Set Ambitious AI Tasks: Guide & Prompt from Anthropic Lead Engineer

Lead engineer at Anthropic wrote a guide on how to truly set more ambitious tasks for AI, and shared a ready-made prompt.

Felix Rieseberg, who works on Claude at Anthropic, notes that everyone advises being “more ambitious” when using AI, but no one explains how exactly. He outlines three approaches that go beyond simple speed‑up. Moreover, they are applicable to any AI.

Instead of asking the AI to fix errors one by one, give it an end goal. Felix himself stopped asking to fix each failure separately and instead set the task of making the programs not crash at all. He gave Claude error reports and computers running Windows, macOS, and Linux so the model could test fixes on its own.

When preparing a quarterly report, first give the AI all the data and ask it to identify what changed, why it happened, and what actions are needed. Then ask it to highlight the three most urgent solutions with their pros and cons. At an even higher level — ask the AI to create something more useful than the report itself, and demonstrate on the company’s historical data that this would allow problems to be detected earlier.

When the AI gives you a small table, ask it to create a database and a statistical model instead, since a leading specialist in the field would use exactly those tools. If a page is supposed to load in 5 seconds, set a goal of 200 milliseconds. And never let the AI evaluate its own work — use independent verification.

To bring in knowledge from other domains, assemble a team: an analyst, a journalist, an editor, and for critical assessment — a person who challenges the company, and an accounting expert. But don’t ask the AI to simply imitate these professionals. Ask it to conduct the research they would conduct.

Felix also shared a ready-made prompt:

Мне нужна более глубокая версия этой задачи, а не более масштабная.
Прежде чем переписывать мой промпт:
1. Расскажи, что сделал бы ведущий специалист в этой области, чего я сейчас не прошу.
2. Назови максимально исходные данные или материалы, которые он использовал бы вместо уже подготовленной для меня информации.
3. Укажи конкретные методы и инструменты, которые он применил бы.
4. Объясни, какие независимые от нас с тобой доказательства подтвердили бы правильность результата.
Затем перепиши мой промпт так, чтобы он требовал именно этого.

This prompt was turned into the /be-ambitious skill.

Simply point your agent to this skill and ask it to install.

The full post by Felix is definitely worth read.