Lesson 7 — Automation with Codex and Playwright

Duration: 120 minutes

Included delivery track: Follow LAB_SETUP.md for the supplied files and local simulation. Required activities use these materials; connected deployments are optional extensions. Instructor debrief: answer key. Real OAuth, cloud releases and payments are simulated or discussed through evidence packets. Deck: 07-automation-with-codex
Lab: Build One Durable Regression Check

Learners will be able to

select a stable candidate for automation;
write a small browser/API assertion using semantic selectors and deterministic outcomes;
use Codex to explain, draft, review, and improve a test while retaining authorship of the intent; and
recognize flaky-test causes and avoid masking them with unlimited retries.

Instructor preparation

Use sandbox/starter.spec.mjs and runtime CANDIDATE values buggy and fixed. Preinstall optional Playwright before class if approved. No login fixture or branch switch is needed. Prepare the unexecuted assertion-design route for blocked installations.

Agenda

Time	Facilitation
0–10	Arrival: teams decide which of four findings deserves automation and defend the choice.
10–25	Teach the automation pyramid and the rule “scripts verify mechanics; humans/AI explore ambiguity.”
25–40	Live build: use Codex to inspect an existing test, propose a semantic locator, add an assertion, run it, and inspect a failure.
40–50	Guided practice: turn an acceptance criterion into Given/When/Then and then a deterministic assertion.
50–60	Break.
60–95	Lab: pairs complete one test. Driver edits/runs; verifier reviews the diff, stability, and evidence. Switch after the first green run.
95–110	Flake clinic: show a dynamic date, animation, and weak selector. Teams choose a principled fix rather than increasing retries.
110–115	Explain CI, artifacts, and why a green test alone does not prove broad product quality.
115–120	Exit ticket: state what makes the new check deterministic.

Formative coaching

Builder route: the same check must fail Candidate A and pass Candidate B with stable selectors and a meaningful name. Foundation route: review a precise unexecuted assertion design, predicted A/B results and limits. Do not demand fabricated logs; record actual execution as a later supervised milestone.

Product Studio connection — Protect your progress

Use up to five minutes of the existing closing/debrief time for a project next-step check; do not extend the 120-minute session. The build work happens as homework. Add a meaningful regression and own-repo CI. Record actual fail-before/pass-after evidence for a repair, then prove the current checks pass without weakening expected behavior.

Follow the milestone map and review checklist. Record support needs privately; practice lab quality criteria are coaching, not an exam or numeric completion gate.