I’m imagining a lonely hearts ad appearing later this year somewhere on LinkedIn.
“Lonely loop, looking for human to help me find relevance. Must be confident, strong-minded, and assertive. No time wasters or sacrificial lambs.”
This is a great post by Neil Perkins from Friday, raising an important question for organisations adopting AI: whether fluent, rapid outputs are quietly weakening the human judgement needed to assess them, that, over time, their own standards of what counts as good may drift downwards and that the remedy is not a token “human in the loop” at the final approval stage, but deliberate human involvement in framing problems, setting fixed quality criteria, resourcing review properly, and giving people real authority to challenge or reject machine-generated work
The post made me smile, as I expect it did for many others with grey hair who remember the early days of Total Quality Management and the Toyota Production System. In particular, the idea of the Andon cord that could be pulled by any operator on the production line to stop the system if they saw a problem was heresy in a manufacturing culture that worshipped throughput and was happy to assess quality at the end of the process rather than during it.
It was a time when algorithmic thinking still resided in human brains rather than silicon ones. The notion that the beautiful plans created by people on the 10th floor could be interrupted by a by an oik in an overall offended all the sensibilities of the people in suits. But we know that over a remarkably short time in which “Made in Japan” went from being a derogatory term to one evoking something like awe, as they put the teachings of W. Edward Deming that were ignored in the West to good use.
In the intervening period, we have moved judgment away from people, increasingly to algorithmic processes operated by people (like Lean Six Sigma), and then away from humans onto silicon.
Measured in money, it’s hard to argue with the success.
Measured in what matters, we have ended up in a cul-de-sac or relevance, with nowhere to go except to trace our steps back a little.
Not to go backwards, but rather to look at where we got lost.
Because whilst AI can reduce complexity in representation; people must judge complexity in reality, because reality resides in all the bits that data cannot represent.
Neil Perkins makes the point well in his post. We can reduce the time to write a monthly report from, say, 3 hours to, say, 20 minutes. What wwe cannot do is check it as we go, so we’re still left with the checking time. Meantime, in the 2 hours 40 minutes that we have saved, we’ve produced another 8 reports, and find ourselves back in the land of the sorcerer’s apprentice, with help we can neither control nor use.
It is not difficult to imagine a complex chain of AI agents, having some sort of a digital frat party, and it leaves us with a real dilemma. The more consequential and less reversible an agent’s action, the stronger and more proximate the human authority should be. If there’s no operator on the line to sense what’s going on and pull the andon cord, we end up back in the 1960s, with quality control departments obsessing and fighting against those who argue for quality assurance.
And if we need not just one, but an ecosystem of humans in the loop, then we have to consider the organisational physics, the nature of leadership, in terms of enabling communication and empowerment. It stops being academic notes in textbooks that end up in the bin after business school, and become something that needs to be lived every day. Then consider that the leadership cannot be remote, back on the 10th floor, it has to be close to where the action is happening.
If that were to be true, perhaps we end up with something where the principles of scale become inverted. It used to be the 500-person company could be more efficient and do more than ten 50-person companies, but when an AI makes a 50-person company equivalent to a 500-person company, the rules change.
Joe Procopio notices this, in his blog, also from Friday:
This is going to get interesting.
Perhaps we are approaching a similar transition with AI: away from the comforting fiction that we can let an ever-faster chain of agents run and then inspect its output at the end, towards a system in which people are close enough to the work to notice when reality has departed from the model and have both the authority and the confidence to stop it. Toyota’s andon made this principle physical: any worker could signal an abnormality and halt the line, not as an act of failure but as the means of building quality into the process rather than sorting defects afterwards. The AI equivalent is not a sacrificial “human in the loop” asked to rubber-stamp eight reports produced in the time once needed for one; it is an ecosystem of strong-minded people with clear criteria, real visibility, and permission to challenge the system before its speed turns small errors into institutional fact.
The lonely loop, in other words, is not looking for a compliant chaperone at the end of the line, but for an operator with an andon cord.
Perkins is on to something. My assumption has been that we will treat AI like we treat every other technology. We capture the low hanging fruit and feel satisfied. I recently spent a week at the beach asking Claude questions about my creative output. We are talking about a million and a half words and 380 podcast episodes. It is leading to two projects. One the follow-up leadership book to Circle of Impact, and the other related to the podcast. I’m challenging the assumption that I know not only what I have created, but its value. I have begun asking it to compare my work to other authors. I’m at 40 writers so far. I’m learning their core ideas, how my work connects, and what’s the gap that could be filled by me. It gave me 3-4 page reports in about two hours. I’m saying AI is being used as a labor saving device, when its real value is leveraging what exists to do what could never be done in a lifetime. I’m already deciding places not to invest time. Ironically, Claude has already asked if I want it to write the first draft of the book. I started laughing. I shut that idea down quickly.