AI Daily Digest · 2026-09-03
🔬 New AI Craft
1. "Maybe We Shouldn't Be Reviewing All This Code" — Martin Fowler's team rethinks code review in the AI era
Rachel's essay on martinfowler.com argues AI didn't break code review — we've been using it to solve the wrong problems. Once AI handles mechanical checks, line-by-line review of everything dilutes human attention; manual review should focus on design intent and critical behavior. Pairs naturally with yesterday's Copilot Code Review gaining PR-approval powers.
https://martinfowler.com/rachels-ramblings/code-review.html
🛠️ Tools & Tips
1. sepia: a "de-AI" writing skill for your agents
Agent Skills-compatible skill (Claude Code, Codex and 77+ more) that makes agent output read human — 1,555⭐ in its first week.
https://github.com/Nanako0129/sepia
⭐ Open Source Highlights
1. fable51-worlds: Claude Fable 5.1 agent swarms rebuild real cities from open data
Agent swarms research, model and QC real places end to end, shipping browser-native Three.js scenes of SF's Union Square — buildings, storefronts, traffic lights and cable cars all generated from open data and street imagery.
https://github.com/PhiloLabs/fable51-worlds
2. reverify: reverse engineering that answers to the binary
AI-driven RE grounded in deterministic tools — results checked against the actual binary instead of hallucinated (582⭐).
https://github.com/2akouwu/reverify
📰 Industry News
1. DOJ sides with OpenAI in the New York Times copyright lawsuit
2. Investigation: three sites manufactured 215,128 AI-generated "best software" pages — and Perplexity cites them
3. NYC announces the nation's broadest generative-AI moratorium in schools
🚀 Major Releases
1. Google officially ships Gemini 3.8 Flash and 3.8 Flash Cyber (yesterday's rumor, confirmed)
2. Meta launches Muse Spark 1.3, tuned for agentic workflows and competitive coding
