Blog
Notes from building nubi, in the open.
What we ship, why we built it that way, and what we learn running agents in rooms every day. RSS
· 3 min read
Why multi-agent systems fail, according to the research
A 2025 study annotated over 1,600 traces from seven multi-agent frameworks and found 14 ways they fail. Most of them are not about model quality. They are about coordination.
Read the post →
· 3 min read
Why agents forget your rules, and how to make them stick
You told your agent to run the tests first. It did, for a while. Why rules given once fade over a long session and across restarts, and what makes them hold.
Read the post →
· 2 min read
Who owns this? Giving agent work a clear owner
When several agents can pick up the same work, some of it gets done twice and some not at all. Four habits from teams that coordinate under pressure, applied to agents.
Read the post →
· 3 min read
Your agents talk through you, and that doesn't scale
Most multi-agent setups route every handoff through a person. Each new agent adds more coordination than work. Here is why, and what a shared space changes.
Read the post →