Miatz playground

Same Prompt, Two Machines

One prompt, two builds, one hidden eval decides.

Coming to the Build-LabBYOK AIProductpm
The Demystify signature

the hidden eval suite reveals only after both builds finish, turning 'AI just knew what I meant' into a scored, visible list of exactly what it guessed wrong

How it works

What goes in, what comes out

What it does

Learner writes one feature request twice — a loose one-line 'vibe' prompt, then a tight spec with explicit acceptance criteria — and both go to a live coding agent in parallel sandboxes. Both resulting builds run against the same hidden eval suite neither prompt-writer ever saw.

You bring

write the vibe prompt; write the spec plus acceptance criteria; submit both to the agent

You get

a side-by-side code diff of both implementations plus a pass/fail eval readout per acceptance criterion — the vibe version typically fails 2-4 criteria it was never told about

You control

feature-request seed picker (5 seeded prompts, e.g. 'add a discount code field'); fixed default agent/model; eval-suite hidden until reveal

Where it's used

Module 'Vibe-Coding a Prototype You Can Ship' L1 demo-to-lab pair + free no-signup landing-page teaser

specification-driven developmentacceptance criteriaagentic coding drifteval-gated shipping
Optional · watch & try

Go deeper, elsewhere

Hand-picked public explainers and open tools that complement this one — always optional, never required, never graded.

Concepts click when you open the machinery.

Three labs are already live and free — the same hands-on style this playground brings to its module.