Lesson 06 · Activity 03

Build, Test & Iterate

You have a spec (Activity 1) and an instruction block (Activity 2). Now stand the assistant up in a real builder, then do the part most people skip: stress-test it. An assistant is not finished when it answers your first easy question — it's finished when it survives the awkward ones. Pick your platform, work the build checklist, then log what breaks so you know exactly which instruction to tighten.

Part A — Build it

Choose where you're building. The checklist updates to match that platform's flow. Your progress is saved on this device for each platform.

Platform

0 of 6 done

Part B — Stress-test log

Run each edge case against your assistant and record what it actually did — not what you hoped it would do. Mark a result, and when something's off, write the one-line instruction fix in the last column. The log is saved on this device.

Edge case	Test prompt you used	What it did	Result	Fix to instructions	Remove row

Iteration write-up

After one round of fixes, what's the single instruction change that improved behavior most — and what still needs work?

Saved automatically on this device.

The loop that makes it reliable: build → test the edge cases → note the misses → tighten one instruction at a time → re-test. Most failures are fixed by a single explicit "do not" rule or one short example, not by rewriting everything. When the same prompt passes twice in a row, that behavior is locked — move to the next failure.