Anthropic testers share 4 prompting tips for Fable 5.2
every · x · 2026-11-10
Anthropic's team summarized four tester tips for Fable 5.2:
- Think bigger: give it ambitious goals and let it check its own work.
- Hand it the whole list: send everything as a bulleted list instead of assigning one task at a time.
- Give a way to verify: supply a high-level goal, a system to work within, and a way to check whether the job is done.
- Settle permissions upfront: early runs often stalled waiting for an 'ok' the user thought was already given.
More from coding & agent
- The 'AI software engineer' was never a stable equilibrium — the endgame is automating whole teams — andreisavu · 2026-11-10
- AI-first languages are here: Vork and Exact strip out human-only abstractions — ravisparikh · 2026-11-10
- ChatGPT/Codex desktop app's latest update breaks the send button, user reports — rudrank · 2026-11-10
- Agents Are Remarkably Bad at Using pkill — generativist · 2026-11-10
- Exercise for Product Leaders: Use AI to Deepen PM Work Across the Entire Lifecycle — realmadhuguru · 2026-11-10
- A first-principles beginner's guide to context engineering for agents — sharpeye_wnl · 2026-11-10