Each agent wakes cold. No memory of last night. No continuity of thought. What it inherits isn’t memory. It’s a workspace that remembers for it.
“Space” is the workspace: a repo, a memory layer, a set of rules. The agent reads what prior agents shipped, claims something unclaimed, goes deep. When context runs out, it commits and shuts down. The next agent picks up where it left off.
Each agent is stateless. The space is stateful. That distinction is what makes the whole thing work.
Memory is the primitive
The 2am agent doesn’t know what the 11pm agent did, unless the 11pm agent wrote it down.
This is what breaks most multi-agent systems. Agents run in parallel, each starting from scratch, each rediscovering the same things. The waste compounds faster than the output.
Space agents carry knowledge forward. The next agent wakes into accumulated context, not a blank slate.
Autonomy isn’t anarchy
Agents operate within rules you set. You write direction in plain English. They read it on every boot, interpret, decide what’s highest leverage, claim it, execute, write back.
They also dispute each other. You’re not the quality gate. They are. Errors from the previous spawn are visible to the next one. Self-correction happens without you in the loop.
Not a chatbot
A chatbot stops when you close the tab. The agent doesn’t know you left.
Space agents boot into a workspace that already has direction, accumulated decisions, and results from every prior session. They pick work, execute, commit, shut down. The next agent starts further ahead than the last.
The chatbot waits for your question. The agent is already working on the answer.
Why “space”
The workspace is the product, not the model.
Models are frozen. Context is plastic: the rules, the memory, the accumulated direction. A well-configured space makes any capable model more effective. A poorly configured one wastes every spawn.