← back
The Missing Primitive for Agent Swarms — Lou Bichard, Ona
Takeaway
Scaling coding agents requires explicit coordination and dependable execution environments in addition to the agents themselves.
Summary
- Bichard defines a software factory as progressively removing proactive human intervention from the software-development lifecycle.
- Swarm patterns fan one intent out to multiple agents and combine results, while fleets apply changes across repositories and event triggers launch work automatically.
- Ona's examples include scheduled or triggered CVE remediation, test-coverage improvements, and organization-wide enforcement using repeatable prompts and scripts.
- Runtime infrastructure and orchestration are described as relatively mature; coordination among agents remains difficult, and Ona favors VM-based development environments for stronger isolation and resource control.
agent-swarmsdevelopment-environmentssoftware-factories
Original description
Stripe called theirs Minions. RAMP called theirs Inspect. Both are internal infrastructure for running fleets of background agents, and both teams built it from scratch. Lou Bichard's argument is that this shouldn't keep happening. The talk breaks down what agent swarm infrastructure actually needs: a runtime (largely solved), orchestration and triggers (solved), and coordination, which is not. Coordination is the gap where agents pick up tasks from each other, pass messages, and verify they have cleared a stage of the development cycle before moving on. GitHub is a poor substitute: noisy, designed for humans, and not built for agents raising hundreds of parallel pull requests. Lou covers what a proper primitive looks like, shows how Owner ships VM level isolation for agent fleets today, and makes the case that the coordination layer probably needs to be a CLI gateway that any local coding agent can invoke to check its progress and proceed. Speaker info: https://x.com/loujaybee / loujaybee Timestamps: 0:00 Introduction and definition of a Software Factory 1:50 Agent swarm patterns: Swarms, Fleets, and Events 3:11 Real-world examples of internal agent infrastructure (Stripe, RAMP) 3:50 How Owner handles agent infrastructure and development environments 4:49 Understanding Harness Engineering 5:43 The three pillars of agent swarm infrastructure: Runtime, Orchestration, and Coordination 7:17 Demo: Running sub-agents and fleets in Owner 10:20 Challenges of building a software factory 11:44 The issue with Context Management and Context Rot 12:16 Why GitHub is a poor coordination layer for agents 12:59 Proposed solutions: State machines, Durable execution, and CLI gateways