Skip to content

How we work

AI builds faster. People stay accountable.

AI makes building faster. We put the time it saves into judgement: specs, review, security, and owning the result. One team across Sydney and Dhaka, with one accountable lead on every project.

Discuss Your Project

Our method

Human-led, agent-built.

AI makes the inner loop of writing, testing and refining code fast and cheap. The outer loop, where decisions, review and accountability live, stays with people.

Human-led, agent-built delivery processFrame, Architect and Plan are human-owned and lead into an AI-accelerated inner build loop of write, test and refine. Every pull request passes a human review gate into Verify, then Release with client sign-off, and learnings feed back into the standards.OUTER LOOP · HUMAN-OWNED01FrameAgreed spec02ArchitectStandards in repo03PlanSmall reviewable slicesWriteTestRefine04BuildAI agents · inner loop05VerifyEvery PR reviewed06ReleaseDemo, sign-off, monitor
Human gateInner loop · AI-acceleratedLearnings feed back into standards

Inner loop

Where AI speeds us up

  • Writing code and tests
  • Refactoring and clean-up
  • Trying options with quick prototypes
  • Drafting task plans

Outer loop

What stays with people

  • The spec and acceptance criteria
  • Architecture and security decisions
  • Reviewing every pull request
  • Releases, sign-off and accountability

Stage by stage

Who owns each step, and what you see.

You approve the spec before we build and sign off each release. In between, every change passes human review.

  1. Frame

    Human-owned

    We work through goals, users and constraints with you, then write a spec with testable acceptance criteria (“when this happens, the system should do that”), what is out of scope, and the risks.

    What you see

    A spec you approve. Nothing is built before it is agreed.

  2. Architect

    Human-owned · AI-assisted research

    A senior engineer makes the architecture and security decisions and records them. Project standards (conventions, testing and security rules) are written into the repository as instructions our AI agents must follow.

    What you see

    Decision records and the standards file in your repository.

  3. Plan

    AI drafts · human approves

    The spec is broken into small tasks that can each be reviewed on their own. An engineer approves the plan before any code is written. Small batches keep changes easy to review and safe to release.

    What you see

    The task plan and a board you can follow.

  4. Build

    AI-accelerated · engineer-directed

    Engineers direct AI coding agents (Claude Code, OpenAI Codex, Cursor, GitHub Copilot and similar tools) to write tests and code, iterating against automated checks: tests, type checks, linting and security scanning. Prototypes are cheap, so we try options before committing to one.

    What you see

    Working increments, early.

  5. Verify

    Human-owned gate

    Automated checks and an AI review pass run first. Then an engineer other than the author reviews every pull request before it merges. Agents never approve or merge their own work, and sensitive areas such as authentication, payments and personal data get extra review.

    What you see

    A reviewed pull request history in your repositories.

  6. Release

    Human-owned

    Small, reversible releases. We demo against the acceptance criteria from the spec, you sign off, and we monitor after release. Every change has a named person accountable for it, and what we learn goes back into the specs and agent standards.

    What you see

    The weekly demo, release notes and your sign-off.

Guardrails

What never changes.

  • Every pull request reviewed by an engineer
  • Agents never merge their own work
  • Standards written into your repository
  • Code in your accounts
  • One accountable lead, a demo every week

Why the outer loop matters

AI output still needs judgement.

Industry research points the same way: AI raises speed, and without strong review it raises risk too.

So we use AI where it is strong and keep people where judgement matters.

Your week with us

A rhythm you can plan around.

  1. 01

    Agree priorities

    At the start of each cycle we agree what matters most with you.

  2. 02

    Build in small slices

    Reviewable pull requests land throughout the week, not in one drop.

  3. 03

    Written update

    A short weekly note: what shipped, what is next, and any decisions we need.

  4. 04

    Demo

    A weekly demo with your accountable lead, against the agreed acceptance criteria.

Ways to work with us

Four ways to work together.

These are independent choices, not steps. Start with any of them; the method above applies to all four.

  • Discovery sprint

    Turning an idea or a problem into a clear plan before committing to a build.

    A short, fixed-scope phase of workshops, research and technical planning with your team.

    You get

    • Agreed scope and priorities
    • Prototype or technical plan
    • Estimate for the next step
  • MVP delivery

    Getting a first release in front of real users.

    A defined first release, designed, built and launched against an agreed scope.

    You get

    • Designed and tested product
    • Production deployment
    • Code and IP in your accounts
  • Ongoing product team

    Products that need steady improvement after launch.

    A dedicated Corexlab team working through your roadmap, with priorities agreed each cycle.

    You get

    • Regular releases
    • Maintenance and support
    • Weekly updates and demos
  • Team extension

    Engineering leaders who need more senior capacity, on a new product or an existing platform.

    Our engineers join your team, working in your repos, rituals and review process.

    You get

    • Senior engineers in your team
    • Your tools and standards
    • Knowledge that stays with you

Questions

AI and your project.

Do you use AI on our project?

Yes, mainly in the build stage, where AI coding agents write and test code under an engineer’s direction. The spec, architecture, review and release stay with people.

Who reviews AI-written code?

An engineer other than the author reviews every pull request before it merges, after automated tests, type checks, linting and security scanning have passed. Agents never approve or merge their own work.

Does AI make projects cheaper?

It shortens the build work, and we put much of that time back into specs, review and quality. We don’t promise a discount; a scoping call turns your scope into a real estimate.

Can you work within our tools and review process?

Yes. We can join your stand-ups and tools, work in your repositories and follow your review rules. You still get a weekly update and demo.

Want to see how this would work on your product?

Tell us where things stand. We'll walk you through how the method would apply, who would lead it, and a sensible first step.

Book a call (opens Calendly in a new tab)