← back
Building Reliable Agentic Systems: Eno Reyes
Takeaway
Reliable agentic systems borrow robotics-style state filtering, MPC replanning, consensus sampling, and unapologetic hardcoded plan criteria.
Summary
- Factory builds 'droids' — autonomous SWE agents for review, docs, testing, refactors — with separate cognitive architectures per task.
- Borrows from control theory and robotics: 'pseudo Kalman filter' to pass intermediate reasoning between plan steps for convergence; model predictive control for adaptive replanning on real-time feedback.
- Decision-making toolkit: self-consistency / prompt ensembles / cluster sampling, explicit/analogical reasoning (chain-of-thought, checklists, chain of density), domain fine-tuning for out-of-distribution decisions, simulation for environments like software dev.
- Argues for explicit plan criteria and hardcoded logic (symbolic-AI lessons) since 'we're not building AGI in six months'.
agentsplanningreliability
Original description
Agentic system design is a rapidly evolving and intellectually fascinating field, with huge potential for transforming how software is used across industries. Unlike traditional software, agentic systems rely on non-deterministic and oftentimes difficult to predict decision making. Taking inspiration from fields like robotics, cybernetics, and biology, we can start to develop intuitions around how to build systems that are ~more~ reliable than the sum of their individual stochastic parts. Recorded live in San Francisco at the AI Engineer World's Fair. See the full schedule of talks at https://www.ai.engineer/worldsfair/2024/schedule & join us at the AI Engineer World's Fair in 2025! Get your tickets today at https://ai.engineer/2025 About Eno Eno Reyes is founder and CTO of Factory.ai, an organization developing autonomous systems called Droids that automate software development tasks. Prior to Factory, he was an ML engineer at Hugging Face working on enterprise LLM deployment, fine-tuning, and productionization.