THE 60-SECOND VERSION
● Claude Code ranks first among agent harnesses for long autonomous coding sessions.
● Cloudflare shipped agents a browser, a wallet and a payment protocol. Agents can now browse and pay with no human in the loop.
● 7 days left on Claude Sonnet 5 at $2/M input, and on GPT-5.4 in Codex.
Claude Code tops the agent harness rankings
New August rankings place Claude Code first on depth of hooks, subagents and dynamic workflows, and name it the default choice for long autonomous coding sessions. Codex CLI leads on cloud pull-request-shaped autonomy, Cursor on in-editor workflows, with Gemini CLI and GitHub Copilot completing the top five.
The useful reading is that these are not competing for one job. Several run the same frontier models — what differs is session management and how gracefully they fail. And a better harness raises throughput without raising speed, which is exactly what Linear's telemetry showed last week.
Full breakdown: all five ranked, and what the rankings do not measure →
Cloudflare built agents a browser, a wallet and a way to pay
Three announcements across three weeks that only cohere when read together. Kitesurf, a stateless browser written in Rust running in V8 isolates on Workers with no Chromium underneath, using 3 to 7 times less CPU and memory on screenshots and HTML extraction. Wallets, letting humans fund an account and delegate capped spending to agents. And the x402 protocol, letting any site or API charge an agent per request in stablecoins, settled at the edge.
Over 20 companies are already in agent-initiated payment flows. Kitesurf is free in beta and speaks Chrome DevTools Protocol, so existing Playwright and Puppeteer clients work.
Full breakdown: the architecture, the honest caveats, and the governance question →
The week ahead
| Date |
What happens |
| Aug 31 |
Claude Sonnet 5 goes $2 to $3 per million input, $10 to $15 output, plus a tokenizer change adding 10 to 35 percent tokens on code |
| Aug 31 |
GPT-5.4 and GPT-5.4 mini leave Codex for ChatGPT sign-in users. Still available via API key |
| Aug 31 |
kimi-k2.5 and moonshot-v1 sunset, migrate to kimi-k3 |
| Early-mid Sept |
Grok 4.7 window per Musk. Still no model ID at docs.x.ai |
| Oct 1 |
OpenAI vs Apple hearing |
| Oct 24 |
deepseek-chat and deepseek-reasoner deprecated |
| ~Nov 21 |
GPT-5.6 Sol promotional pricing ends, $4/$20 back to $5/$30 |
Two of those land on the same Monday and pull opposite ways. If you run coding workloads across both Anthropic and OpenAI, next week is a repricing and a migration at once.
Still open
- CVE-2025-62593 in Ray. Federal deadline passed 20 August. Anything below 2.52.0 is still exposed, and the attack path is a developer's browser rather than a server. What to check →
- OpenAI's public S-1. Filed confidentially 8 June, still not on EDGAR.