What Drew Breunig Just Got Right About the Hidden Cost of AI at Work

Drew Breunig — chief executive officer (CEO) and co-founder of cmpnd.ai, and author of the forthcoming The Context Engineering Handbook — is not a conference-circuit futurist. He is a working technologist with experience across multiple eras of computing, and that depth is exactly why his recent essay on prompt debt deserves more attention from senior HR and operations leaders than the average AI manifesto currently flooding inboxes.

The core finding, stated plainly: organizations that deploy AI tools without disciplined context management accumulate a form of invisible technical and organizational debt — what Breunig calls "prompt debt" — that degrades both AI output quality and employee trust in those tools over time, in the same way poorly maintained codebases eventually collapse under their own weight. O'Reilly Radar (2025-07-01) This is not a metaphor. It is a structural diagnosis.

What exactly is prompt debt, and why should HR leaders care?

Prompt debt, as Breunig defines it, is the accumulation of poorly specified, inconsistent, or unreviewed instructions that organizations embed into their AI systems over time. Think of it as the AI equivalent of undocumented processes — everyone knows the system exists, nobody agrees on what it is supposed to do, and fixing it costs more than building it correctly would have. The related phenomenon Breunig calls "fighting the weights" describes what happens when employees try to coax AI models into behaviors that conflict with those models' trained tendencies, burning time and generating frustration rather than productivity. O'Reilly Radar (2025-07-01)

The reason this lands squarely on the desk of a chief human resources officer (CHRO) or chief operating officer (COO) — not just an IT director — is that prompt debt is a people problem dressed in a technology costume. The accumulation happens because no single team owns the context that flows into AI systems: HR writes policies, managers issue instructions, legal adds caveats, and none of it is coherent when assembled into a system prompt. The result is an AI that gives different answers to the same question depending on who asks, which, as any experienced manager knows, is the fastest way to destroy workforce trust in any new initiative.

Why does this matter more than Zuckerberg's latest manifesto?

Mark Zuckerberg's widely circulated AI vision — HR Executive (2025-07-01) arguing for "more jobs, fewer workers and personal superintelligence" — gets the headlines, but Breunig's work gets the mechanism. Zuckerberg's framing treats AI adoption as an inevitability that organizations simply need to absorb. Breunig's framing treats it as an engineering and governance problem that organizations can actively manage or actively mismanage. For a board conversation, the difference matters enormously: one is a fatalistic narrative, the other is an accountability framework.

The broader adoption research supports Breunig's concern. A HR Executive (2025-07-01) study on AI adoption patterns found that organizations were consistently asking the wrong question — "how do we get people to use AI?" — rather than "how do different groups experience AI differently?" That misdiagnosis is precisely how prompt debt forms: one-size-fits-all AI deployments generate inconsistent outputs for different user groups, employees distrust the results, and the organization quietly retreats to the spreadsheets it was trying to replace.

What is the one thing Breunig's work demands of leadership?

The actionable frame from Breunig's analysis is governance before scale. Before your organization deploys another AI tool to another department, assign ownership of context. That means a named person — not a committee, not a working group — is responsible for the quality, consistency, and review cadence of the instructions feeding your AI systems. This is not an IT function. It is a workforce architecture function, which means HR leadership either claims it or watches it default to whichever engineer happened to write the first system prompt.

The parallel to manager readiness is not accidental. A HR Dive (2025-07-01) report on manager skill gaps noted that "the role of the manager has fundamentally changed" — a statement so obvious it has become invisible. What Breunig makes visible is the specific dimension of that change: managers now need to understand how context shapes AI output, or they cannot meaningfully supervise the AI systems working alongside their teams.

Breunig is doing the unglamorous work of naming a problem that most AI vendors have a financial interest in ignoring. That is a service to every organization currently discovering that their AI rollout is generating as many questions as answers. The field owes him a careful read.