How to be more ambitious: A tutorial for humans with models
Source
felixrieseberg.com
Date
Context
Felix Rieseberg, who works on Claude at Anthropic, describes one form of ambitious AI use, which he calls altitude: giving a model responsibility for an outcome rather than an individual task. His example is moving from requests to fix individual crashes toward keeping his team's apps crash-free. That required access to crash reports, a way to connect application versions to source code, and computers where the model could reproduce failures and check repairs. The broader assignment depended on changing the working environment. His essay also recommends outside validation, ideally through tools that provide immediate feedback.
Key takeaways · AI-distilled
Felix Rieseberg, who works on Claude, splits AI ambition into four types: Speed (same work faster), Altitude (ask for the outcome, not the task), Depth (work the way a field-leading expert would), and Range (pull in methods from other fields).
For altitude, his example is moving from asking Claude to fix individual crashes to asking it to keep the apps crash-free, which first required giving it crash reports, version-to-commit mapping, and Windows, macOS and Linux machines to reproduce and verify fixes.
For depth, he warns that models told to act like professionals often change vocabulary without changing methods, so he insists on experts' instruments (databases and statistical models instead of markdown tables) and asks the model to test assumptions against raw data.
He advises against letting a model judge its own work as of October 2026, preferring an outside check, ideally a tool the model can run for immediate feedback.
For range, he suggests restating the task as verbs and assembling a cross-field crew, but cautions against role play: ask the model to do what a short seller does to research a position, not to write in a short seller's voice.
