Skip to content
Zaman Javed.

02 / Process

AI makes code cheap. Process makes it correct.

This is the differentiator: a five-stage method where every AI-assisted session starts from documents and ends in a reviewed, tested pull request. Scroll through it the way an engagement actually runs.

Documents before code
ambiguity is killed where it's cheap
Same gate for all code
AI output passes human CI, always
Context is curated
scoped sessions, never vibe coding
  1. 01 / define

    Discovery & PRD

    Every engagement starts with questions, not code. Stakeholder interviews, competitive context, and hard constraints become a PRD that names the audiences, the success metrics, and — critically — the non-goals. Ambiguity dies here, where it's cheap.

    • PRD document
    • Success metrics
    • Non-goals list
    • Scope & phasing

    01definefig. 1 / 5

  2. 02 / constrain

    Rules & standards

    Before implementation, the constitution: coding standards, conventions, quality bars, and taste — written as rules every human and AI session must obey. Review stops being debate and becomes diff-checking.

    • Rules file
    • Strict TS & lint config
    • Testing thresholds
    • A11y & perf budgets

    02constrainfig. 2 / 5

  3. 03 / design

    Architecture & docs

    System boundaries, data flow, and key decisions documented before they're expensive to change. Short, opinionated documents sized to fit a context window — one shared mental model for everyone (and everything) building.

    • Architecture doc
    • Decision records
    • Data model
    • Integration contracts

    03designfig. 3 / 5

  4. 04 / build

    Context-managed AI development

    The fast part — because everything above exists. Features are built in scoped sessions with Claude Code and Cursor, each loaded with exactly the right documents: the PRD section, the relevant architecture doc, the rules. Every session ends as a reviewed, typed, tested pull request.

    • Scoped sessions
    • Curated context
    • PR-sized changes
    • Human review gate

    04buildfig. 4 / 5

  5. 05 / verify

    Testing & shipping

    AI-accelerated code passes the same gate human code does: CI enforcing types, lint, and tests; staged rollouts; monitoring. The process buys speed at the specification end — never by loosening the quality end.

    • CI pipeline
    • Staged rollout
    • Monitoring
    • Handover docs

    05verifyfig. 5 / 5

Have a project in mind? Let’s make it inevitable.

The case studies show this process against real deadlines — or book a call and we'll walk through how it maps onto your project.