Dominika Michalska

Tools

Practical tools for evaluating AI and deciding where people stay involved. Start with a worked example, then adapt it to your team.

Main tool

TRACE autonomy checklist

About 15 minutes, per automated step

Five questions to answer before you let a system decide something on its own: Track, Rank, Assign, Challenge, Evaluate. You get a list of what needs work before launch, each item with an owner and a next step. It saves your work in your browser, and has a worked example.

Open the checklist

Also

Eval set planner

About 10 minutes

Plan the five kinds of cases an eval set needs, and copy the plan.

The long versions

GitHub, CC0

The autonomy boundary checklist and how to build an eval set that survives a real product. Use them, change them, no credit needed.

Trust-Critical AI

Notes on building AI products, testing decisions and learning from what goes wrong. Sent when I have something worth sharing.

Read and subscribe to the newsletter ↗