Workspace
Where you build.
The AI-native build environment, editor, tooling, and assistant in one place.
The research goal: can one platform let a person with an idea and no engineering department ship software at the standard a large one would? Big organizations get quality from headcount, review layers, and internal libraries. This project tests whether an AI build environment, a vetted module library, and enforced quality gates can do the same job for one person.
It is the largest project in the lab and the one the others feed. The orchestration engine runs the AI that builds inside it. The knowledge graph tells a builder what already exists so nothing gets rebuilt.
The platform is organized as rooms, each one a working surface: a build environment with an integrated editor and AI assistant, a module library where vetted work is reused instead of rewritten, a research library connected to arXiv, Semantic Scholar, OpenAlex, and PubMed, a task system with typed work contracts, a collaboration space, and a five gate quality pipeline covering build, test, security, compliance, and human review.
Payment is the other half of the experiment. A bounty economy lets a founder with no cash offer shares instead, with contracts that convert to real equity on reviewer sign off. The lab runs its own development through this system, so the first user is us.
Built so far: the working prototype with a 161 table schema, around 100 API routes, the task and share contract system running end to end, GitHub ingestion, and the research library connectors.
The surfaces
Where you build.
The AI-native build environment, editor, tooling, and assistant in one place.
Where it is verified.
Two-stage quality and security gate, mapped against recognised standards.
Where it grows.
Project incubation and the module library everything is assembled from.
Where it gets paid.
Cash or share-backed contracts, vesting on reviewer sign-off.
What the lab built
At a glance
A goal argued down into buildable parts
The AI build environment
Five quality gates, then human review
Vetted work enters the module library
The next builder starts from it instead of from zero