Anthropic Planted Three Flags in One Week.
The Enterprise Developer Layer May Never Be the Same.

Three moves. One week. One layer of the AI stack that nobody was defending.

Between May 12 and May 18, 2026, Anthropic acquired Stainless for at least $300 million, brought Andrej Karpathy into the fold, and locked in a distribution alliance with KPMG and its 270,000 person consulting army. Any one of these would be a headline. All three in the same week is a thesis statement.

The thesis: the enterprise developer layer, the SDKs, the connectors, the reference architectures that determine which model becomes the default inside Fortune 500 companies, is the most valuable unclaimed territory in AI. Anthropic just planted three flags on it before OpenAI, Google, or Meta even packed their bags.

Here is how that works, why it might not, and what it means for anyone building on top of these platforms.

The Plumbing Thesis

Most people think the AI race is about models. Bigger parameters. Better benchmarks. Flashier demos. The truth is simpler and far less glamorous.

The company that wins enterprise AI will be the one that controls the plumbing: the SDKs developers install, the connectors agents use to reach internal systems, and the consulting playbooks that CIOs trust enough to sign seven-figure contracts around.

Call it The Plumbing Thesis. Models are commoditizing. Plumbing is not.

Here is the mental model. Picture the AI stack as four layers. At the bottom: foundation models and raw compute. Above that: APIs, orchestration, and routing. Then the enterprise developer layer, where internal teams actually wire models into databases, SaaS tools, and compliance workflows. At the top: finished applications.

Layers one and four get all the attention. Layer three decides who actually gets paid. Anthropic's trifecta targets layer three with surgical precision. Stainless owns the SDK surface area. Karpathy owns developer mindshare. KPMG owns the procurement relationships. Together, they form a full-stack play for the connective tissue of enterprise AI.

Simple scales, complex fails. And nothing is simpler than being the default import statement in every enterprise developer's codebase.

The Three Flags: What Anthropic Actually Bought

Let me walk through each move and what it really controls.

Stainless: the front door to every API call. Stainless, founded around 2022, built automatic SDK generation for API-driven services. Before the acquisition, it generated production-grade client libraries in 7 to 10 languages for OpenAI, Anthropic, Google, and Cloudflare.

Here is the number that matters: roughly 60 to 80 percent of production API calls from medium and large customers go through official or semi-official SDKs rather than raw HTTP. Stainless built many of those SDKs. Anthropic now owns the factory.

Katelyn Lesse, Anthropic's Head of Platform Engineering, stated it plainly: "Agents are only as useful as what they can connect to." Stainless also has first-class support for Anthropic's Model Context Protocol, launched in November 2024. SDK generation and agent connectivity tooling now sit under one roof.

The competitive implication is sharp. OpenAI and Google relied on Stainless. TechCrunch reported on May 18 that Anthropic will wind down all hosted Stainless products. Rivals will need to fork, replicate, or build from scratch. Not fatal, but expensive and slow.

Karpathy: the most expensive credibility signal in AI. Andrej Karpathy's resume (ex-Director of AI at Tesla, founding member of OpenAI, creator of some of the most-watched AI engineering lectures on the internet) makes him the single most recognizable name in applied AI research. His hire is not about one person writing code. It is about what his presence signals to the 1 to 2 million developers actively working with LLM APIs in 2026.

Developer trust compounds like interest. When an engineer choosing between Claude and GPT-4 sees Karpathy's name on Anthropic's roster, it shifts the default. Not for everyone. But at the margin, where enterprise decisions actually get made, margins are everything.

I think this is the most underrated of the three moves. SDKs can be rebuilt. Consulting deals can be renegotiated. But developer credibility at this level takes years to earn and seconds to lose.

KPMG: 270,000 salespeople you do not have to hire. KPMG pulled in roughly $36 to $37 billion in global revenue for FY 2023. Its advisory arm touches thousands of Global 2000 companies across financial services, healthcare, government, and energy.

But here is the honest hedge: it is unclear whether this deal delivers meaningful exclusivity. KPMG, like every Big Four firm, partners with multiple vendors simultaneously. PwC committed $1 billion to Microsoft and OpenAI in 2023. Accenture pledged $3 billion across AI investments. KPMG will almost certainly keep OpenAI and Google in the mix.

