From AI to Agents to Agencies
blog.nishantsoni.comThe idea of swarming multiple agents on a task isn't new. And honestly we haven't seen it really work in practice. We've tested multi-agent systems a bunch with OpenHands [1] and have never really seen a bump on benchmark scores despite the massive increase in complexity. There's nothing that many different agents can do that a single generalist can't accomplish on its own.
That said, they can potentially get you a speedup if you have a neatly separable task, and can parallelize the work. But it doesn't lead to some quantum leap in what agents are able to accomplish unsupervised.
I do think some form of multi-agent workflow is going to become important over the next few years, but more because it fits our mental model of the world rather than being some big technological unlock.
Same observations we had talking to many of our users. The trick to build reliable systems, is minimize the complexity to the max, not the other way around.
I think the theoretical value of multi-agents is collaboration with external agents (outside your code base). Other than that, there is a very little use cases where it make sense (e.g. https://www.anthropic.com/engineering/built-multi-agent-rese... ), and building it / debugging them take much much longer and is much harder. So unless you have the ressources, not worth the trouble
This isn't multi-agents at all. Infact if you read the article in detail, you will realize that the author goes in detail to explain how this system is different from multi-agents. And this is exactly why the author calls it "Agency" because it is fundamentally different from multi-agents.
I agree that multi-agent doesnt work in practice. But this isnt that.
How is this different from multiple agents? Are you saying using different models for different parts of the task is a fundamental difference from using one model for different parts of the task?
Using different models for different things isn’t new at all. The article seems like an excuse to get some marketing out there (and it’s poor at that - they got me looking at what was built with their product but I can’t see the actual code. Feels scammy.)
1. Multi-Agent is divide a part into tasks and hand off each part to a different Agent. This is different in the sense that a task is not divided into parts aprior. When the agent gets to a roadblock - lets say it is unable to fix a software issue - it rolls up to a deep think model to unblock. But you might be right that the difference is too subtle too notice.
2. "they got me looking at what was built with their product but I can’t see the actual code. Feels scammy" - What do you mean by you can't see the actual code ? You can just signup and use NonBioS to build software. And you can see the code written by NonBioS in multiple ways - ask it give you a downloadable zip, ask it to checkin the code to github, ask it to show you the code on the screen. Infact that the black boxes which scroll up, you can just expand them and see the code it is writing directly.
Sounds like 3 card monte, sounds like the fast but short thinkers are running out of analogies, they might actually have to think and realize that the common assessment of AI as an automation technology is not correct. It's a muse and a Socratic mentor, a lobotomy when tasked to think for you, and a Rube Goldberg Machine when applied to automation.
1. Why is the author calling this agencies. He is talking about multi-agent systems, a space with research spanning decades (check https://en.wikipedia.org/wiki/Multi-agent_system ). Renaming this to agencies is weird.
2. Creating single agent systems is already quite tricky. The best practices and LLMOps workflows are far from mature. Jumping to multi-agent systems is very early imo. My suggestion to any builder in this space is to start simple, very simple, and then add complexity, instead of building a house of cards.
He is NOT talking about multi-agent systems, which is exactly why he is calling it an Agency. The author goes to great length to explain why this is NOT a multi-agent system because it can be easily misunderstood to be that.
Can we stop it with the AI spam ads on HN already?
until some of the significant flaws of agents are addressed (hallucination, explainability, bias), I'm not really all that interested in extending this model further.
Agentic AI definitely works for software engineering because we have suitable mitigations for its limitations. It is unclear what those mitigations might be in other fields of application.