How to be more ambitious: A tutorial for humans with models

Open original

Felix Rieseberg, who works on Claude, argues that most people use AI for basic tasks like speed gains when they could tackle much harder problems. He breaks ambition into four categories — speed, altitude, depth, and range — and walks through concrete techniques for each, using a quarterly business review as a running example.

  • Altitude means shifting from completing individual tasks to solving the underlying outcome; instead of asking Claude to fix crashes one by one, ask it to keep your app crash-free entirely, which often requires restructuring your workflow and data access.
  • Depth means pushing models to use the same tools, methods, and raw data that field experts would use, then validating results against external judges rather than asking the model to judge itself.
  • Range means pulling techniques from multiple disciplines — treating a quarterly deck not as "a deck" but as "getting busy people to believe a story and decide under uncertainty," then assembling a cross-disciplinary team (analyst, journalist, short seller, forensic accountant) to bring in different perspectives.

Felix Rieseberg, who works on Claude at Anthropic, argues that people consistently underuse AI's actual capabilities. He's observed incredible applications this month alone—helping discover a new planet, creating compelling AI music videos, advancing mathematics—yet most users still treat these models as fancy search engines. The gap exists because model abilities are uneven. They excel in some directions while falling short in others, and without deliberate effort, people default to trivial tasks. Rieseberg frames his advice around four types of ambition: Speed (faster execution of existing work), Altitude (targeting higher-level outcomes), Depth (working like field experts), and Range (pulling from multiple disciplines). Speed is already well-established, so he focuses on the other three. Altitude means reframing what you're actually trying to accomplish rather than just automating the immediate task. Instead of asking Claude to fix individual app crashes, Rieseberg started asking it to keep the entire system crash-free—a shift that required restructuring workflows so the model could independently fetch crash reports, map versions to commits, and access multiple operating systems. For a quarterly business review, the altitude approach escalates from "make slides from this template" to "identify the three most urgent decisions we face and show arguments for each" to "redesign how we catch problems early instead of waiting for quarterly reviews." The catch: this works best when you're chasing a real outcome and have meetings about it, not for creative work where the process matters as much as the result. Depth means pushing the model to work like the best professionals in a field rather than accepting surface-level execution. With the churn problem, this means insisting the model use actual statistical tools and databases instead of markdown tables, setting seemingly impossible targets to force deeper thinking, and most importantly, using raw data to challenge assumptions rather than accepting digested conclusions. Rieseberg emphasizes getting outside validation—don't let the model judge its own work. He offers a concrete prompt: before rewriting a task, have the model identify what a top professional would do that you're not asking for, what raw materials they'd start with, what specific methods they'd use, and what external measure would prove it's right. Range, the fourth type, involves pulling techniques and knowledge from multiple fields at once, something that's only become economical now that you can ask an AI to do the synthesis work.

Ask about this article