Source: ai-research/be-more-ambitious-altitude-depth-range-2026-10-27.md (primary essay, fetched 2026-10-27) + raw/Anthropic_just_shared_how_they_use_Claude_differently_Master_it_in_12_mins.md (secondhand YouTube walkthrough by Jay of RoboNuggets)

Author: Felix Rieseberg (“I work on Claude”) URL: https://felixrieseberg.com/how-to-be-more-ambitious-a-tutorial-for-humans-with-models/ Published: 2026-10-13

Felix Rieseberg’s essay “How to be more ambitious: A tutorial for humans with models” turns the slogan “be more ambitious” into concrete techniques. He describes four types of ambition: speed, altitude, depth and range. He skips speed as already well understood. Altitude means asking for the outcome instead of the task. Depth means doing the task the way a field-leading expert would. Range means bringing in other fields. The essay includes a reusable four-question prompt for depth. The walkthrough video that brought it into the inbox covers only three types and some of the techniques, so this article is built from the essay. Rieseberg says his advice “is not model-specific”.

Key Takeaways

  • Altitude: ask for the goal beyond the goal. “What do you want to happen next? What’s the goal beyond the goal?” He moved from asking Claude to fix individual crashes to asking it to “simply keep our apps crash-free”.
    • Altitude needs workflow changes. To make crash-free possible, he first gave Claude crash reports, a mapping from app version to git commit, and Windows, macOS and Linux machines to reproduce and verify fixes.
    • When to use it. It “works best if you really are after some ultimate outcome; it works less well for creative endeavors”. His test: you are going to hold meetings about the work. Limits remain: “Claude, go make money” does not work for most businesses in 2026.
    • QBR ladder (his example). First “Tell me what changed this quarter, why, and what we should do about it”. Then have Claude work out the three most urgent decisions with for/against arguments. Finally ask it to “Design a better way to do that than a quarterly deck, show from our own history that it would have caught things sooner, and run it beside the current one for a quarter.”
  • Depth: change the method, not the vocabulary. “When prompted to act like a professional, models often switch vocabulary without actually switching methods.” He uses five techniques:
    • Insist on expert instruments. When Claude reaches for markdown tables, insist on databases and statistical models. “You don’t necessarily have to know what the experts use, Claude usually does.”
    • Set unreasonable goals. Need 5 s instead of 10 s? Ask for 200 ms. If the model says it’s impossible, tell it the problem “is very much solvable, it just needs to keep going”.
    • Challenge what you know. Give it raw data and have it test your assumptions adversarially: “We believe churn is an onboarding problem. Try to reproduce that from the raw data.”
    • Produce new knowledge. Have it simulate customers from your logs, then find “the earliest signal of churn we don’t know about yet” and show it predicts on held-out data.
    • Use an outside judge. “Do not ask the model to judge itself.” In October 2026, he insists on outside validation, ideally a tool the model can run for immediate feedback.
  • Range: bring in other fields. “Every single heist movie spends the first 30 minutes assembling a diverse cast.” Multi-agent setups make this cheaper than it used to be.
    • Replace nouns with verbs (from The Art of Innovation, Kelley and Littman). A quarterly review isn’t “a deck”; it’s “getting busy people to believe a story about cause and effect, and to decide under uncertainty, in 30 minutes”.
    • Assemble the crew. His QBR crew: an analyst, a journalist and an editor; a motion and audio designer if it will be presented; and a red team of a short seller and a forensic accountant.
    • Avoid role-play. “Don’t ask the model to write like a short seller, ask it to do what a short seller does to research their position.” Role-play makes the transcript weirder without improving the outcome.
  • Prerequisite for all of it. The model needs “a computer with the ability to install software and access to all kinds of relevant data (either using MCP connectors or other means)”.
  • Self-efficacy close. People who believe they can do more accomplish more. He cites Locke and Latham’s 35 years of goal-setting research and three hacks: believe it is possible (George Dantzig’s “homework”), watch someone else do it (Bannister’s four-minute mile), and give yourself permission to fail.

The depth prompt (verbatim from the essay)

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.

How the secondhand video differs

  • The video (Jay, RoboNuggets) presents three methods. The essay has four types and explicitly skips speed.
  • The video covers three depth techniques. The essay has five; the video omits “challenge what you know” and “produce new knowledge”.
  • The video’s crew example (analyst, sales manager, graphic designer) is the creator’s own, not Rieseberg’s. It also leaves out his warning against role-play.
  • The video describes Rieseberg as lead engineer on the Claude app, Cowork and the Claude Code desktop app, and says the essay “now has over 4.5 million views on X”. The essay itself says only “I work on Claude”; the view count is the creator’s claim.
  • The video’s marketing examples are the creator’s own. A pitch deck whose goal is “impress the client” might become interactive HTML slides or a 3D walkthrough of a venue. A house listing becomes “helping a family picture their life in this house before they ever visit”.
  • The video promotes a paid community and course.

Try It

  1. Before your next big prompt, run the depth prompt above on it and use the rewritten version. Check whether it names a real outside check (a test, a speed benchmark, a data holdout).
  2. Altitude check. If you will hold a meeting about the work, ask for the decision the meeting exists to make instead of the deliverable. Then list what access the model would need to own that outcome (data, logs, machines) and grant it.
  3. For a client deliverable, write the verb. For example, “a dental practice website” becomes “getting a nervous patient to book a first visit”. ^[inferred] Then ask Claude which fields that verb calls for, and have each field do its own work (research, data, design), not role-play.
  4. Add one outside judge to any agent loop you rely on: a script, a test, a subagent with a rubric, or real data. Do not let the model grade itself.

Open Questions

  • The essay’s view count on X (4.5 million per the video) is unverified.
  • Rieseberg expects self-judging to “eventually be solved with increasing model intelligence”; no timeline or evidence is given.