Creator of Anthropic's Claude Code wants you to stop micromanaging your AI

• Claude Code creator Boris Cherny said users are underestimating the capabilities of newer models.
• He advised users to stop giving a model overly specific instructions and let it "cook."
• Cherny said some prompting habits could be obsolete as models continue to improve.

If AI users are the new bosses and agents are the new employees, then the creator of Anthropic's popular coding agent has a piece of evergreen management advice: Stop micromanaging your AI.

Boris Cherny, the creator of Anthropic's Claude Code, said at a Y Combinator event on Saturday that one common mistake he sees from Claude users is giving overly detailed, step-by-step instructions because they underestimate the tasks a model can handle.

"They're like, 'I want you to do this, but I want you to do it in this way, this way, this way. You must do one, then two, then three, then four,'" Cherny said. "For modern models, that's actually really not the way to do it."

Instead, Cherny said users should be specific about the outcome while giving the model more freedom to decide how to get there. He recommended describing the task, setting the guardrails, and defining what a successful completion looks like.

"Then just go let the model cook and come back in a little bit," he said. "I think it'll surprise you."

Cherny's advice reflects how quickly AI models can improve, potentially rendering older prompting habits obsolete. Cherny said his approach may not have worked six months ago, but it does with today's more capable models.

Google Brain cofounder Andrew Ng shared a similar idea last year, calling it "lazy prompting" — when users provide minimal instructions as models get smarter.

"We add details to the prompt only when they are needed," Ng said.

Cherny said Anthropic employees are also finding new capabilities as they experiment with the company's models. He estimated that there are probably dozens, if not hundreds, of useful abilities that have yet to be discovered.

So, AI power users: Don't repeat the same mistakes bad managers make. You could be getting in your AI's way.

Have a tip? Contact this reporter via email at [email protected] or Signal at lloydlee.71. Use a personal email address, a nonwork WiFi network, and a nonwork device; here's our guide to sharing information securely.