Darius: People who work on AI have this rallying cry right now. You see it everywhere: 'We ought to be more ambitious.'
Harper: Right. 'You're not using our incredibly powerful, expensive tool correctly.' I always read that as a sales pitch.
Darius: See, I think that's the common reaction. It sounds like a complaint. But this piece I read by Felix Rieseberg, who works on Claude, he says he understands how tricky that advice is to actually follow. So he wrote a tutorial.
Harper: A tutorial for ambition. Okay.
Darius: And it’s not just 'think bigger.' He breaks ambition down into three types you can actually use: Altitude, Depth, and Range. He skips the first one, Speed, because 'do the same work faster' is what everyone's already doing.
Harper: That's 99% of my usage. 'Make this email sound less angry.' Done.
Darius: Exactly. So his first real category is Altitude. It's about going up a level of abstraction. Instead of giving the AI a task, you give it the outcome you want.
Harper: What does that look like?
Darius: His example is a quarterly business review, a QBR. The low-ambition version is, 'here's a template and some new numbers, make me a deck.' We've all done that.
Harper: Yep.
Darius: A higher 'altitude' version is giving it all the raw data you can find and saying, 'Tell me what changed this quarter, why, and what we should do about it.' You're asking for the insight, not the slides.
Harper: Okay, that makes sense. You're asking it to do the analysis part.
Darius: But then he goes even higher. The ultimate goal of a QBR is for leadership to spot problems and move resources. So the most ambitious prompt is: 'Design a better way to do this than a deck. Show from our own history that your new process would have caught problems sooner. And then run it in parallel to the old way for a quarter.'
Harper: Whoa. So you're asking it to fire itself. Or, fire the whole quarterly review process.
Darius: You're asking it to solve the real problem, which is 'how do we catch problems early,' not 'how do we make a deck.' You're moving from task to outcome. But he does say, to do this, you have to give the AI access to your systems, your data, a computer it can run things on... it's a change in your whole workflow.
Harper: There's the catch. It’s not just a better prompt, it's an infrastructure project.
Darius: Right. Okay, so the next one is Depth. This is for when you don't want a different outcome, you just want the task done at a world-class level. His example is customer churn. A model might give you a nice markdown table. He says you have to insist on the same tools an expert would use. Tell it, 'No, use a real database, build a proper statistical model.'
Harper: But what if I don't know what tools the experts use?
Darius: This is the part... I'm not totally sure about. He says you can just ask the model what an expert would do. 'Tell me what a top professional would do here that I'm not asking for.' Which feels a little like pulling on your own bootstraps, but he claims it works.
Harper: Huh. You're asking the intern to tell you how the CEO would do their job.
Darius: Kind of. And then you tell it to challenge your own assumptions. Like, 'We think churn is an onboarding problem. Try to prove that's wrong using our raw data.' You use it to find your own blind spots.
Harper: Okay, the adversarial check is interesting. So what's the last one? Range?
Darius: Range. This is where he says the most magic happens. The idea is to bring in perspectives from completely different fields, like a heist movie.
Harper: A heist movie.
Darius: You're assembling a crew. For that QBR deck, don't just ask an 'analyst' AI. Bring in a journalist to find the story, an editor to tighten the language, a motion designer for the presentation. And then, his example, bring in a forensic accountant and a short seller to 'red team' the whole thing and find the weaknesses.
Harper: Okay, wait. That just sounds like roleplaying. 'Pretend you're a skeptical short seller.' It's going to generate a list of generic business risks. It's not going to have a genuine insight.
Darius: He's very specific about that. He says don't ask it to be a short seller, ask it to do what a short seller does to research a position. Use their methods on your data.
Harper: But the methods of a real short seller involve calling people, using a network, having decades of intuition about market sentiment. An AI has none of that. It has an archive of the internet. It can pattern-match what a short seller's report looks like, but it can't replicate the actual work that produces the insight.
Darius: I see your point. But maybe the value is having a completely unbiased viewpoint look at your raw numbers. A human short seller has biases. The AI just has the data you give it. Maybe it finds a correlation that every human in your industry has been trained to ignore.
Harper: I'm not convinced. It feels like a way to generate very plausible-sounding nonsense. I think there's a limit to what you can get by just mashing up different fields.
Darius: That's fair. The author does wrap up by explaining why he thinks all of this is even necessary. It comes down to a psychology concept: self-efficacy.
Harper: Self-efficacy.
Darius: Basically, your belief in your own ability to do something. And research shows that people who are given specific, hard goals perform way better than people who are just told 'try your best.' He's saying we can use AI to set and chase those harder goals—goals we'd normally think are impossible for us. It’s a tool to hack our own sense of limitation.
Harper: So it’s not about prompt engineering. It's about ambition engineering. For yourself.
Darius: That's it exactly. I'm Darius.
Harper: I'm Harper.