Claude Code lead's tutorial: set absurd goals and push models beyond their comfort zone

量子位 · wechat · 2026-10-18

Felix Rieseberg, who leads Claude Code desktop, published a long tutorial on "being more ambitious with AI," organizing ambition into speed, height, depth, and breadth.

Core techniques

Set unreasonable goals: want load time cut from 10s to 5s? Demand 0.2s, then insist the problem is solvable and keep pushing.
Height: raise the abstraction level—for a QBR, have Claude analyze what changed and why, surface the three most urgent decisions, or even redesign the review mechanism and validate it against historical data.
Depth: force expert-grade tooling (databases and statistical models over Markdown tables), raw sources over pre-digested content, challenge your assumptions, hunt for unknown unknowns, and use external validation instead of self-grading—he shares a reusable prompt that rewrites your task the way a top expert would frame it.
Breadth: replace nouns with verbs ("make a PPT" → "convince busy people of an explanation in 30 minutes and make decisions under uncertainty"), then assemble a cross-disciplinary team of approaches—a skeptic or forensic accountant's method, not role-play.

He ties this to self-efficacy research, the Dantzig anecdote, and the 4-minute mile: clear, challenging goals measurably improve performance. All methods assume the model can operate a computer and access data (e.g., via MCP connectors).

Original post →

Claude Code lead's tutorial: set absurd goals and push models beyond their comfort zone

More from coding & agent