[Anthropic] MLE round where you use their own model as a collaborator

Question

Unusual Anthropic MLE detail worth knowing: in one technical round they hand you access to their model and watch *how you work with it*, not whether you can solve it bare-handed.

It rewards people who actually know how to drive a model well: how you prompt it, when you trust it, when you check it. Felt like the most "2026" interview round I've done, and a preview of how the job actually works there.

Anonymous #1 · Answer

This is such a smart round design honestly. Half the skill of the job now is knowing when the model is confidently wrong.

Anonymous #2 · Answer

Did they give guidance on what tools were allowed or was it open-ended?

Anonymous (OP) · Answer

Open-ended in my round; that was kind of the point. They wanted to see what I'd reach for unprompted.

[Anthropic] MLE round where you use their own model as a collaborator

Comments (3)