Skip to content
Just IoT
Menu
How we build

AI writes the software. Evidence decides if it ships.

AI coding agents are fast, confident and sometimes wrong. Our method keeps the speed and removes the guesswork: work is split across specialised agents, and nothing moves forward without evidence a person can check. It is how Chimes is built, every day.

The evidence gates

Red before green

A test for a bug fix must be shown to fail on the code before the fix. A test that passes either way is not evidence, and is reported as such.

Cold review

The reviewer must not be the author. Review runs in a fresh context, and every finding is checked against the code before it is fixed or declined.

Real devices

Simulators first, then real phones for what simulators cannot show: GPS on the move, background behaviour, push delivery and sound.

Screenshots, not step logs

A device test is not a pass until its screenshots are read. A test runner records what it sent, not what the app did.

Rules from incidents

Every operating rule is written down with the incident that caused it, so the method learns from its own mistakes.

Specialised roles

Each role is a separate agent with its own instructions and its own written output.

  1. 01Product spec. Turns an idea into user stories and acceptance criteria.
  2. 02UX flow. States, flows and accessibility before any code.
  3. 03Architecture and data. Maps the change onto existing services and failure modes.
  4. 04Delivery planning. Breaks the work into small slices, each with its first tests.
  5. 05Implementation. Builds slice by slice, test-first where it can.
  6. 06Developer tests. Unit and integration tests against the acceptance criteria.
  7. 07Challenger. Argues against the spec, the design and the conclusions.
  8. 08Code review. Reviews the diff cold: never the agent that wrote it.
  9. 09QA cases and automation. Human-approved test cases, then device automation.
  10. 10Release readiness. Docs, changelog and operational risk before merge.

Rigour that fits the risk

Not every change needs every role. Four levels, chosen by risk; anything touching sign-in, permissions, data or payments moves up automatically.

Lite
Small, well-understood changes: one review.
Standard
A written spec and one review.
Thorough
Spec, a design challenge and a code review.
Full audit
Every role, with human approval at each gate. For security, data and payment changes.

Want this for your project?

We take on selected builds of location and IoT apps, and evidence-based reviews of AI-written code.

Talk to us