How We Surfaced Our Engine to Users — Engine + Studio + Lab — FrootAI Blog

Why transparency wins

Most AI code generators treat their pipeline as a black box. You paste a prompt, get generated infrastructure, and have no idea which modules back each resource, what confidence level the resolver used, or whether a WAF check even ran.

We took the opposite approach: **expose everything**.

The `/engine` page shows every stage of the 7-stage harvest pipeline and 5-component AVM chain — parsed live from our masterplan markdown at build time. The `/studio` page lets any visitor browse the AVM catalog, try a composition, and see cost projections without signing in. The `/lab` page publishes reproducible benchmarks with runnable scripts and CC0-licensed datasets.

This post covers **why** we built these three surfaces, **what** shipped, and **what the data says** so far.

---

The three surfaces

/engine — How FrootAI Works

The engine tour discloses the full pipeline: S1 Discover → S2 Fetch → S3 Extract → S4 Retrieve → S5 Scaffold → S6 Compose → S7 Customize, plus the AVM Intelligence chain (C1 Catalog → C2 WAF/CAF → C3 Resolver → C4 Composer → C5 Validator+Emitter).

Every number on the page — milestones shipped, AVM modules indexed, retros published — is parsed from source at build time. Zero hand-typed marketing copy.

**Why it matters**: a CIO clicking "How was this made?" gets an audit-ready answer. Competitors cannot replicate this without publishing comparable engineering artifacts.

/studio — Try Before You Buy

Studio is the only try-before-buy surface in the agentic-IaC category. Visitors can:

Browse 471 Azure Verified Modules by provider, type, and tag
Explore 101 reference Solution Plays with confidence scores
Preview a composition workbench (read-only until V3 ships)
Run a live harvest demo (safe-listed repos, 1 run/hour free tier)

No sign-in required for read-only flows. The trajectory projects +25% paid-tier sign-ups attributable to Studio touch.

/lab — Cite-able Research

Lab publishes three benchmark families today:

1. **AVM vs Hand-Authored Cost** — monthly cost delta across 100 enterprise workloads 2. **Carbon per Region** — carbon footprint across 200 compositions by Azure region 3. **Pipeline Speed** — harvest throughput across Node vs Python vs azd template ingest

Every benchmark ships a runnable script. Datasets are CC0-licensed Parquet. Procurement teams and Microsoft FastTrack partners forward Lab posts — that's earned distribution.

---

What the A/B test says

We ran a 7-day A/B test (`engine.panel_ab_test`) with a 50/50 split on `/orchard/<slug>` traffic:

**Control** (panel-hidden): provenance panel not shown
**Treatment** (panel-shown): full provenance disclosure inline

Panel funnel

| Stage | Metric |
|---|---|
| Panel View → Panel Expand | Tracks whether users engage with the disclosure |
| Customize Start → Complete | Tracks whether transparency drives customization |
| Checkout Start → Payment | Tracks conversion attribution |

Guardrails

A mid-test guardrail auto-pauses the experiment if the treatment variant shows >5pp conversion drop at any funnel stage. Slack alerts fire immediately for founder review.

Studio funnel

The Studio funnel tracks: AVM Browser View → Composition Attempt → Live Demo Run → Sign-in CTA Click → Paid Signup. This gives us the full picture of how the sandbox experience drives revenue.

---

Distribution channels

The engine surfaces ship with a full distribution stack:

**5 RSS feeds**: `/engine.rss`, `/roadmap.rss`, `/changelog.rss`, `/blog.rss`, `/schemas.rss`
**Newsletter**: weekly digest with top shipped items, gated behind founder approval
**Social share buttons**: X, LinkedIn, copy-link on every blog post and changelog entry
**Embed widgets**: badge SVGs + catalog iframe for partner sites
**Partner pages**: `/partners/microsoft-fasttrack` with ready-to-copy embed snippets

---

What's next

The i18n scaffolding is in place (en-US + de-DE + es-ES + ja-JP). Translation drift detection runs in CI. The locale-switcher appears in the footer when the feature flag is enabled.

The quarterly retro will cover:

A/B test results with statistical significance
Studio funnel drop-off analysis
Lab citation count and external references
Partner embed adoption
i18n translation coverage

---

Try it yourself

[How FrootAI Works](/engine) — the full engine tour
[FrootAI Studio](/studio) — try the AVM catalog + harvest demo
[FrootAI Lab](/lab) — reproducible benchmarks + public datasets
[What's New](/whats-new) — latest shipped capabilities
[Changelog](/changelog) — every shipped tag

---

*This post covers the FrootAI Website Engine launch. Engineering artifacts are maintained internally.*