Appaya is a software studio with three lines: its own products, fixed jobs and contracting. Luke Czak directs a fleet of AI agents, and an independent review challenges what they build before it reaches you.
One inspects a system you already have, one builds a defined outcome, one decides what to build. Each has a written scope, and a statement of work comes before any invoice.
01 Inspects
Inspects a system you already have.
02 Builds
Builds one defined outcome.
03 Decides
Decides what to build. It does not inspect your code.
Agents repeat, fast. A person decides what ships. The engine that writes a change never reviews it.
Every job starts as one written requirement. The acceptance criteria are set before any code is written.
The work is split into bounded tasks. Each runs in its own lane, on its own isolated branch, side by side.
Several AI coding engines work in parallel, one task each. The judgement about what to build and what to accept stays human, with one accountable person.
The engine that writes a change never reviews it: a different engine, from a different vendor, attacks every change looking for failure modes.
Work that fails the review goes back to the start of its lane, still open. It does not merge; it is built again.
A change merges only after the independent review passes and a repeatable test checks the behaviour.
Appaya builds its own software with the same fleet and the same review its fixed jobs use. FitKind and Demix are internal builds, not client work.
Own product
A workout tracker built natively for Apple and Google, with offline logging that syncs once you are back online. Personal workout tracking: not social, and not AI coaching.
Own product
A command centre for AI agent sessions: a macOS workspace, iPhone and Apple Watch apps with widgets, and a shared daemon. It tracks Claude, Codex and other agent sessions across Mac and mobile, so you can approve, answer, interrupt or resume work from the right device.
Appaya's own internal tooling: what runs its builds and its review gate. It is shown here as part of how the work is done, not as something offered.
Internal tool
The control plane for AI coding tools. It keeps accounts, usage limits, workspaces and sessions for Claude Code, Codex and other CLIs in one place, with isolated profiles.
Internal tool
Visual sign-off and independent review. It screenshots a running app and grades each acceptance criterion; its review broker dispatches the reviewer and signs the receipt, so the author produces only the patch.
Internal tool
A local supervisor around the build and review process. It assigns, observes, feeds and verifies work and proposes merges; it never writes product code and never merges.
Internal tool
The project knowledge layer. It owns vision, architecture, roadmap, tasks, specs and decisions for every project, and keeps them in sync across agents and machines.
Internal tool
Lane naming and lifecycle for supervised agent workers across machines. Who owns which worker is derived from live state, never read from a stored record.
Internal tool
Simulator and emulator allocation for several agents at once. Each agent claims a session and works through one command, so agents stop booting and killing each other's devices.
Contract Luke Czak through Appaya Ltd: ongoing senior technical leadership, part-time, working with your existing team.
Luke's past employers, not Appaya clients
HSBCZopaSmarketsShieldpayIrysNot sure which job fits? Start with a fit check. One enquiry path covers all three jobs and contracting, and the fit check comes before any scope or invoice.
Or write to [email protected]. Hai is hi, plus AI.