Coordination SystemsAbout

Intelligence is a coordination system.

Coordination Systems is an AI research lab working on a single thesis: intelligence is produced by coordination among people, models, programs and memories. General intelligence, on that account, is the coordination intelligence between specialised models competing under evolutionary pressure. We work on the arena rather than the athlete: the conditions under which specialists compete, are scored and amplify one another. We call the layer where those parts act on one another's behalf the Agentic Internet, and our work studies and builds the protocols, runtimes and local computers that make those relationships explicit.

The thesis unfolds through three research programs. Cultural Intelligence asks how a group turns signals into judgments about what deserves attention and trust. Compositional Intelligence asks how those judgments can guide the runtime arrangement of specialist models, programs, tools and memories. Open Epistemic Intelligence asks how the resulting learning can travel without stripping away its author, context or local authority. Cultural judgment supplies evaluation, composition turns evaluation into working structures, and the open epistemic layer lets those structures compound across a network.

Three software layers make that sequence operational. gOS keeps files, memory, tools and human judgment within an application boundary on a computer its user owns. AGInetwork is the runtime in which those machines exchange learning events and compose expert components, and it is what settles the exchange: it routes a job to the nodes that can do it, prices the result, and pays the contributors out of a staked pool. Geranium is the research layer designed to learn from evaluated traces of those compositions. Nothing in that loop asks participants to share a canonical state. It asks for enough evidence about origin, trust and consequence that distinct participants can build on one another.

The thesis is tested inside businesses. We deploy with companies where coordination changes what they can produce, build their agentic workflows alongside their teams, and hand the workflows over as those teams take them on. Deployment is how the research earns its evidence: paid work under a deadline supplies the evaluated traces and the situated judgment that the three programs need and that no benchmark contains. What survives contact with a client is what the network keeps.