Updated 20 hours ago
Anthropic Wants You to Build on Claude — Then It Competes With You

Anthropic Platform Risk

Anthropic Wants You to Build on Claude — Then It Competes With You

Anthropic is aggressively expanding into enterprise software with Claude Sheets, Whiteboards, and industry‑specific tools, raising alarms among developers who say the company is following Amazon's playbook: watch what sells, then ship a competing version.

The Amazon Playbook, AI Edition

Anthropic's November 10 product blitz — Claude Sheets, Whiteboards, Claude Data, and a preview of Claude Meetings — signals an aggressive push into enterprise software. But for developers building businesses on Claude's API, the expansion looks less like a platform and more like a trap. The New Stack reports that founders are drawing direct parallels to Amazon's private‑label strategy: watch which third‑party products sell best, then launch your own competing version.

Gerald Zingraf, president of GlacierGrid, told The New Stack that Anthropic's move "makes me nervous for wrappers and pure SaaS companies," comparing it to how Amazon uses marketplace data to inform its own product decisions. Todd Saunders, CEO of Dalton Mills, went further, calling Anthropic "the new Amazon" and warning that the company could observe application usage through its SDK and eventually ship competing functionality directly into Claude.

Spellbook vs. Claude for Legal: When Assurances Don't Hold

The most concrete example involves Spellbook, a legal AI startup that uses Claude models for contract review, revision suggestions, and workflow automation — processing 530,000 contract reviews monthly. The New Stack reports that Anthropic employees had assured Spellbook they had no plans to compete in legal software. Then Anthropic launched Claude for Legal, which includes contract‑review capabilities that overlap with Spellbook's core business.

Spellbook remains a Claude customer and is still featured in Anthropic's marketing, but CEO Scott Stevenson told The New Stack the experience was eye‑opening: "Anthropic has been a great partner, but it made us realize that they are a massive company and two parts of the org can have conflicting mandates. Ultimately this is inevitable," Stevenson told.1 There is no evidence Anthropic used Spellbook's proprietary data to build its legal offering, but the timeline of assurances followed by a competing launch has left developers wary.

A History of Cutting Off Competitors

Anthropic's move into enterprise apps follows a pattern of aggressive platform enforcement. 2 documents multiple instances where Anthropic cut off access to competitors: in April 2025, it blocked Windsurf (a coding app startup) when it was about to be acquired by OpenAI, just as Anthropic was ramping up Claude Code. In August 2025, it cut off OpenAI's access to Claude models, citing use for testing GPT‑5. In January 2026, it blocked xAI's access through Cursor to prevent acceleration of xAI's model development.

Anthropic's commercial terms explicitly forbid using its service "to build a competing product or service," including reselling the services without express approval, according to terms cited by.2 OpenAI and Google have similar clauses. The Brookings analysis notes that all three major AI model providers — Anthropic (40% market share), OpenAI (27%), and Google (21%) — have policies allowing them to cut off competing AI app developers. Combined, they control roughly 90% of the $37 billion enterprise LLM market.

What Anthropic's Policies Say and Don't Say

Anthropic's privacy policy, effective September 10, 2026, acknowledges it has access to information about how customers interact with third‑party apps and services, The New Stack notes. The commercial terms state that business customers retain rights to inputs and own outputs, and Anthropic agrees not to use customer content for model training. Feedback data excludes content pulled through connectors and MCP servers unless users paste it directly into a conversation.

What remains unclear is whether Anthropic uses aggregated integration data — such as which connectors or third‑party apps are most popular — to inform its own product roadmap. Anthropic did not respond to The New Stack's request for comment.

Why AI Companies Can't Stay Platform‑Only

The push into applications isn't just ambition — it's economics. Brookings notes that AI model provision risks becoming a commodity margin business as models rapidly converge in capability. With Microsoft, Meta, Alphabet, and Amazon projected to spend $610 billion in capex in 2026, AI companies "need application‑layer revenue to justify the capital expenditure." As Paul Kedrosky told Brookings: "applications become the margin layer," as Kedrosky told.2

Anthropic's Claude Code, launched February 2025, already competes with Microsoft's GitHub Copilot and Cursor — a top Anthropic customer. The three now have roughly equal shares of the AI coding tools market. The company's February 2026 release of specialized tools for legal and financial work similarly placed it in direct competition with RELX and Thomson Reuters, whose stock fell significantly after the announcement.

What This Means for AI Builders

For developers building on Claude's platform, the risk is no longer hypothetical. Any AI app developer whose product overlaps with Anthropic's expanding enterprise portfolio faces the possibility of being cut off — as Windsurf, OpenAI, and xAI discovered. The switching cost is theoretically low today since all three major model providers offer comparable frontier models, but the oligopoly structure means all three are moving into applications simultaneously.

The tradeoff, as 1 frames it: developers gain access to Anthropic's growing enterprise customer base by building on Claude. In return, they accept that Anthropic controls the interface and can introduce competing features without consulting them. Whether that relationship stays sustainable "may depend less on what Anthropic promises today than on what it decides to ship next," 1 concluded. For builders whose businesses depend on a single model provider, the lesson is clear: platform risk in AI looks exactly like platform risk everywhere else.

Sources

  1. 1.The New Stack(thenewstack.io)
  2. 2.Brookings(brookings.edu)

Share this article

PostShare

More on This Story

Related News