# Claude Code Just Topped the Agent Harness Rankings - and the > Source: > Published: 2026-08-24 13:04:38+00:00 # Claude Code Just Topped the Agent Harness Rankings - and the August agent harness rankings put Claude Code first on depth of hooks, subagents and dynamic workflows. Codex CLI, Cursor, Gemini CLI and Copilot follow,… Ask a developer which coding tool is best and you'll get a model comparison. Ask the August agent harness rankings and you get a different answer entirely, one that says the model stopped being the point a while ago. [Claude](/glossary/claude) Code ranks first on depth of hooks, subagents and dynamic workflows, and is named the default choice for long autonomous coding sessions. Codex CLI leads on cloud, pull-request-shaped autonomy. Cursor holds in-editor workflows, with [Gemini](/glossary/gemini) CLI and GitHub [Copilot](/compare/github-copilot-vs-cursor) rounding out the top five. ## Same Models, Different Harness The revealing detail is that several of these tools run the same frontier models. Claude Code and Codex CLI aren't competing over whose underlying brain is smarter. What differs is session management and how gracefully they fail. That should reframe the entire question. For the last two years the industry has been arguing about which model wins. The August rankings quietly move the battleground to the harness, the layer that manages context, coordinates subagents and decides what happens when something breaks halfway through a long task. ## Throughput Is Not Speed And here's the punchline, because it's the same lesson twice in one month. A better harness raises throughput without raising speed. That's exactly what Linear's telemetry showed last week, when coding agents tripled pull requests without cutting cycle time because the bottleneck sat in review, not generation. You can hand a developer the fastest model on the market and watch them ship slower than a colleague on an older model with a harness that manages failure well. The constraint has moved. It's not the brain. It's the scaffolding around the brain, and the teams that figure that out stop buying bigger models and start fixing their session management instead. ## What the Rankings Don't Measure The rankings are useful precisely because they admit their own limit. They score depth of hooks, subagents and workflow dynamics. They don't measure the things that actually eat a quarter: onboarding friction, cost per session, or what happens to your repo when an agent goes wrong and nobody notices. So the honest read isn't "Claude Code wins." It's that we've finally started measuring the right layer. The model wars were a distraction. The harness wars are the real fight, because that's where throughput is actually won or lost. *Sources: August 2026 agent harness rankings; Linear telemetry on coding-agent throughput; AI Tools Recap daily briefing, August 24, 2026.* Get AI news in your inbox Daily digest of what matters in AI.