Field notes on autonomous engineering
Practical writing on the multi-agent loop, cost ceilings, watch-only mode, and what actually decides whether AI agents ship code.
- Engineering
How to roll out AI coding agents to your team without breaking trust
A rollout playbook for engineering leads: pick the first repo, start watch-only, climb the graduated trust ladder, and track the three metrics that matter.
Jul 6, 2026 · 9 min read - Comparison
GitHub Copilot coding agent review: what happens when you assign an issue
An honest review of GitHub Copilot's coding agent: how assigning an issue works, what it costs after the AI Credits change, and where backlog agents differ.
Jul 6, 2026 · 9 min read - Engineering
Devin pricing explained: the ACU, the $20 entry, and the real monthly bill
Devin pricing explained: how ACU billing and usage allowances work, a worked example of a realistic monthly bill, and how flat-quota alternatives compare.
Jul 6, 2026 · 8 min read
All blog
17 more- ComparisonJul 6, 2026 · 8 min read
Devin AI review 2026: what it does well, what it costs, who should buy it
An honest Devin AI review for 2026: what Cognition's autonomous engineer does well, where usage costs surprise teams, and when a backlog agent fits better.
- EngineeringJul 6, 2026 · 9 min read
AI coding agents vs human developers: what the evidence says
Where AI coding agents beat human developers, where humans stay better, what METR and DORA data show, and the math for when delegating a ticket pays off.
- EngineeringJul 6, 2026 · 9 min read
AI coding agents on legacy code: why they fail and how to prepare
Why autonomous AI coding agents fail on legacy codebases — no test signal, hidden conventions, dead code traps — and the preparation that makes them work.
- EngineeringJul 6, 2026 · 11 min read
The AI coding agent security checklist: 8 questions to ask before adopting
A practical security review checklist for autonomous AI coding agents: sandbox isolation, secrets, least-privilege tokens, prompt injection, audit trails.
- EngineeringJun 6, 2026 · 8 min read
How to read AI coding agent benchmarks without fooling yourself
SWE-bench scores sell tools but rarely predict them. A practical guide to what coding-agent benchmarks measure, where they mislead, and the metric that matters.
- ComparisonJun 6, 2026 · 9 min read
The best AI coding agents in 2026, sorted by the job you're hiring for
A field guide to the 2026 AI coding agent landscape — Cursor, Copilot, Claude Code, Codex, Devin, Cline, Windsurf, Aider — sorted by the job each one is best at.
- ai coding agent safetyMay 27, 2026 · 8 min read
Safety-First AI Engineers: The Case for Watch-Only Mode
Why watch-only mode — not auto-merge from day one — is the right default for autonomous AI coding agents. How it builds trust and what enabling auto-merge actually looks like.
- multi-agent codingMay 27, 2026 · 8 min read
Multi-Agent vs Single-Agent AI Coding: What's the Difference?
Multi-agent coding pipelines vs single-agent sessions: how each works, when the complexity pays off, and what it means for PR quality on your real codebase.