ENGINEERING STUDIO · TAHITI
Software and agents built for real work. Think data contracts, business tools, tests and rollback. We like problems that need more than a copy-paste.
// made in Tahiti. maintained in reality.
stack.ts
// understand the problem, then pick the tools.
App Router, Server Components, static rendering when it fits.
Explicit contracts, readable components, deliberate state.
Query, Table, Router: the right tools for your data.
Tokens and interfaces versioned with the code.
iOS and Android, with a shared codebase.
Field data capture, local storage, sync when the connection returns.
Business APIs, asynchronous jobs and integrations.
When the project and its ecosystem call for it.
Data modelling, migrations and queries.
OpenAI, Anthropic, Google, Mistral, open source: choose for the task.
Business tools exposed with precise permissions.
Context, execution, traces and evaluation in one loop.
Use-case audits, architecture and team training.
Preview environments and automated deployments.
Tests, logs, errors and rollback paths.
Business connectors that keep data moving.
devlab.shell
devlab --mode nerd --interactive
studio: Devlab / engineering
location: Tahiti · Pacific/Tahiti
team: 14 humans
focus: software · agents · integrations
shipping_since: 2014
Type help. The cursor is yours.
pipeline.yml
From first commit to field feedback. Short steps, explicit decisions and software that evolves with its users.
./shipBreak down the problem. Build a usable version. Put it in the hands of the people who will use it.
./loopObserve usage, read feedback, fix things. The next commit addresses an identified need.
./aiGeneration, review, tests: agents speed up the work. Architecture and approval remain human decisions.
./prodTypes, permissions, critical paths and observability. Prepare the rollback before you need it.
feedback → backlog → commit → repeat
agents.config.ts
Useful inputs, scoped tools, verifiable outputs. The reference architecture follows the business task through to its outcome, with a path to a human.
Map the code, select useful documents and keep the task state. Every token should support a decision.
Define available tools, permitted data and actions that require human approval.
Keep decisions and completed steps so a task can resume with its context, without starting over.
Test normal, ambiguous and degraded cases. Log actions and surface what the agent cannot resolve.
principles.diff
A proven dependency can be a very good architecture decision.
Readable code, documented choices and the ability to evolve your tools.
Verify a feature in the workflow that matters to its user.
Understandable changes, fast feedback, focused fixes.
AI accelerates execution. Technical judgement remains a human responsibility.
Ambitious software, built and maintained here. The timezone is part of the architecture.
TODO: solve your next problem.
05 / next-commit.sh
30 minutes with someone who codes. Bring your architecture questions, stubborn integrations or agent idea. The terminal doesn’t bite. Neither do we.
Book a slot