You picked your project management tool after weeks of evaluation. You read comparison posts. You watched demos. You argued with your team about Kanban versus timeline views. Nobody in that process chose which AI model would run inside the tool. Your vendor did that while you were comparing board layouts.
That should bother you — not because Claude is bad, but because the last time a company pulled this exact move, everyone woke up a decade later wondering why their payment processing fees had quietly tripled.
Seven lines of code, then the trapdoor
Stripe launched in 2011 with seven lines of code and a promise: never think about payments again. By 2015, ripping Stripe out of a production application meant rewriting billing, subscriptions, fraud detection, disputes, and compliance simultaneously. The switching cost was never the API. It was every feature the platform built on top of the API.
Anthropic is running this playbook with surgical precision. Managed Agents — their hosted agentic infrastructure that launched on April 8, 2026 — is the "seven lines of code" moment. Embed Claude as your product's brain. Anthropic handles compute, memory, tool orchestration, scaling. The platform builds features. Users never see the Anthropic logo.
Here's how Stripe's lock-in actually worked, and why each step has an Anthropic equivalent.
Step one: make integration trivially easy. Stripe: paste a snippet, accept money. Anthropic: call an API, deploy an agent. When standing up AI capabilities takes one week — as Rakuten demonstrated across five departments in an April 8 case study published alongside the Managed Agents launch — ripping them out takes a fiscal quarter, three steering committee meetings, and a CTO willing to break production features.
Step two: encourage building on top. Stripe didn't just process charges. Platforms built subscription logic, metering, and revenue recognition using Stripe's primitives. Now SaaS platforms are building document generation, task delegation, and autonomous debugging on Claude's reasoning capabilities. Those aren't integrations. They're product differentiators that happen to run on rented intelligence.
Step three: make the platform's customers depend on the features. This is the trapdoor. Once end users rely on their workspace AI to draft reports and manage tasks, the platform can't remove it without a user revolt. And they can't swap the underlying model without regression-testing every AI-powered feature against a completely different reasoning engine. The switching cost isn't engineering hours. It's user retention.
Step four: raise prices. Stripe's effective take rate on US transactions crept steadily upward over a decade — base fees, Radar, Billing, Connect, premium tiers — all stacking incrementally. By the time a platform noticed, the migration cost exceeded the cumulative price increase by an order of magnitude.
The meter you can't see
Anthropic doesn't even need to raise list prices. They just need platforms to ship features that consume more tokens per user action. A quick AI summary runs 500 tokens. An autonomous agent that researches, drafts, revises, and delivers? Fifty thousand. The meter spins 100x faster, and the user sees it labeled "more powerful AI features." The platform sees a line item that grows 40% quarter-over-quarter with no rate change. Everyone's happy — until someone runs the unit economics.
Where the analogy cracks
The Stripe parallel breaks in one critical place, and not in Anthropic's favor. Stripe's moat included regulatory barriers: PCI compliance, money transmitter licenses, banking relationships. Legal switching costs reinforced technical ones. AI models have zero regulatory moat. Nothing prevents a platform from swapping Claude for Gemini tomorrow except inertia.
But inertia is a hell of a moat. Platforms don't accumulate contractual lock-in — they accumulate thousands of carefully tuned system prompts, tool schemas, and output formatting expectations calibrated to Claude's specific reasoning patterns. Engineers call that technical debt. Anthropic's sales team calls it customer retention. Same thing, different slide deck.
Stripe also dominated a thin competitive field for years. Anthropic doesn't have that luxury. OpenAI's Responses API and Google's Vertex Agent Builder offer comparable agentic infrastructure right now. The window to lock in platforms before multi-model orchestration layers commoditize the whole stack is 12 to 18 months, tops. Anthropic knows this — the hundred-million-dollar partner fund (announced March 12) and the consulting-firm onboarding blitz aren't growth investments. They're land-grab tactics on a timer.
Check the receipt
Look at your SaaS invoices. You won't find a Claude line item. Look for the "AI" tier your tools introduced in the last six months — the $8 or $12 per-user add-on that appeared in your renewal quote. A cut of that flows to Anthropic, priced per token, negotiated by your vendor. You didn't get a vote. You probably didn't get a notification.
Stripe turned every internet business into a Stripe customer without most of them realizing they'd made a choice. Anthropic is doing the same thing with reasoning — and unlike Stripe, they don't even need their name on the receipt. You chose your Kanban board carefully. The AI model underneath it chose you.
