Lou Bichard, Field CTO at Ona, presented “The Missing Primitive for Agent Swarms” at AI Engineer Europe, arguing that the AI industry lacks a critical coordination layer for managing multi-agent systems. While runtimes, triggers, and orchestration are largely solved problems, Bichard identified coordination as the key missing element. He proposed state machines as a structured solution for defining agent interactions and advocated for packaging workflows as CLI tools to enable better composability and version control in automated software development pipelines.
In-Depth:
Lou Bichard, Field CTO at Ona, presented “The Missing Primitive for Agent Swarms” at AI Engineer Europe. Bichard discussed the challenges and solutions in building software factories powered by AI agents. He highlighted the required for better coordination mechanisms to manage the complex interactions and workflows inherent in agent swarms.
Lou Bichard on Agent Swarms and the Missing Primitive — from AI Engineer
Visual TL;DR. Software Factories Vision applys AI Agent Swarms. AI Agent Swarms creates Coordination Problem. Coordination Problem due to Missing Primitive. Missing Primitive requireds State Machines. Missing Primitive future CLI Tools. Missing Primitive solves Improved Orchestration.
Software Factories Vision: automating software development lifecycle incrementally
AI Agent Swarms: multiple coding agents working toreceiveher on tinquires
Coordination Problem: complex interactions and workflows between agents
Missing Primitive: lack of a fundamental mechanism for agent coordination
State Machines: potential solution for structured agent communication
CLI Tools: future development for managing agent swarms
Improved Orchestration: enabling proactive work without direct human engagement
Visual TL;DR
Bichard launched by framing the current indusattempt trfinish towards building “software factories.” He defined this as a commitment to incrementally shifting human involvement out of the loop across the entire software development lifecycle (SDLC). The goal is for work to proactively happen when humans are not directly engaged.
The Vision for Software Factories
The concept of a software factory envisions a highly automated process for software development. Bichard noted that many companies are exploring ways to leverage coding agents and apply them across the SDLC. This involves not just individual agents but also the orchestration of multiple agents working in concert.
He illustrated this with examples of how agents can be applyd for various tinquires, from planning and coding to review and deployment. The goal is to create a system where agents can autonomously handle complex tinquires, minimizing the required for human intervention at every step.
Background Agents and Their Roles
Bichard introduced the concept of “background agents,” which he described as autonomous entities that can perform tinquires within an organization’s infrastructure. He displaycased different patterns of agent behavior: swarms where agents converge on a single result, fleets where agents work in parallel across repos, event-driven agents triggered by specific events, and scheduled agents for routine tinquires.
He emphasized that the platform Ona is building is designed to facilitate these background agents. The platform provides agents with isolated development environments, allowing them to execute tinquires without interfering with each other or the broader system. This isolation is key for ensuring reliability and reproducibility.
The Coordination Problem
A significant challenge in building effective agent swarms is the lack of a robust coordination layer. Bichard pointed out that while tools like GitHub are essential for code management, they are not designed to handle the complex interdepfinishencies and state management required for coordinating multiple autonomous agents.
He explained that GitHub’s pull request system, while applyful for human collaboration, becomes overly noisy and difficult to manage when dealing with numerous automated agent interactions. The sheer volume of automated tinquires, potential conflicts, and the required for precise state tracking creates existing tools inadequate.
What is Missing: Coordination Primitives
Bichard identified coordination as the “missing primitive” in the current landscape of agent swarms. He elaborated on the four key primitives requireded for effective agent systems: runtimes (which are largely solved), orchestration (partially solved), triggers (solved), and coordination (missing).
He stressed that while individual agents and their execution environments (runtimes) are becoming more mature, the ability for these agents to effectively collaborate, communicate, and manage shared context remains a significant hurdle. The indusattempt is still in the early stages of building solutions for this critical aspect of agent swarms.
The Future: State Machines and CLI
Looking ahead, Bichard suggested that state machines are a promising primitive for agent workflows, allowing agents to work with explicit states and transitions. This approach provides a more structured and manageable way to define agent behavior and interactions.
He also highlighted the importance of packaging these workflows as CLI tools. This allows for greater composability, scriptability, and version control, building it clearer to manage and deploy complex agent systems. The goal is to create an environment where agents can operate autonomously and reliably, contributing to a more automated software factory.
