Most AI Agents Fail: Why Workflows Matter More

Why this matters

Chandra's warning lands because most enterprise AI failures we've tracked trace back to exactly what he describes: bolting agents onto processes nobody bothered to fix first. That's a governance problem dressed up as a technology problem. For founders pitching agentic AI, the pitch can't just be "our agent automates step X." It has to include an honest answer about whether step X should exist at all.

For developers building orchestration layers, this is a reminder that connecting agents to data and systems safely matters less if the underlying workflow logic is broken; you're just automating dysfunction faster. And for researchers, it's a useful signal about where the real bottleneck sits, not in model capability but in organizational readiness. NiCE has an obvious commercial interest in selling orchestration and workflow tooling, so read the specifics with that in mind.

But the core observation, that scaling agents exposes process debt rather than fixing it, tracks with what we're hearing elsewhere. Worth watching whether "workflow redesign" becomes its own product category before agentic AI matures much further.