~/guides/AMBITION.md
One paragraph that makes Claude think deeper: the ambition prompt from an Anthropic engineering manager
Anthropic's Felix Rieseberg splits ambition with AI into four types and shares a paragraph to paste after your prompt. The text, four questions, examples.
Most of us ask Claude small things: write this email, fix this error. Felix Rieseberg, an engineering manager at Anthropic who has led engineering for Claude.ai, Cowork and Claude Code Desktop, wrote a tutorial in October 2026 saying we are not ambitious enough with AI, and he shared a short paragraph: paste it after your prompt and Claude first works out how a top expert would approach the task, then rewrites your prompt at that level. This is the full version of the AMBITION reel: the paragraph (in English and in our Chinese translation), the four things it makes Claude answer, the four types of ambition and the examples from his essay.
01The result first: the same Claude, from "fix this crash" to "keep the app crash-free"

By the end of this guide you will have:
- A paragraph you can copy and paste after anything you were going to ask (in English and Chinese).
- Claude answering four things first: what a top expert would add, what raw material they start from, which methods and tools they use, and what would prove the result right.
- Your prompt rewritten into a deeper version, which you then use for the real work.
- The four types of ambition (speed, altitude, depth, range), explained with the examples from Felix's essay.
02Step 1: paste this paragraph after your prompt
![Reel frame: the essay's prompt block, first sentence highlighted, a card reading [your prompt] + Felix's paragraph](/assets/articles/felix-rieseberg-ambition-prompt-claude/prompt.webp)
Write what you want Claude to do as usual, then add this paragraph after it (our Chinese translation of the original):
[你原本的提示詞,例如:幫我做一個咖啡店的菜單網頁] 我要的是這個任務更深入的版本,不是更大的版本。改寫之前,先:(1) 說出這個領域的頂尖專業人士會做、但我沒有要求的事;(2) 說出他們會從哪些最原始的素材開始,而不是別人已經幫我整理好的東西;(3) 說出他們會用的具體方法和工具;(4) 說出除了你和我之外,什麼能證明結果是對的。然後改寫我的提示詞,明確要求這些。
[your 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.
03Step 2: read its four answers, then start from the rewritten prompt

Claude answers these four first, and each is worth a look:
- 1. Expert extras: what the best people in the field would do that you didn't think to ask for.
- 2. Raw material: they start from the rawest source (raw data, logs, the original text), not a summary someone already made.
- 3. Methods and tools: the methods and tools they actually use, not just expert-sounding words.
- 4. Outside proof: something other than you or Claude that shows the result is right, such as a test it can run, real data or a number you can measure.
If the direction looks right, tell it to go ahead with the rewritten version:
Good, start from the version you rewrote. If you are missing data or access (for example: [files, a site's admin, data]), list it and ask me first.
04The four types of ambition: speed, altitude, depth, range

- Speed: the same work in less time. This is how most people already use AI, so Felix skips it.
- Altitude: go up a level and aim at the outcome, not the task in front of you. His examples: not "fix this crash" but "keep the app crash-free"; not "make the quarterly review slides" but asking what decisions the review is really for.
- Depth: work like the field's best expert, down at the underlying data, tools and problems. When Claude reaches for markdown tables, insist on databases and statistical models; your site needs to load in 5 s instead of 10 s? Ask for 200 ms.
- Range: look at the work from several fields. Turn the noun into verbs (a quarterly review isn't "a deck", it's getting busy people to believe a story about cause and effect and decide in 30 minutes), then ask which experts you'd bring in: an analyst, a journalist, an editor, even a short seller and a forensic accountant.
To use altitude, start your prompt with the goal behind the goal:
Right now I need to [the task in front of me, e.g. put together this quarter's sales report], but what I really want is [what should happen next, e.g. leadership catches problems early and moves money and people to the right places]. First tell me whether there is a better way than my original approach to reach that goal, and how we would show it's better.
05Optional: install it as a Claude Code skill

Don't want to paste it every time? Someone on GitHub turned Felix's essay into a Claude skill (its README says all the ideas are his). It is a community project, not something Felix or Anthropic published. In a terminal:
git clone https://github.com/robonuggets/be-ambitious ~/.claude/skills/be-ambitious
Then, in Claude Code:
/be-ambitious [what you want done, e.g. make a menu web page for a coffee shop]
06Good to know
- Telling it to "act like an expert" is not enough: Felix says models often switch vocabulary without switching methods. Ask it to do what the expert actually does, with the expert's tools.
- Deeper, not bigger: if a suggestion only makes the job larger without making it better, drop it.
- Get an outside check: don't let Claude tell you it did well; find a test, real data or a measurable number that shows it.
- All of this needs the model to have a computer, be able to install software and reach the relevant data (for example through MCP connectors); in a plain chat window it can do less.
- The Chinese version is our translation; for the exact meaning, Felix's English original wins.
- You found this guide by typing AMBITION on the site from the link in bio; nothing is sent by DM.
Sources
What should the next reel cover?
I test Claude and AI tools every day and turn the useful ones into reels. Follow @vivianlin.tech to catch each new guide.