Don't Build Multi-Agents
The contrarian essay from the Devin team that reset the industry's defaults.
Open cognition.com →
Building & Product
Article
Thread
Thread
Article
Article
Article
Article
Article
Article
Tool
Article
Article
Article
Article
Article
Article
Article
Video
Should I build one agent or a team of agents working together?
The short answer
Start with one. The famous industry debate (Cognition's "Don't Build Multi-Agents" vs Anthropic's multi-agent research system) actually agrees on the fundamentals: multiple agents working from partial context make conflicting decisions, and multi-agent setups burn roughly 15x the tokens. Go multi-agent only when the work is genuinely parallelizable and read-heavy, like broad research, and even then use one lead agent delegating to sub-agents, not a committee of equals.
A quick orientation. The real value is below: resources worth your time, from people who've actually done it.
18 resources worth your time
Walden Yan: principles for building agents
The thread version of the argument, with the mistakes he keeps seeing.
Open x.com →Walden Yan: a year later, what changed
The honest update: some multi-agent setups now work, most sexy ideas still don't.
Open x.com →Multi-Agents: What's Actually Working
The follow-up naming the specific multi-agent patterns that survived production.
Open cognition.com →How we built our multi-agent research system
The other side of the debate: a 90% performance gain, at 15x the tokens.
Open anthropic.com →When to use multi-agent systems (and when not to)
Anthropic's own decision rules for graduating beyond a single agent.
Open claude.com →Single vs Multi-Agent System?
Shows why Cognition and Anthropic are both right about different problems.
Open philschmid.de →How Anthropic Built a Multi-Agent Research System
The architecture broken down with diagrams a non-researcher can follow.
Open blog.bytebytego.com →Why Do Multi-Agent LLM Systems Fail?
200+ analyzed tasks and 14 failure modes, the evidence behind the caution.
Open arxiv.org →MAST: Multi-Agent System Failure Taxonomy
The open dataset and annotator to diagnose your own multi-agent failures.
Open github.com →MAST project page
The digestible summary of where multi-agent systems break.
Open sky.cs.berkeley.edu →AI Leaders Clash Over Agent Architecture
A neutral recap of the whole debate if you want it in ten minutes.
Open ctol.digital →Multi-Agent vs Single Agent: When It's Worth the Cost
Frames the choice as a coordination-cost calculation, which is the right lens.
Open cognilium.ai →The Art of Multi-Agent Collaboration: Reflections on Anthropic's Research System
Extracts the engineering lessons hiding inside Anthropic's post.
Open llmmultiagents.com →Why do multi-agent LLM systems fail (and how to fix)
Maps each MAST failure mode to a practical fix.
Open futureagi.substack.com →Why Do Multi-Agent LLM Systems Fail? (explainer)
The paper's findings retold for people who won't read a paper.
Open thegrigorian.medium.com →Anthropic's multi-agent research system (LLMOps case study)
The case study written for someone deciding whether to copy the pattern.
Open zenml.io →
Harrison Chase on Deep Agents, LangSmith, and Earning Trust
The emerging alternative: one deep long-running agent instead of many shallow ones.
Watch on YouTube youtube.com →More in AI Agents
- What actually is an AI agent, in plain language? 18 →
- Why would a founder use agents instead of a chatbot or automation? 18 →
- What real tasks are founders delegating to agents today? 20 →
- How do I build my first agent without being technical? 20 →
- Which platform or framework should I build my agents on? 18 →
- How much do AI agents cost to run, and how do I keep the bill under control? 18 →
- How do I test an AI agent and know when it's reliable enough to trust? 18 →
- What can go wrong when an agent acts on my behalf, and how do I keep it safe? 16 →
- What is MCP, and why does it matter for my agents? 18 →
- Can coding agents like Claude Code actually build my product? 18 →
- Is there a real startup opportunity in building vertical AI agents? 19 →
- How should I price a product built on AI agents? 18 →
- Why do most AI agent projects fail, and what do the winners do differently? 19 →
- How do I give my agent the right context and memory so it stops forgetting? 18 →