Collaboration
How humans and AI work together in Catalyst projects.
Read this when starting a project, onboarding team members, or clarifying who does what.
Useful for solo builders and teams working with AI.
Why This Matters
AI can build fast, but it can’t decide what to build. The best results come when humans and AI play to their strengths — humans set direction and make decisions, AI handles execution and brings evidence.
Without clear responsibilities, projects drift. It’s easy for AI to build the wrong thing, so this page gives you a simple framework for collaborating together — whether you’re a solo builder or part of a larger team.
Who Owns What
Make responsibilities explicit. This prevents confusion and makes feedback clearer.
Humans Own
- Intent — What we're building and why
- Constraints — What must not change
- Quality bar — What 'good' looks like for this stage
- Decisions — Trade-offs, priorities, approval
AI Owns
- Execution — Code, refactors, scaffolding
- Exploration — Options, alternatives, edge cases
- Evidence — Tests, verification, screenshots
- Packaging — Summaries, PR descriptions, docs
When something goes wrong, this split helps diagnose it. UX not right? That's an evidence gap. Wrong feature entirely? That's an intent gap.
Working Modes
Different tasks need different mindsets. Even solo builders switch between these modes.
Directing
Set the target. Define what success looks like, set constraints, and make scope decisions. This is human territory.
Building
Produce the work. Write code, create content, implement features. AI excels here with human guardrails.
Verifying
Prove it works. Run tests, check quality, validate against requirements. AI provides evidence, humans judge.
Shipping
Get it live. Review, approve, deploy. Humans own the final call; AI helps with safe rollout.
Solo builders: You wear all four modes. Making them explicit helps you give AI clearer instructions — and reduces the back-and-forth when expectations are misaligned.
Across the Loop
Each phase of the delivery loop has a different collaboration pattern.
| Phase | Who Leads | What Happens |
|---|---|---|
| Brief | Human-led | You define what to build. AI flags gaps and suggests acceptance criteria. |
| Build | AI-led | AI implements against the brief. You provide guardrails and course-correct. |
| Review | Human-led | You judge if work meets the bar. AI summarises changes and highlights risks. |
| Refine | AI-led | AI applies feedback and re-verifies. You batch comments and clarify priorities. |
Across Stages
Collaboration rules change as projects mature. More rigour, more gates.
| Stage | Focus | Callouts |
|---|---|---|
| POC | Speed — validate the idea fast | Manual checks are fine. Focus on the critical path only. |
| MVP | Balance — real users, real feedback | Add repeatable tests. Completion packets become mandatory. |
| MMP | Rigour — production-like quality | Gated merges, release notes, stronger UX bar. |
| PROD | Safety — auditable and recoverable | Strict permissions, audit trails, rollback plans. |
Next Steps
Now you understand how to collaborate, explore how the loop works in practice.