The value is not exclusivity. It is presence. When a KPMG partner walks into a Fortune 500 boardroom with a "reference architecture for compliant AI," Claude will be on the slide. Being on the slide is how defaults get set in enterprises where 70 to 80 percent are experimenting with generative AI but only 15 to 25 percent have production deployments. The gap between experiment and production is where consulting relationships convert.

2031

Three signals inside the same shift

Pull back five years. Where does the Plumbing Thesis land?

Two scenarios worth holding in tension. In the first, Anthropic's bet works. The enterprise developer layer consolidates around a small number of vertically integrated platforms. Whoever controls the SDKs, the agent connectivity protocols, and the consulting playbooks becomes the AWS of AI middleware. In this world, Anthropic's May 2026 week looks like Amazon's 2006 launch of S3 and EC2: the moment infrastructure became platform.

The asymmetric advantage is real. Models will keep improving across all providers. The gap between Claude, GPT, and Gemini on raw capability will narrow. But the switching cost of ripping out SDKs, rewriting agent connectors, and retraining 50,000 KPMG consultants on a different stack compounds every quarter. By 2031, the enterprise developer layer could be worth $10 to $50 billion annually in integration budgets alone, according to current trajectory estimates on agent connectivity spend.

In the second scenario, the plumbing stays fragmented. Enterprise developers, especially in regulated industries, resist single-vendor lock-in at the connective tissue layer. LangChain, LlamaIndex, Semantic Kernel, and open-source MCP implementations keep the middleware layer vendor-neutral. KPMG uses Anthropic for some clients and OpenAI for others. Stainless's competitors emerge within months. Karpathy, who has left major organizations before, moves on after 18 months.

My read: the truth will land somewhere between these poles. Anthropic will not "win" the enterprise developer layer the way Google won search. The layer is too fragmented and buyers are too sophisticated. But Anthropic does not need to win it outright. It needs to become the default starting point. The first SDK installed, the first model in the KPMG slide deck, the first name a senior engineer trusts.

Defaults are not monopolies. They are something more durable: habits.

There is also a deeper strategic tension worth naming. Anthropic's brand is built on safety, on constitutional AI, staged rollouts, and alignment research. Winning the developer layer often requires the opposite: fast iteration, aggressive pricing, lower friction. If Anthropic maintains strict safety gating on every new capability, it may move slower than OpenAI backed by Microsoft's enterprise sales machine or Google bundling Gemini into Workspace and GCP. If it relaxes guardrails to ship faster, it risks the very credibility that makes Karpathy's hire and KPMG's partnership valuable.

This is the central paradox of Anthropic's 2026 strategy. The company that promises to move carefully is trying to move fastest where it counts. Whether those two impulses can coexist for five years is the real question.

What to Build This Weekend

You do not need to wait for Anthropic's strategy to play out to act on what it reveals. The Plumbing Thesis applies at every scale.

Step one: audit your SDK dependencies. If you are building on Anthropic's Python or TypeScript SDK, check whether your client library was Stainless-generated. Most official Anthropic SDKs were. This is now first-party infrastructure, which means faster updates but also tighter coupling. If you are building on OpenAI's SDK, watch for migration notices. The Stainless tooling that generated it is no longer neutral.

Step two: test Model Context Protocol. MCP launched in November 2024 and is Anthropic's standard for how agents connect to external tools and data sources. Set up one MCP server this weekend. Pick a simple use case: connect Claude to a local database or an internal API. The documentation is public. You do not need a CS degree. You need about two hours and a willingness to break things.

Step three: build for portability. The biggest lesson from this week is that neutral infrastructure can become proprietary overnight. If your AI workflow depends on a single provider's SDK, add an abstraction layer now. It does not need to be fancy. A thin wrapper that lets you swap model providers without rewriting your application logic will save you weeks when the next acquisition reshuffles the stack.

Step four: follow the consulting playbooks. KPMG and the other Big Four firms will publish reference architectures for Claude integration in regulated industries over the next 6 to 12 months. These documents are free market research. They tell you exactly what compliance-sensitive buyers care about: logging, audit trails, approval workflows, data residency. Build those features into your product now, before they become table stakes.

The enterprise developer layer is where the real money moves. Anthropic just told you exactly where they think it is. The question is whether you are building on top of it or waiting for someone else to build it for you.