The Accountability Gap

When speed is the metric, nobody names who checks the work.

When the sprint ends, the question nobody asks

The sprint review ends the way it usually does. The numbers are good. Output is up. The team shipped more this month than any previous month on record.

Nobody asks who reviewed it.

The question doesn’t fit the mood. The mood is celebration. A product manager thanks the team. A director flags the numbers in a message to leadership. Everyone files out.

Somewhere in that output is a decision nobody clearly made.

Not a catastrophic one. Not yet. A judgment where the AI tool generated something, a team member approved it without deep review, a manager signed off because the process said to, and leadership counted it as a win. The chain of custody for that judgment is genuinely unclear.

Nobody is hiding. Nobody is lying. The accountability gap isn’t a character problem. It’s a structural one.

Speed moves faster than clarity

When organizations introduce tools that accelerate output, they rarely redesign accountability structures at the same rate.

Sociologist Charles Perrow identified something important about complex systems: when components become tightly coupled and their interactions grow opaque, failures stop belonging to individuals. They belong to the system.¹ He was writing about industrial disasters. The principle applies to organizational ones.

AI-assisted teams are building tightly coupled systems without noticing. The tool generates. A team member reviews. A manager approves. Leadership ships. Each person in that chain made a reasonable local decision. Nobody made the whole one.

What fills the gap is assumption.

The reviewer assumes the tool is reliable. The manager assumes the reviewer checked closely. Leadership assumes the manager would flag a problem. When something goes wrong, the post-mortem becomes an archaeology project. Where was judgment actually exercised?

The moral crumple zone

Researcher M.C. Elish coined a term for what follows: the moral crumple zone. When a complex automated system fails, accountability doesn’t distribute evenly across the chain. It collapses onto the nearest human operator. The team member who clicked approve. The manager who signed off. Even when their actual control over the outcome was limited.²

The system is protected. The human absorbs the impact.

Sociologist Diane Vaughan’s study of the Challenger disaster traced how organizations drift into dangerous practices not through recklessness, but through repetition.³ What didn’t fail last time becomes the template for next time. The deviation normalizes. The standard quietly moves.

AI-assisted teams do something similar. Not with bad intent. They optimize for what the organization rewards: speed, output, throughput. Nobody explicitly decides who is responsible for quality when the tool does most of the generation. The question just doesn’t come up, because it never had to before.

Karl Weick’s research on organizational sensemaking found that shared understanding of responsibility is not a given.⁴ It is constructed through practice, language, and explicit structures. When those structures are absent or outdated, people act but nobody is certain who the action belongs to.

That is the condition AI acceleration creates inside teams that haven’t caught up to it.

A design problem, not a technology problem

The accountability gap is not a technology problem. AI tools are doing what they were designed to do.

It is a design problem. Organizations haven’t answered the question that speed made urgent: when the tool generates and the human approves, who owns the result?

That sounds simple. It isn’t.

Ownership isn’t only about assigning blame after a failure. It’s about the standing conditions of judgment. Who is positioned to evaluate quality? Who has the authority to slow things down when something feels wrong? Who is trusted to name a problem before it ships?

In teams where velocity is the celebrated metric, those questions feel like friction. They slow the sprint. They interrupt the mood. They require someone to say, in a room full of good numbers, that not everything that shipped is something they can vouch for.

That is not a comfortable position to occupy.

But capable leaders who occupy it anyway are doing something more durable than the ones who keep their heads down and their numbers up. They are building the conditions for accountability before a failure arrives rather than after.

The gap is not inevitable. It is a design choice made by omission.

The leaders positioned to close it are the ones willing to ask, in a room where the mood is celebration: who checked the work?

That question isn’t friction. It’s function.

Sources & References

1. Perrow, C. (1984). Normal Accidents: Living with High-Risk Technologies. Basic Books. ISBN: 978–0691004129
2. Elish, M.C. (2019). Moral Crumple Zones: Cautionary Tales in Human-Robot Interaction. Engaging Science, Technology, and Society, 5, 40–60. DOI: 10.17351/ESTS2019.260
3. Vaughan, D. (1996). The Challenger Launch Decision: Risky Technology, Culture, and Deviance at NASA. University of Chicago Press. ISBN: 978–0226851754
4. Weick, K. E. (1995). Sensemaking in Organizations. Sage Publications. ISBN: 978–0803971776