Thinking about LLM teams as distributed systems actually makes a lot of sense. Coordination, latency, and failure modes start to look very similar to traditional distributed architectures.

That&#x27;s all nice &amp; well but which protocol &amp; topology will deliver the most dollars from investors?

&gt; Next up, LLMs as actors &amp; processes in π-calculus.You jest, but agents are of course already useful and fairly formal primitives. Distinct from actors, agents can have things like goals&#x2F;strategies. There&#x27;s a whole body of research on multi-agent systems that already exists and is even implemented in some model-checkers. It&#x27;s surprising how little interest that creates in most LLM &#x2F; AI &#x2F; ML enthusiasts, who don&#x27;t seem that motivated to use the prior art to propose &#x2F; study &#x2F; implement topologies and interaction protocols for the new wave of &quot;agentic&quot;.

Abstractly? 100%. Realistically? Depends on how many trillions we can get from investors.

Next up, LLMs as actors &amp; processes in π-calculus.

Language Model Teams as Distrbuted Systems