AI Agents Are Just Distributed Systems Now — Salman Munaf, TikTok
AI agents require distributed systems thinking because they now cause real-world side effects.
“we shouldn't ideally have uh allow AI agents to delete production databases.”
A TikTok engineer argues that AI agents have crossed an architectural threshold — they are no longer isolated text processors but probabilistic coordinators interacting with external systems and causing real state changes. The talk frames agent reliability failures (Replicate's deleted production DB, Air Canada's hallucinated refund policy) as distributed systems problems solvable with familiar patterns: scoped authority, robust backups, and authoritative retrieval. The framing is notable because it repositions agent safety as an engineering discipline problem rather than a model quality problem.