How we build
Senior people make the decisions. AI agents speed up the work around them. Every change takes the same reviewed, tested path to production.
AI accelerates the work. It never owns the decision.
We use AI agents wherever they make the work faster or more thorough. Responsibility stays with people: every decision that shapes the product is made and reviewed by a senior founder.
Who does what
The split is explicit on every project. Judgement, direction and accountability belong to people; agents take the work that benefits from speed and repetition.
Owned by people
- Strategy
- Taste and design direction
- Decisions and trade-offs
- Architecture
- Priorities and scope
- Risk assessment
- Final quality review
- Accountability to the client
Assisted by AI
- Research
- Analysis
- Implementation
- Repetitive work
- Tests
- Documentation
- Localisation
From first question to steady operation
Twelve stages, each with an owner. People lead the stages that need judgement; AI agents carry much of the work in the others, always under review.
- People with AI agents
Research
Users, market, competitors and existing systems are studied. Agents speed up the analysis; we draw the conclusions.
- Led by people
Product framing
We agree on the goal, the users, the constraints and what success means.
- Led by people
Strategy
What to build first, what to leave out and why. Decided by a founder and written down.
- Led by people
Architecture
Data model, boundaries, integrations and risks are defined before screens are built.
- Led by people
Design
Flows, interface and visual system are designed together with the architecture.
- People with AI agents
Implementation
Typed code in small, reviewed changes. Proven, well-supported technology over novelty.
- People with AI agents
AI agent execution
Agents take scoped tasks: scaffolding, repetitive code, tests, documentation. Nothing merges without review.
- People with AI agents
Quality assurance
Automated tests, accessibility and performance checks, and manual review on real devices.
- Led by people
Security
Least-privilege access, no production secrets in agent workflows, dependency and header checks.
- People with AI agents
Localisation
Locale-aware routing, translatable content, localised metadata and formatting, right-to-left layouts where needed.
- People with AI agents
Deployment
Reproducible builds, versioned releases, health checks and a tested rollback path.
- Led by people
Iteration
Measure, learn and improve in short cycles, with the same review discipline.
How we work with AI agents
Agents handle research, drafting, scaffolding, testing and checks. They work under explicit rules:
- Each agent gets a clearly scoped task and only the access that task needs.
- Agents have no access to production secrets or client credentials.
- Every change is reviewed by a person before it is merged or deployed.
- Client data is used only for the client’s project.
- When an agent is unsure, the decision goes back to a person.
Want to see how this would work on your project?
Send a brief. We will outline how we would approach it.