The flat team is the year’s favourite slide. One amplified operator, a stable of agents, headcount falling, capital per head rising. Tobi Lütke told Shopify in April 2025 that staff must prove a task cannot be done by AI before asking for more people. Brian Armstrong cut roughly a seventh of Coinbase and told the survivors there would be no more pure managers — every leader an individual contributor too. The executive coach Joe Hudson, writing for Lenny’s Newsletter in June, gave the shape a name: teams will stop looking like factories and start looking like NBA rosters, each player carrying far more of the outcome. The roster is sold as the brave, AI-native move.
It is also the supermarketer, arriving by reorganisation rather than by demo.
I described that figure in May: not a single human doing more work, but a human orchestrating a stable of agents, each with bounded authority, each accountable to outcomes the human agreed to be measured on but did not personally execute. The argument then was that we had learned how to build the supermarketer and not how to manage one, and that the unsolved problem was organisational. Who escalates the goal. Who approves expanding an agent from recommendation to autonomous spend. Who answers when it goes wrong. I called the missing role a controller for the autonomy ledger, and noted that the org chart had no row for it.
Focus On: Where the ledger goes
Hudson’s claim is that the scarce input has changed. When knowledge and effort are nearly free (and inside the frontier labs he coaches, they are closer to free than anywhere else), the differentiator becomes emotional clarity: the capacity to feel what you feel without being run by it. The four traits he sees in people who thrive he calls a wisdom stack: discernment, productive conflict, a willingness to fail, and the tone of one’s own internal voice.
Set aside, for a moment, that he sells the cure. The diagnosis points somewhere real, and it is not where the consensus is looking.
Recall the supermarketer’s escalation ladder. An agent good enough to recommend an audience is, with one configuration change, good enough to assemble it; with another, to launch the campaign; with another, to commit budget with no human in the loop. In a layered organisation each rung is a decision that diffuses upward. A manager co-signs. A committee shares the call. The hierarchy quietly guarantees that no single name carries the consequence alone. Flatten the organisation to build the supermarketer, and the diffusion leaves with the layers. The ledger outlives them, and lands on one person, undivided.
That is the migration the brave slide omits. The constraint the May issue called organisational was always, underneath, a question about where the consequence of a wrong call is permitted to land. The flat team answers it by accident, and the answer is: on you.
The one thing the agent cannot do
The agent will do almost everything. It will run the analysis, draft the action, model the downside, and rehearse the difficult message ten times over. What it cannot do is hold the consequence. A model has no career, no reputation, and no skin in the outcome. Accountability does not transfer to it, and on a flat team it no longer diffuses across colleagues. The binding act — the one that does not scale and cannot be delegated — is committing to a call under irreducible uncertainty, knowing the machine has removed every excuse you might once have offered for getting it wrong.
That act is emotional before it is rational, and the evidence for this is older and harder than anything in a coaching deck. Antonio Damasio’s patients with damage to the ventromedial prefrontal cortex keep their intelligence, memory, and reasoning intact, and yet lose the ability to make ordinary decisions, because they have lost access to the emotional signal that ranks the options. Reasoning stripped of affect does not produce caution. It produces paralysis. So when AI commoditises the reasoning, the affective capacity to decide becomes the work that is left.
Two further mechanisms explain why that bottleneck tightens precisely as the team flattens. The brain runs a failure brake: the lateral habenula suppresses dopamine in anticipation of a poor outcome, dampening the will to take the next shot. On a flat roster every shot is more visible and more consequential, the felt cost of a miss rises, and the brake bites harder. Self-criticism, meanwhile, is not free fuel. Worry and rumination (what the literature calls perseverative cognition) prolong cortisol output and depress heart-rate variability; the body answers a thought about a threat as it answers the threat itself. The hostile internal voice that once pushed people to outwork the person beside them now taxes the exact composure the flat team demands more of.
AI removes the excuses. Flattening removes the shock absorbers. The nervous system explains why what remains is harder, not easier, at the moment it becomes most valuable.
The marketing test case
The CMO running agentic campaigns is the first to feel all of this at once. The agent assembles the audience and drafts the creative; it even forecasts the lift it will be judged on. The human owns the number, and the brand it could quietly degrade within a quarter. When I piloted an agent for personalised outbound at S&P Global, the technology was not the obstacle. The pilot stalled on a single question — at what volume does this stop being marketing-led and become compliance-led — and nobody had been empowered to own it. That was the ledger landing on people who had not agreed to carry it.
The May prescription was to hire the accountable person and write the org chart that lets them do the job. The amendment is this: hiring the name is not enough if the structure gives their wrong calls nowhere to land but their own bloodstream. Designing where consequence lands is concrete work, not a values statement. Adriano Olivetti built Ivrea on the principle that the factory served the people inside it, not the reverse; the flat AI organisation inverts him, asking one person to absorb whatever the machine’s output costs. The fix is structural and dull. It means explicit decision rights for each step up the autonomy ladder, with a named owner and a budget rather than an implied volunteer; and a review of the misses that interrogates the decision and not the person, so the organisation absorbs the variance instead of the individual’s cortisol. Do that, or the role burns through its holder inside a year, and you begin again with someone less willing to decide.
What I would watch
The next eighteen months will produce two stories that are the same story told twice: a loud market in resilience coaching for the AI age, and a quieter epidemic of decision paralysis on flattened teams. The firms that come out ahead will not be the ones that sent the supermarketer on a mindfulness retreat. They will be the ones that rebuilt, on purpose and in the design of the work, a place for a wrong call to land.
Emotional clarity can be trained. An organisation that makes the first unhedged decision unsurvivable cannot be trained out of itself. Fix the second before you pay to fix the first.
Two questions worth sitting with this week. If one of your amplified operators makes a wrong call next quarter, where does the consequence land, and is it somewhere they can come back from? And if you could fund only one thing this year — an individual’s composure or the redesign of the organisation that taxes it — which are you actually budgeting for?
Follow me
That’s all for this week. To keep up with the latest in generative AI and its relevance to your digital transformation programs, follow me on LinkedIn or subscribe to this newsletter.
Disclaimer: The views and opinions expressed in Chronicles of Change and on my social media accounts are my own and do not necessarily reflect the official policy or position of S&P Global.