One human sets the direction and the quality bar. AI agents do the building. mapelabs exists to prove that AI, used with clear standards and real accountability, can ship ambitious products that hold up in the real world.
The model
mapelabs is deliberately small: a single person runs it, and the code, tests, translations and documents are produced by AI agents working inside written standards.
Luís Miguel Massih Pereira decides what gets built and why, designs the architecture, sets the standards and reviews what the agents deliver. Nothing goes public without his sign-off, and he answers for everything that does.
Agents write the code, the tests, the infrastructure, the translations and the documentation. They work across the whole stack, often in parallel.
Quality isn’t a feeling. The rules are explicit: how code should read, what must be tested, what blocks a release, how users are treated.
Every change is verified before it goes live: type checks, automated tests, a local run and a look at the real thing in a browser.
Results are reported as they are, failures included. No inflated numbers, no claims the work can’t back up.
The standards
The same gates apply to everything, from a one-line fix to a whole new product.
Every task starts with a stated goal and its constraints. Ambiguity is resolved by the human, never guessed by an agent.
Agents follow each project’s conventions and limits. They match the surrounding code, and never spend money or publish anything on their own.
Type checks, tests and a real local run. Anything users will see is checked in a browser before anyone else sees it.
The human reviews, approves and ships. Accountability never moves to a machine.
Proof, not promises
A free space MMO that runs in the browser, set in the real universe: real stars and confirmed exoplanets at true scale, planetary landings, pirate hunts, corporations with wars and alliances, player trading and voice chat.
Built entirely with this model: game client, multiplayer server, payments, data-protection compliance, automated tests and a website in three languages. It’s live, and anyone can play it.
Play Estelar Online →Client case
A Brazilian live-commerce marketplace where sellers run live video auctions and “buy now” sales, with real-time bidding where latency and order matter.
Luís defined the architecture of the whole platform and owned quality assurance from the first commit. Execution was also done by AI agents working within his standards: the buying experience end to end (bidding, checkout and card payments, orders, cancellations and disputes) and the path to production on AWS, with fully automated releases. From the first commit to launch took less than three months.
Visit PQN Live →Lab product
A free desktop app that mirrors every terminal window on your computer, including Claude Code sessions, straight to your phone. Scan a QR code and keep working from anywhere on your Wi-Fi, with nothing going through a server.
Built by the lab with the same model, for people who work with agents. The first version shipped through Homebrew 17 days after the first commit, and 24 versions have shipped so far for macOS, Linux and Windows.
Get FloatCode →Contact
Have an ambitious idea, or want to talk about the model, the standards or the human behind mapelabs? Write in.
hello@mapelabs.io