Ford recently rehired around 300 engineers it had let go as part of a push to automate more of its engineering work. The headline writes itself as a story about AI falling short, about technology overpromising and underdelivering. But that reading misses the more useful point.
This was not a story about AI failing. It was a story about an organisation discovering, at some cost, that it had not decided who was responsible for the judgement calls that AI cannot make. The engineers were not brought back because the tools stopped working. They were brought back because the tools were working, and no one in the room was qualified to know whether to trust what the tools were saying.
The gap that does not show up on a roadmap
Most businesses thinking seriously about AI are focused on the right things: which processes to automate, which tools to evaluate, what the integration looks like, what it costs. These are legitimate questions and they deserve careful attention.
But there is a quieter question that tends to get skipped, not out of negligence but because it is harder to put on a slide. When the AI produces an output, and something about it feels off, or consequential, or simply important enough to pause on, who in your business is the person that call belongs to?
Not who presses the button. Not who reads the output. Who owns the decision about whether to act on it.
In many organisations, that question has no clean answer. The tool sits inside a workflow, the workflow sits inside a team, and the team assumes that someone upstream or downstream is holding the judgement. Often, no one is. The output moves forward because it came from a system, and systems feel authoritative in a way that a colleague’s rough draft does not.
Authority without accountability
This is the specific risk that the Ford story illustrates. AI outputs carry a kind of implicit authority. They arrive formatted, confident, and fast. They do not hedge the way a junior analyst might. They do not say “I am not sure about this part.” They present, and the human on the receiving end has to decide whether to interrogate or accept.
That decision, the one about whether to interrogate or accept, is a skilled act. It requires domain knowledge, contextual judgement, and a clear sense of what the stakes are. It is not something that can itself be automated. And in organisations that have moved quickly to reduce the number of people with that domain knowledge in the room, the capacity to make that call well has quietly eroded.
The engineers Ford rehired were not there to do what the AI could not do. They were there to know when the AI was wrong, and to understand why it mattered.
A practical question worth sitting with
For most businesses, the risk is not as acute as it was for Ford. But the underlying dynamic is the same, and it scales down to smaller decisions than automotive engineering.
Think about the AI-assisted work already moving through your business. Strategy documents, client briefs, research summaries, financial models, creative outputs. At each point where that work becomes a decision, or informs one, is there a named person whose job it is to evaluate it critically? Someone with enough context to know what a good output looks like, and enough standing to push back when it does not?
If the honest answer is “roughly, yes, but it is not explicit,” that is worth making explicit. Not because something is about to go wrong, but because clarity here is cheap and the alternative is not.
The businesses that use AI well over time are not necessarily the ones with the most sophisticated tools. They tend to be the ones that have thought carefully about where human judgement sits in the workflow, and have made sure that judgement is held by someone with the knowledge and the authority to exercise it.
AI is very good at producing answers. The question of whether this is the right answer, right now, for this situation, is still yours.
That is not a limitation to work around. It is the thing worth protecting.
