Collaboration

Workflow~5 min

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

  • IntentWhat we're building and why
  • ConstraintsWhat must not change
  • Quality barWhat 'good' looks like for this stage
  • DecisionsTrade-offs, priorities, approval

AI Owns

  • ExecutionCode, refactors, scaffolding
  • ExplorationOptions, alternatives, edge cases
  • EvidenceTests, verification, screenshots
  • PackagingSummaries, 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.

PhaseWho LeadsWhat Happens
BriefHuman-ledYou define what to build. AI flags gaps and suggests acceptance criteria.
BuildAI-ledAI implements against the brief. You provide guardrails and course-correct.
ReviewHuman-ledYou judge if work meets the bar. AI summarises changes and highlights risks.
RefineAI-ledAI applies feedback and re-verifies. You batch comments and clarify priorities.
Learn more about the Delivery Loop

Across Stages

Collaboration rules change as projects mature. More rigour, more gates.

StageFocusCallouts
POCSpeed — validate the idea fastManual checks are fine. Focus on the critical path only.
MVPBalance — real users, real feedbackAdd repeatable tests. Completion packets become mandatory.
MMPRigour — production-like qualityGated merges, release notes, stronger UX bar.
PRODSafety — auditable and recoverableStrict permissions, audit trails, rollback plans.

Next Steps

Now you understand how to collaborate, explore how the loop works in practice.