WhyCBuild Academy · learn-while-you-ship
Tutorial mode: ON Arize track · demo
LESSON 00 · why this exists

While they hire interns, you learn how to ship.

Paste a job-posting URL. WhyC builds a working Next.js preview in under a day — and explains every design choice as the agents make it. Each tooltip is a lesson YC charges 7% equity for.

Pipeline progress · spec-fit score

62% → climbing

1 · Spec extract

Gemini · 41s · ✓ done

2 · Codegen

ADK · 3m12s · ✓ done

3 · Deploy

Cloud Run · running…

4 · Self-improve loop

Phoenix MCP · queued

Live lessons grounded in traces

The agent narrates its own reasoning as it works. Hover a card for the trace span ID.

Tutor: Notice the spec-fit only hit 62% on first deploy — that's by design. We ship fast, then let the loop close the gap. Why hybrid? Because waiting for 96% before a URL exists is how you become the YC company that hires for 6 months.
Lesson 01 · spec → eval rubricwhy first?

We extract the spec before writing a single line of UI. Reason: that same spec doubles as the LLM-as-judge rubric in stage 4. One artefact, two jobs.

jobs_to_be_done.functional → judge.criterion[0]
Lesson 02 · design tokens firstwhy tokens?

Generating OKLCH design tokens before components means later regenerations stay visually coherent — the loop can rewrite a flow without re-skinning the whole app.

--accent: oklch(72% 0.19 55) // parodic, not trademarked
Lesson 03 · stateless containerwhy Cloud Run?

Cloud Run scale-to-zero matches the spec's usage_frequency: episodic. A founder who pastes a URL once a week shouldn't pay for an idle VM.

Lesson 04 · regenerate only weak flowswhy partial?

The loop targets flows scoring under threshold, not the whole app. This is the difference between self-improvement and thrash — and the reason Phoenix traces matter.

Lesson recap · what you can steal for your own startup tomorrow

Three takeaways the agent earned by doing — not by reading a Substack.

1Spec is your eval. If you can't write the rubric, you can't ship the agent. Extract structure first, code second.
2Ship before you converge. 10 minutes to a live URL beats 6 months to a perfect one. Let the loop close the gap.
3Regenerate the weak flow, not the app. Trace-driven, surgical fixes are the Arize-track moat. Whole-app rewrites are how token bills explode.