Cosine's 'swarm' — an orchestrator delegating to sub-planners who delegate to up to 253 worker sub-agents — can one-shot a complex, novel project that no single frontier model can do out of the box.
Alistair describes Cosine's 'swarm' system — one orchestrator breaking work into sub-planner verticals that delegate to a flat layer of workers (253 in one case) — successfully one-shotting a from-scratch mechanical-watch SDK project that base models like Opus or Gemini can't do alone. ✦ AI generated
Alistair Pullen · Machine Learning Street Talk · 2026-07-13 · original ↗
starts at this moment · 42:10
“I mean how do you see that evolving over time because now it's quite a manual process but do you think well I mean you've got this swarm thing maybe.”
as soon as you put them in a swarm and for what swarm looks like for cosign is you have one orchestrator at the very top it breaks down a problem into um sub problems for uh basically product managers or or or whatever you want to call them. We call them subplanners but they own verticals of this.
verbatim transcript · starts at 42:10
41:52able to like specify mechanical watches in code because I don't know herology, but I want to be able to do it anyway. Um, I want it to be physically congruent. I want you to use some kind of physics engine. Um, and then I also want like a 3D viewer to be able to like see the thing running. Um, and it all needs to be possible in real life. You
42:10can't have things like intersecting each other wouldn't be possible and so on. Um, and that is something that out of the box Kimmy cannot do. There's no way, not even close. Like it would be terrible. Um, in fact, even Gemini 3.5 and Opus and 5.5 can't really do it. But as soon as you put them in a swarm and for what swarm looks like for cosign is
42:29you have one orchestrator at the very top it breaks down a problem into um sub problems for uh basically product managers or or or whatever you want to call them. We call them subplanners but they own verticals of this. So like within that task you would have had um a subplanner to do the SDK. You would have had a subplanner to do the 3D viewer. had a subplanner to write the
42:53documentation and so on. Um, and then those subplanners could then delegate to workers and then they have like a flat layer of of as many workers as they like. Um, and for that problem we use 253 sub aents which I think is more than you tend to see in a claude code session and so on. I think I think you you'd probably hit your usage limit pretty
43:12quickly that way. Um, but when you do that it is possible and you can do that entire project in one shot. Um and and and I am contradicting myself quite badly because I've just said harnesses don't matter but in that respect they do obviously matter. >> Oh indeed very excited about that. But you know um it raises the question that when first of all when when you start to
43:32have loads and loads of agents you have more understanding debt and less interactivity because you know for me the the the lack of interactivity is part and parcel of the understanding debt. And and sometimes you you want to interject you want to say oh you know you've gone slightly wrong there. I want to change what this agent's doing. And what many folks have found when they build these agent systems is that the
43:50agents interfere with each other. They kind of overwrite its own. They go into deadlock. How how are you dealing with all that? >> So it's a hard problem and we experience all those things. Um in terms of you can it so one of the key things that we did is we made you we gave the ability to interject to like an agent on the lowest level. So say you had a