AI Agent Lock-In Is Here
Claude Code Hit $2.5B. Amazon Engineers Can’t Use It. This Is the New Vendor Lock-In.
Claude Code just hit a $2.5 billion run-rate — doubled since January 1st. Yet 1,500 Amazon engineers are fighting for permission to use it, steered toward AWS Kiro instead. This is vendor lock-in repackaged for the AI agent era. Platform-native vs platform-agnostic is the new architectural fault line. Here's what every cloud team needs to decide.
The $2.5 Billion Signal
Claude Code just hit a $2.5 billion annualized run-rate. That number doubled since January 1st. Business subscriptions quadrupled in the same period. Enterprise customers now account for more than half of all Claude Code revenue.
$2.5B
Run-Rate ARR
4%
GitHub Commits
2x
Weekly Users Since Jan
9 mo
Zero to $2.5B
According to a SemiAnalysis report published February 5, 2026, Claude Code now authors 4% of all public commits on GitHub worldwide — double the share from one month earlier. At the current trajectory, they project Claude Code will author 20%+ of all daily commits by the end of 2026.
Claude Code went from zero revenue to $2.5 billion in approximately nine months. It’s part of Anthropic’s broader $14 billion ARR — up from $1 billion just 14 months ago. Eight of the Fortune 10 are now Anthropic customers. Over 500 customers spend more than $1 million annually, up from roughly a dozen two years ago.
The Amazon Paradox
Amazon has invested $8 billion in Anthropic. They sell Claude Code to customers through AWS Bedrock. Yet roughly 1,500 Amazon engineers are fighting internally for permission to use Claude Code in production work.
Instead, Amazon is steering teams toward Kiro — their in-house AI coding agent built on top of Anthropic’s Claude models combined with Amazon’s proprietary tooling and infrastructure. Internal guidance limits the use of third-party AI coding tools for production code without formal approval.
Sales engineers who sell Claude Code to customers through Bedrock have expressed concerns that it’s difficult to convince customers to use a product they haven’t officially approved for internal production use.
The irony is thick. But it’s not really about Amazon. This is a preview of tensions every major cloud vendor will face. Your dev team wants the best tool. Your procurement team wants the integrated one. When those two forces diverge, you get a new flavor of lock-in.
Meanwhile at Microsoft
Microsoft has taken the opposite approach — actively encouraging engineers to test Claude Code alongside GitHub Copilot and share performance feedback. The data flows back into improving Copilot itself.
Two tech giants. Same investor in Anthropic. Completely opposite internal strategies. The contrast tells you everything about where the AI agent wars are heading.
Two Camps Are Forming
Platform-Native Agents
Kiro, GitHub Copilot
Deep integration with specific cloud ecosystems
Access to proprietary APIs and internal services
Optimized for that vendor's workflow and toolchain
But tied to a single platform
The trade-off: Maximum ecosystem integration at the cost of portability. When you build your dev workflow around Kiro, you’re deepening your commitment to AWS — not just for infrastructure, but for how your team writes code.
Platform-Agnostic Agents
Claude Code, OpenAI Codex
Work across any cloud, any stack, any language
Stronger general reasoning capabilities
Portable between projects and employers
But less deeply integrated with specific platforms
The trade-off: Flexibility and best-in-class reasoning at the cost of native ecosystem depth. Your workflows are portable, but you may need to bridge gaps that native tools handle natively.
Why This Matters Now
AI coding agents aren’t just writing code anymore. They’re becoming infrastructure.
CI/CD Pipelines
Agents triaging failures, generating fixes, and auto-merging patches
Issue Triage
Agents classifying, prioritizing, and drafting responses to production incidents
IaC Generation
Agents writing Terraform, Pulumi, and CloudFormation from architectural intent
Code Review
Agents reviewing PRs, suggesting improvements, and enforcing standards at scale
Once you build your development workflow around a specific agent, switching costs are real. Custom prompts, AGENTS.md files, tool integrations, CI/CD hooks, team muscle memory — all of it creates inertia. This is cloud lock-in, but at the developer tooling layer.
Three Questions Every Cloud Team Should Ask
Does your AI agent choice lock you into a cloud vendor?
