About
The definitive resource on why agents break
How Agents Fail is a research collective studying failure modes in production AI agent systems — the gap between what autonomous AI promises and what it delivers at scale.
Why this exists
The AI agent space is moving at a speed that outpaces its own failure taxonomy. Teams are shipping autonomous systems into production before the field has developed shared vocabulary for what goes wrong, why it goes wrong, and what to do about it.
Most writing about AI agents is promotional. It's demos, benchmarks, and capability showcases. The hard stuff — the memory failures, the planning collapses, the tool-use edge cases, the silent degradation under load — lives in Slack threads, post-mortems, and conference hallways.
We're bringing it into the open.
What we do
We study the architecture of AI systems, listen to engineering teams about their production experiences, and publish rigorous analysis of failure patterns. Our research is empirical, not theoretical — we care about what actually breaks, not what the benchmark says might break.
We're building across every medium where someone might first learn about agent failure: long-form research, a weekly newsletter, expert profiles, and eventually video and audio. If you're trying to understand why your agent isn't working, we want to be the place you end up.
Who we are
We're a small team of researchers and practitioners with backgrounds in systems engineering, machine learning, and production software. We've seen these failures from the inside — as engineers, as consultants, and as researchers — and we think the field needs a dedicated institution to make sense of them.
Contribute
If you've encountered an agent failure that taught you something — and you want to help others learn from it — we'd love to hear from you. We work with practitioners to turn real production experiences into research that protects confidentiality while generating value for the field.