METR on Anthropic: Staff Feel 12x, Progress Is 2x
METR estimates Anthropic now improves its models about twice as fast, mostly thanks to AI, while staff self-report 12x speedups. What the gap teaches teams measuring AI coding gains.
AI coding agents, compared like a developer would.
Daily, sourced write-ups on Claude Code, Codex, Copilot, Cursor, Devin, MCP, and the models behind them: what shipped, what broke, and what it costs you.
Latest
New posts on coding agents, their models, and the infrastructure around them.
Conductor Joins Meta: Move Your Agent Workspaces
Conductor Cloud shuts down November 20 after the Meta deal. How to save your workspaces and branches, and where to run parallel Claude Code and Codex agents next.
Revolut's Self-Hosted GLM: When the Math Works
Revolut says self-hosting GLM on 56 B200 GPUs cut its AI coding bill about 90%. What that figure leaves out, and the team size where self-hosting starts to pay.
Infostealers Now Target Claude Code and Codex Tokens
WardenStealer and Aoki copy login tokens from Claude Code, Codex, Copilot, and Cursor. Where each tool stores its login and how to revoke it on the server.
Claude Message Threads: Faster Turns for Agent Loops
The Messages API can now hold a conversation server-side for five minutes so each agent turn sends only new messages. What it speeds up, what it doesn't, and how to handle expiry.
DeepSeek Super-agent: What the Open Prototype Does
DeepSeek open-sourced dsh-superagent, a coding agent that takes requests from GitHub and Feishu. What it does by default, how it decides who can ask, and why its workers need an isolated machine.
Claude's Nov 30 Cutoffs: Opus 4.8 Fast and Sonnet 4.5
Fast mode on Opus 4.8 and the Sonnet 4.5 model both start returning errors on November 30. What breaks, how to find it in your code and settings, and what Opus 5.5 fast mode costs instead.
Backdoored Claude Code Extension in Cursor: What to Do
Cursor's marketplace served a tampered Claude Code extension for 11 hours that planted an SSH key and a startup script. How to check your machine, clean it up, and set an install cooldown.
Head-to-head comparisons
Pick the layer you're deciding on.
- Cursor vs GitHub Copilot vs Windsurf in 2026: Which AI Code Editor Fits Your Team?
A practical three-way comparison of the editor agents developers actually evaluate now: context quality, pricing predictability, GitHub integration, and review burden.
- Cursor vs GitHub Copilot in 2026: An Honest Comparison
After Copilot's billing change and Cursor's VS Code dominance, which AI code editor wins for real development workflows? Context quality, pricing, and ecosystem lock-in compared.
- Amazon Q Developer vs JetBrains AI in 2026: Enterprise IDE Integration Compared
For AWS-heavy teams and JetBrains shops, these are the tools Cursor vs Copilot debates miss. An honest comparison of ecosystem fit, security scanning, pricing, and where each breaks down.
What We Cover
Browse posts by the kind of tool.
⌨
Editor Agents
Cursor, GitHub Copilot, Windsurf, JetBrains AI, Zed, and Replit: the AI built into the editor you write code in all day.
💬
CLI Agents
Claude Code, Codex CLI, Aider, Continue, and Cline: agents that run multi-step changes from your terminal, with your tests in the loop.
⚙
Autonomous Agents
Devin, OpenHands, and SWE-agent: tools that take a ticket and come back with a pull request for you to review.
🔓
Models and protocols
Claude, GPT, Gemini, Llama, Mistral, and Hermes, plus MCP, A2A, and the frameworks that connect agents to tools.