What actually broke when we put agents on back-office work
Summary
A practitioner shares real-world failure modes of running AI agents on back-office workflows, covering ambiguous timeouts, poorly contextualized human review queues, and shared credential management, along with fixes for each.
Similar Articles
Where AI agents actually break in real workflows (not demos)
A discussion on where AI agents fail in real workflows, highlighting issues with coordination, reliability under messy inputs, and the challenge of reducing human intervention in production.
AI agents fail in ways nobody writes about. Here's what I've actually seen.
The article highlights practical system-level failures in AI agent workflows, such as context bleed and hallucinated details, arguing that these are often infrastructure issues rather than model defects.
AI agents usually break in production for boring backend reasons, not model reasons
An experienced backend/distributed-systems engineer argues that AI agent failures in production usually stem from plumbing issues—non-idempotent tool calls, missing approval steps, uncapped costs, and poor state persistence—rather than model limitations, and shares practical hardening tips.
AI agents are starting to expose how broken most workflows already were
The article argues that AI agents are revealing how unstructured and chaotic many corporate workflows actually are, suggesting that successful automation depends more on clean systems and documentation than on advanced models.
how to scale AI agents in production workflows when the underlying business process is broken?
A practitioner shares challenges scaling multi-agent AI systems in production, including dealing with shadow workflows (undocumented Slack threads and spreadsheets), context loss across different systems (ERP to CRM), and cross-departmental ownership issues. They seek advice from others who have navigated these real-world problems.