infrastructure-level influence over the developer experience for building with any major AI API, not just Claude. Whether that plays out as neutral vendor value or active competitive advantage depends on how Anthropic manages the business post-acquisition, and those are choices the company hasn't had to make before.

The broader context here involves consolidation. The AI industry's infrastructure layer is increasingly getting absorbed into major lab and cloud ecosystems. Neutral tool providers that served the whole market, for SDKs, evaluation, observability, and orchestration, are becoming acquisition targets as the big players build out platform strategies. The Stainless deal, if it closes, is an early and unusually consequential example of that pattern.

The Information's reporting indicates advanced talks, not a signed agreement. Either side could walk away. Terms could change. The deal could collapse over regulatory concerns, price disagreements, or strategic recalculation.

But advanced talks for a $300 million deal involve legal review, due diligence, and alignment on structure. The fact that The Information, which has a reliable track record on acquisition reporting, is covering it suggests the probability of completion is meaningful.

For developers building with Claude, GPT, or any other major AI API today, the immediate answer is: nothing changes yet. Stainless's SDKs continue to be the official, maintained client libraries for OpenAI, Google, and the rest. The technical quality of those SDKs won't drop overnight.

The medium-term picture depends on how Anthropic structures the acquisition. If Anthropic treats Stainless as a neutral infrastructure vendor and continues investing in its product for all customers, developers across all the major AI platforms benefit from a better-funded team building agent-ready SDK infrastructure. If Anthropic prioritizes Claude's SDK development at the expense of competitors', developers on other platforms will eventually notice the divergence.

For teams building multi-platform AI applications, the acquisition is a signal to start thinking about SDK dependencies more carefully. The ownership chain of your core developer tooling now matters in a way it didn't a week ago. Teams that depend on Stainless-generated SDKs for OpenAI or Google integrations should understand the implications if those SDKs eventually fall behind their Claude counterparts.

There's also a version of this story where the acquisition accelerates the entire ecosystem. Stainless under Anthropic ownership with a larger engineering budget and deeper model knowledge could build better agent SDKs for everyone, not just Claude users. The MCP server benchmarks Stainless already published, 98% task completeness with open-source methods, suggest the technical work is ahead of where most SDK tooling sits today. More investment could accelerate that work across the board. That outcome requires Anthropic to run Stainless as a genuine platform business rather than a captive Claude advantage, but it's a plausible path if the economics support it.

The Stainless acquisition, at $300 million and first-ever status for Anthropic, would mark a significant shift: a company known for safety research and model quality deciding that owning the infrastructure developers use to build on AI is worth the cost, the competitive complexity, and the strategic departure from three years of building everything in-house. The deal hasn't closed. But if it does, the developer infrastructure landscape for AI looks different than it did yesterday.