@sleepy0x13: I increasingly feel that many people's imagination around AI has been boxed in by their past decade-plus of work experience. Today I came across an article by Felix Rieseberg, engineering lead at Claude, titled "How to be more ambitious," about how to use AI with greater ambition…

I’m increasingly convinced that most people’s imagination about AI has been rigidly limited by their past decade-plus of work experience.

Today I came across an article by Felix Rieseberg, the engineering lead at Anthropic behind Claude, called “How to be more ambitious” — a guide on how to use AI more ambitiously. I think every heavy AI user should read it, myself included.

He lays out a concrete methodology.

Take a company’s quarterly business review. Most people’s first instinct is to have Claude help build a PowerPoint deck; the more ambitious ones might toss in some data and ask it to analyze business performance.

Felix argues that this is barely the starting point.

You could keep pushing: ask the AI to identify the three most critical decisions worth surfacing to leadership, and analyze the pros and cons of each. Then go further — since the purpose of a quarterly review is to catch problems early and allocate resources wisely, why does it have to be a PowerPoint at all? Could you design a better mechanism, validate it against historical data to see if it surfaces problems earlier, and then run it in parallel with the existing process for a quarter?

He’s lived through this kind of shift himself.

He used to have Claude fix one software crash at a time. Later, he started simply asking Claude to continuously maintain the application’s stability. To do that, he had to give the AI access to crash logs, code version history, and test environments across different operating systems.

I think this is a particularly telling example, because it shows that “being more ambitious” often means you have to re-examine the ways of working you’ve long taken for granted.

Felix also highlights two other dimensions.

One is depth — having the AI actually use the methods, tools, and raw data of top experts in the field, rather than merely mimicking how experts talk.

The other is breadth — making cross-disciplinary collaboration routine, something that used to be prohibitively expensive. To analyze a company, you could simultaneously bring in the investigative methods of a journalist, a data scientist, a short seller, and a forensic accountant.

He also emphasizes, of course, that final outputs must be validated externally — the model shouldn’t grade its own homework.

After reading it, I kept thinking about one thing.

We’ve gotten quite good at having AI do more and more of our work. One person spins up a dozen agents, pushes a dozen tasks forward simultaneously, and stays frantically busy all day. But if you look closely, these tasks might be no different from what they were five or ten years ago.

Many so-called AI workflows are just slapping a faster engine onto the way work was already being done.

What I think will truly set people apart going forward is the ability to redefine the problem itself — and even the courage to abandon workflows you once considered non-negotiable.

If your understanding of the job stops at “deliver a PowerPoint,” then no matter how powerful the model you’re given, the best you’ll get out of it is a faster, prettier PowerPoint.

AI’s capabilities can undergo a seismic shift every few months, but human inertia doesn’t auto-upgrade alongside it.

That may be the hardest part of “using AI more ambitiously.”

Felix Rieseberg (@felixrieseberg): Hi, I’m Felix. I’m an eng lead working on Claude. I love watching people take flight and do things they never thought possible.

We often describe that as “being more ambitious” when using AI, which is tricky advice to follow. I wrote a tutorial: