Claude engineering lead's guide: ambition with AI as altitude, depth and range

alex_verem · x · 2026-10-31

Felix Rieseberg, engineering lead behind Claude's apps, published a tutorial on being more ambitious with AI. His premise: model capabilities grow unevenly, so most people never discover what models can do — and "be more ambitious" is useless advice unless you know what to change.

Altitude: ask for the result after the result. He stopped asking Claude to fix individual crashes and asked it to keep his apps crash-free — which only worked once he built the setup: easy crash-report access, every app version mapped to a git commit, and Windows/macOS/Linux machines for Claude to reproduce and verify fixes.
Depth: work like the field's best would — real tools (databases, statistical models) instead of tidy tables, concrete targets (200ms load time, not 5 seconds), and an outside judge: a test or tool the model can run decides when the job is done.
Range: pull in multiple fields. For a quarterly business review, have the model do the actual work of an analyst, journalist, editor, and a red team with a short seller and forensic accountant — not roleplay.

He cites Locke & Latham's 35-year goal-setting review (specific hard goals beat "do your best") and Bannister's sub-4-minute mile, broken by others 46 days later: the limit is usually our own idea of what we're allowed to ask for.

Original post →

Claude engineering lead's guide: ambition with AI as altitude, depth and range

More from coding & agent