If your AI coding assistant only works well with AWS services, you’ve added another lock-in vector you didn’t have before. Your infrastructure was already locked in. Now your development process is too. That’s a compounding effect that makes migration exponentially harder.
What happens when your vendor’s agent competes with the best agent?
AWS pushing Kiro over Claude Code is a preview of tensions every major cloud will face. Your dev team wants the best tool. Your procurement team wants the integrated one. When the vendor-preferred tool and the developer-preferred tool diverge, someone has to make a deliberate choice — not let it happen by default.
Are you building agent-portable workflows?
The teams that abstract their AI tooling will have the same advantage as teams that went multi-cloud early — flexibility when the landscape shifts. Use open standards like MCP and AGENTS.md (both now under the Linux Foundation’s Agentic AI Foundation). Keep your agent orchestration layer swappable. Treat AI agents like you treat cloud services: with an exit strategy.
The Numbers Behind the Shift
Anthropic (Feb 2026)
$14B annualized revenue (up from $1B in Dec 2024)
$380B valuation after $30B Series G
500+ customers spending $1M+ annually
8 of 10 Fortune 10 companies are customers
30,000 Accenture professionals being trained on Claude
Claude Code Specifically
$2.5B annualized run-rate (18% of total revenue)
4% of all public GitHub commits worldwide
135K+ commits per day on GitHub
4x business subscription growth since Jan 1
20%+ projected daily commit share by end of 2026
Meanwhile, Amazon claims 70% of its software engineers have used Kiro at least once as of January 2026. But “used at least once” and “prefer to use daily” are very different metrics. The 1,500-engineer pushback suggests the gap between mandated adoption and organic preference is wide.
The Platform Engineering Playbook
The $2.5B run-rate tells you where developer preference is heading. The Kiro push tells you where vendor incentives are heading. The gap between those two forces is where platform engineering teams need to make a deliberate choice.
Audit your agent dependencies
Map every place an AI agent touches your workflow: IDE, CI/CD, code review, IaC, incident response. Identify which are vendor-locked and which are portable.
Adopt open agent standards
MCP (Model Context Protocol) and AGENTS.md are now under the Linux Foundation. Build your agent integrations on open standards, not proprietary SDKs.
Keep the orchestration layer swappable
Your AGENTS.md files, prompt libraries, and workflow definitions should work with any capable agent. Don't hardcode agent-specific behaviors into your pipelines.
Measure what your team actually uses
Track which agents your engineers choose when given freedom vs which they're told to use. The gap between those two metrics is your lock-in risk.
Negotiate agent flexibility into contracts
When signing cloud or enterprise AI agreements, ensure your team retains the right to use best-of-breed agent tooling — not just the vendor's preferred agent.
The Bottom Line
Don’t sleepwalk into AI agent lock-in the way many teams sleepwalked into cloud lock-in a decade ago.
The vendor lock-in debate has a new front: the tools your engineers use to write, review, and ship code. Make it a deliberate architectural decision, not an accidental procurement outcome.
Sources: Anthropic Series G • SemiAnalysis Report • TechRepublic • SiliconANGLE
Related Posts
GitHub Agentic Workflows: The Decision Framework Nobody's Talking About
Everyone's excited about AI in CI/CD. Nobody's asking when to use it vs when not to. GitHub Agentic Workflows just entered technical preview — the architecture is solid. But the real decision isn't which agent to pick. It's when to use agentic workflows vs deterministic ones. Here's the decision framework, the adoption pattern, and the three questions to answer before you deploy.
GitHub Agent HQ: Claude & Codex Join Copilot in a Unified AI Coding Dashboard
GitHub just launched Agent HQ — a unified dashboard inside GitHub, GitHub Mobile, and VS Code that lets Copilot Pro+ and Enterprise users run Claude, OpenAI Codex, and Copilot agents without leaving their repo or PR. With 20M+ Copilot users and 90% Fortune 100 adoption, the "best AI coding tool" debate just became "best AI coding workflow."
The 2026 Blueprint: Why DIY Infrastructure Is Becoming a Liability
My prediction for 2026: enterprises still running DIY infrastructure will fall behind. With 89% multi-cloud adoption, explosive edge growth, and a $5.5 trillion skills gap looming—the complexity has crossed a threshold.