Your idea
Raw Input
Any format, any state
- Voice memo transcript
- Slack thread or email chain
- Meeting notes or bullets
- Napkin sketch description
Human Automation Labs · 2026
AI coding agents didn't
fix the funnel.
They exposed it.
The bottleneck was never the builder. It was always the intake. We fix the intake — turning your raw idea into a build-ready spec that any AI coding agent can execute without guessing, without interpretation, without losing context.
Before — Broken intake
Product definition
Vague PRD. Stakeholders argue. Scope creeps.
Engineering scoping
T-shirt estimates. Questions the PRD can't answer.
Build
"That's not what I meant."
Before first user feedback
8–16 weeks
After — Fixed intake
Chancery — AI CPO interrogation
Kill or proceed. 29-check PRD. Compliance detected.
DesignerSDD — Spec compilation
65 quality checks. Build-ready package. Zero ambiguity.
Handoff
Your team. Your tools. Your AI builder. Ready to execute.
Before first user feedback
Hours, not months
The funnel has been broken for years. AI made it worse.
Every team felt it. Nobody named it. The problem was never that builders were slow — it was that the intake was broken before anything reached the builder.
01
PRDs are written for humans who can ask clarifying questions.
"The system should be user-friendly" worked when the engineer could walk over and ask what that meant. It doesn't work with distributed teams, outsourced development, or AI coding agents that build exactly what the spec says.
02
Nobody challenges the product definition.
The PM writes the PRD. The PM approves the PRD. There's no CPO in the room pushing back on weak problem statements, unvalidated assumptions, or missing compliance requirements.
03
The spec and the build are permanently disconnected.
Even a good PRD loses context at the handoff. NFRs are adjectives — "fast," "secure," "scalable" — instead of numbers. Acceptance criteria that can't be tested aren't criteria.
04 — The shift
AI coding agents exposed the intake problem completely.
Claude Code, Cursor, Codex — they build fast. But they build exactly what the spec says. If the spec is vague, they build something vague. A bad spec used to produce a wrong prototype in 8 weeks. Now it produces one in 8 hours. The faster the builder, the more damage a broken intake does.
Two stages. One output. Any builder.
Your raw idea goes in. A build-ready specification package comes out. What you build it with — Claude Code, Cursor, your own team — is your choice. That's the point.
Stage 1 — 2–4 hours
Chancery
AI Chief Product Officer
- Kill or proceed — Phase 0
- Live competitive scan
- Domain-adaptive interrogation
- Compliance detection, 10 frameworks
- 13-section PRD, 29 quality checks
- Score ≥70 required to advance
Stage 2 — 30 minutes
DesignerSDD
Spec Compiler
- 28 quality + 8 buildability checks
- Auto-improve loop until both pass
- ≥80% acceptance criteria verifiable
- 7-document output package
- COMPLIANCE.md if applicable
The handoff is the product. You receive CONSTITUTION.md, SPEC.md, PLAN.md, TASKS.md, HANDOFF.md, README.md, and COMPLIANCE.md if applicable — a complete build package that any AI coding agent or dev team can execute without clarifying questions. What you build it with is entirely up to you. If you want us to execute it too, that's the Prototype tier.
Seven documents. Zero ambiguity.
Not a summary. Not a slide deck. A structured build package with a decision audit trail, a task backlog, and compliance docs if your domain requires it.
CONSTITUTION.md
Values, constraints, and hard limits for the build. What the agent is and isn't allowed to do.
SPEC.md
Full product requirements — functional, non-functional, scope. Every NFR is a number, not an adjective.
PLAN.md
Technical approach, stack decisions, architecture decisions with rationale.
TASKS.md
Phased task list with dependencies, acceptance criteria, and "when to stop and ask" rules. Your dev backlog.
HANDOFF.md
400+ lines of build instructions. Every requirement in FR-N format. Zero interpretation required.
COMPLIANCE.md
Generated if your domain triggers any of 10 frameworks. HIPAA, PCI-DSS, GDPR, SOC 2 — designed in from day one.
