How to Be More Ambitious — A Tutorial for Humans With Models (Felix Rieseberg)

An 8-minute essay by Felix Rieseberg, an engineering lead on Claude at Anthropic, published October 13, 2026. It turns the stock advice “be more ambitious with AI” into a method. One prompt from it went viral on its own (Anthropic’s lead engineer shares a… (tweet)).

The framework: four kinds of ambition

• Speed: the same work, faster. This is the established case, so the essay skips it.
• Altitude: go up a level of abstraction and aim at the outcome, not the task. He went from asking Claude to fix individual crashes to asking it to keep the apps crash-free, which first meant giving it crash reports, build-to-commit mapping and machines it could use to reproduce bugs. Worked example: a quarterly business review moves from “fill in the template” to “what changed and why,” then to “the three most urgent decisions,” and finally to “design a better mechanism than a quarterly deck, prove from our history it would have caught things sooner, and run it alongside for a quarter.” A tell that altitude is right: there will be meetings about the work.
• Depth: do the task the way the best person in the field would.
  • Insist on the expert’s actual instruments, such as databases and statistical models instead of markdown tables. Models tend to change their vocabulary without changing their methods.
  • Set unreasonable goals (200ms, not 5s) and tell the model they are reasonable.
  • Have it adversarially test your assumptions against raw data, then hunt for new knowledge, such as an earlier churn signal validated on held-out data.
  • Use an outside judge. Never let the model grade itself, at least as of October 2026.
• Range: bring in more fields than used to be economical.
  • Replace nouns with verbs: a QBR is “getting busy people to believe a story about cause and effect, and to decide under uncertainty, in 30 minutes.”
  • Then assemble the heist crew, for example an analyst, a journalist, an editor, a short seller and a forensic accountant. Ask what a short seller does, not how one writes, or you get role-play instead of research.

The prompt

I want a deeper version of this task, not a bigger one. Before rewriting it: (1) say what a top professional in this field would do here that I’m not asking for; (2) name the rawest material they would start from instead of what has already been digested for me; (3) name the specific methods and tools they would use; (4) say what, other than you or me, would show the result is right. Then rewrite my prompt to ask for exactly that.

Why it’s necessary: self-efficacy

The closing section argues the bottleneck is belief, not capability. People with high self-efficacy accomplish more. Specific hard goals beat “do your best” (Locke and Latham). People do hard things when they believe them possible (George Dantzig solving two open problems he took for homework), after watching someone else do it (Bannister’s four-minute mile, and two runners going sub-two-hour in the same 2026 London Marathon), and when they give themselves permission to fail (The Inner Game of Tennis; Linux as “just a hobby”). He also points to Patrick Collison’s “Fast” list. Because capabilities now jump every few weeks, the techniques that let you attempt what you didn’t think possible matter more.

Where it fits

A practical companion to Ambition and AI Workflows. It is also the most concrete answer in the corpus to AI Homogenization: depth means starting from raw material and outside checks, which is how you get off the population mean that Miles Cranmer warns about.