Five labs for the AI jobs that happen at work
Pick a role, run a prompt, inspect what left, keep what worked.
Labs
Every lab uses the same six-step bench: Scenario, Work area, Run, Inspect, Judge, and Debrief.
Compare: one task, two briefs
Lab 1.1Write two versions of a brief on the same synthetic material, run both, see how...
What leaves and what stays
What leaves: The scenario material and the two briefs you run. What stays: Your drafts before you press Run, and any recipe you keep on this device.
Ask a document, then verify it
Lab 1.2Ask a question over a synthetic policy or handbook, inspect which passages were...
What leaves and what stays
What leaves: The retrieved passages and your question when you press Run. What stays: The full document in browser memory and your verification calls.
Make sense of a messy table
Lab 1.3Inspect a messy synthetic spreadsheet, spot missing rows and unit traps, and check...
What leaves and what stays
What leaves: Only the column header and at most twenty sample rows when you press Run. What stays: The full table and all local row counts in your browser.
Watch an agent take steps, and stop it
Lab 1.4Give a helper a multi-step job on a synthetic inbox and calendar, set which actions...
What leaves and what stays
What leaves: The goal and the synthetic inbox items for the current step. What stays: Every tool call stays inside the local sandbox and touches nothing real.
Build a reusable work helper
Lab 1.5Write a short reusable helper with clear rules, test it on three cases including a...
What leaves and what stays
What leaves: Your helper instructions and the three test inputs when you press Run. What stays: Saved helpers in your local recipe shelf until you choose to share.