We'll tell you not to build it.
Most service providers take every engagement because revenue. Phase 0 assesses four signals before the spec work starts. When all four are weak, we recommend killing the idea — with evidence, not opinion.
01
Competitive landscape. Three YC companies already shipping the identical feature is a weak signal, not a market validation.
02
Signal strength. One enterprise customer asking at dinner is a conversation. Not evidence.
03
Timing. "Someday" with no regulatory deadline, competitor launch, or cohort event is not a market window.
04
Effort vs. impact. A quarter of engineering runway for marginal value is a kill, not a sprint.
Kill decision pricing: Tier 1 pays $500, not $5,000. Tiers 2–4 pay $1,000 for Phase 0 + competitive findings. Financially painless for you, reputation-building for us.
"We took a $7,500 engagement from a Series A startup. After 2 hours of discovery, Chancery found that three YC companies had already shipped the exact same feature, the client's only evidence was one enterprise customer asking at dinner, and there was no timing signal — no regulatory deadline, no competitor launch, no cohort event.
We told them: this isn't worth building right now. Here's what we found. Here's what would change our mind. They paid $500 for the Phase 0 assessment and we parted ways."Six months later they came back with a different idea — one that passed all four signals. We built a prototype in 48 hours they used to close their Series B.
Four tiers. One pipeline.
Every tier runs the full Chancery → DesignerSDD quality chain. The core product — the spec — is the same at every level. The tiers differ in what happens after the handoff.
Validate
24-hour delivery
$5,000
Your idea, interrogated and compiled into a build-ready spec. Take it to any builder — AI or human.
- Phase 0 kill or proceed assessment
- 13-section PRD, scored & quality-gated
- Full SDD package — 7 documents
- Compliance requirements if applicable
- dashboard.html interactive spec viewer
Prototype
48-hour delivery
$7,500
Validate plus we execute the build. Working prototype deployed to staging, ready to demo.
- Everything in Validate
- Working prototype on staging
- One revision round
- CI/CD pipeline configuration
- Test suite from acceptance criteria
Launch
5 business days
$15,000
Production deployment, compliance docs, three revision rounds, performance testing.
- Everything in Prototype
- Production deployment
- Three revision rounds
- Performance testing vs. NFR thresholds
- Security hardening + monitoring
Governed
10 business days
$25K+
Federal, healthcare, fintech, and AI companies where governance isn't optional.
- Everything in Launch
- Castellan governance spec
- SENESCHAL runtime enforcement
- Aegis audit report
- Vigil continuous monitoring
Kill decision pricing: If Phase 0 recommends against building and you agree — Tier 1 pays $500 only. Tiers 2–4 pay $1,000 for Phase 0 + competitive research. Full price applies if you proceed against the recommendation.
The pipeline that validates your spec is itself validated.
Numbers, not adjectives.
Chancery v0.2
AI Chief Product Officer
| Language | Python 3.11+ |
|---|---|
| LLM backend | Anthropic Claude API |
| Test suite | 211 tests (pytest) |
| State enforcement | GateNotApprovedError — code, not policy |
| Compliance detection | 10 frameworks |
| PRD checks | 29 (20 deterministic + 9 semantic) |
| Scoring gate | ≥70 required to compile |
DesignerSDD
Spec Compiler
| Language | Python 3.10+ |
|---|---|
| Test suite | 720 tests (pytest) |
| Quality checks | 28 weighted + 8 buildability |
| Buildability gate | ≥80% verifiable acceptance criteria |
| Auto-improve loop | Iterates until both gates pass |
| Export artifacts | 7 core + conditional COMPLIANCE.md |
| Context compiler | Charlotte — 103K tokens |
Send us the idea.
We'll fix the intake.
Voice memo, Slack thread, bullet points — whatever form the idea arrived in. You'll have a build-ready spec in 24 hours that any AI coding agent can execute without guessing.
Tim Wolfe · GitHub · Telegram · rtwolfe@gmail.com