My most productive day last month, I was the only human in a meeting of six.
The other five were agents. One had spent the night reading a competitor's documentation and had a summary waiting. One was holding the thread of a half-finished refactor I'd abandoned on Friday. One was drafting outreach. One was checking the drafts against what we'd actually shipped, so I wouldn't promise something that didn't exist. The fifth mostly listened, took notes, and reminded me of the three things I'd said yesterday that I'd already forgotten. I talked. They worked. By lunch I'd done what used to take a team and a week, and I'd done it without scheduling anything, without a standup, without waiting on anyone.
I want to be careful here, because this is the kind of sentence that gets a founder mocked in six months. So let me be precise about what happened and what didn't. The agents were not brilliant. Two of them produced work I threw away. One had quietly drifted off the brief and spent the night refining the wrong half of the problem, because I'd been sloppy in how I set it up. The thing that changed my morning wasn't that the machines were smart. It was that there were five of them, each pointed at a different part of one problem, and I was the only one holding the whole shape of it in my head.
That arrangement has a name, and we've had it for four hundred years. It's an orchestra. One person stands in front, holding the score, and the sound comes from everywhere else.
The score is the same; now one person plays all the parts
For most of the history of work, the conductor was a luxury good. You needed an orchestra before you needed someone to conduct it — sixty musicians, a hall, a payroll, a season. The whole apparatus existed first, and then a single person was placed at its center to shape it. Conducting was a role you earned by climbing through an institution large enough to require coordination.
The strange thing happening now is that the apparatus is getting cheap and the conductor is becoming the common case. The musicians are software. They don't need a hall or a salary or a season. They need a score and a downbeat. And it turns out that the skill that matters most in front of an orchestra was never playing an instrument better than the players. It was holding the whole piece in your head and deciding, second to second, what should be loud and what should be silent.
This is the part most people get wrong when they imagine AI at work. They picture a smarter tool — a faster violin. Or they picture replacement — a player piano that needs no one. Both pictures keep the old org chart and just swap who sits in the chairs. The more interesting thing is structural, and it has almost nothing to do with how good any single model is. It's that the cost of standing up a team has collapsed to roughly the cost of describing what you want.
When the cost of a thing falls far enough, you don't get more of the same thing. You get a different thing. Cheap printing didn't give us more monks copying manuscripts by hand; it gave us newspapers, pamphlets, a reading public — a whole new shape of who could put words in front of whom. Cheap coordination won't give us more middle managers coordinating people. It will give us a new smallest unit of work: the individual, with their machines.
The unit of work is shrinking to one
Consider how a company has been built, structurally, since the railroads. You hire a person to do a thing. When the thing gets too big for one person, you hire a second, and now you need a third person to keep the first two pointed the same way. The org chart is, at bottom, a coordination tax. Every box above the people doing the work exists because humans are expensive to align and slow to inform. The pyramid is not a picture of ambition. It's a picture of the cost of getting people to agree on what's true and what's next.
Take a large slice of that cost to near zero and the pyramid stops being the obvious shape. If a single person can stand up five capable workers in the time it takes to write a paragraph, and direct them with a sentence, then a great deal of what we built management to do — the assigning, the chasing, the status-gathering, the translating of one person's output into another's input — happens inside one head and one set of tools. The coordination didn't disappear. It moved. It became something the individual does for themselves, continuously, the way a conductor coordinates an orchestra without a single memo.
I think this is the shift, and I think most of the conversation is aimed in the wrong direction to see it. The dominant framing in 2025 is the autonomous agent that replaces a function — the AI that does sales, the AI that does support, a new department made of software sitting beside the old departments made of people. That will happen in places, and it will be oversold in more. But it keeps the old picture: agents as a layer, above or below the humans, another tier in the pyramid.
The picture I'd bet on is different. Not a tier of agents. A ring of them. Around each person, a small, owned, persistent team — three to seven workers who know how that person works, hold their context, and extend their reach. Not the company's agents, loaned out. The employee's agents, the way a craftsman owns their tools. The unit you hire, evaluate, and pay is no longer a person who can do the work of one. It's a person who can conduct the work of a section.
Why a ring and not a tier
The reason this matters isn't org-chart aesthetics. It's that a ring and a tier produce different work.
A tier of shared, central agents optimizes for the company's average case. It is built once, by a platform team, to serve everyone, which means it serves no one's actual texture. It doesn't know that you always forget to check the contract before you send the proposal, or that your best ideas come out sideways in the third draft and never the first. Central tooling can't hold that, because holding it for ten thousand people is the same as holding it for none.
A ring is personal by construction. The agents around you accumulate your particular context — the way you phrase things, the mistakes you reliably make, the standards you won't compromise, the five sources you actually trust. Over months, that ring becomes less like software and more like a small staff that has worked with you long enough to anticipate you. The value isn't in any single model's intelligence. It's in the fit. And fit is the one thing a centralized platform can never manufacture and a person can never be talked out of valuing.
This is also why I think the durable products in this space will be installed at the level of the individual, not procured at the level of the enterprise. I'll admit I'm not a neutral observer here — it's the bet my company is built on, that you put the agents next to the person doing the work rather than in a platform two floors up. But the logic stands apart from any company. The enterprise buys the average. The individual owns the fit. And the work that wins is the work with fit. We've watched this movie before: the personal computer didn't beat the mainframe by being more powerful. It beat it by being yours.
The conductor's actual job
If the unit of work becomes the individual with their machines, then the scarce skill is the conductor's skill, and it's worth being honest about what that skill actually is, because it is not what most people are training for.
It is not knowing more than your players. The agents will, in narrow stretches, know more than you — read faster, recall more, never tire. The conductor in front of a great orchestra is frequently the worst instrumentalist in the room, and it does not matter, because their job was never to play. Their job is taste and direction: holding the whole piece, hearing the thing that's slightly wrong before anyone else does, and deciding what gets emphasis and what gets cut. Those are judgment skills, and they have been quietly undervalued for a generation that optimized for being the smartest individual contributor in the room.
It is also, and people skip this part, the skill of writing a clear score. An orchestra plays from notation precise enough that sixty people can be coordinated by marks on a page. Most knowledge work has never had to be that precise, because a human collaborator fills the gaps with common sense. Agents fill the gaps too — just not always with sense. The discipline this demands is the discipline of saying exactly what you mean, which is rarer than it sounds and which most organizations actively punish, because vagueness is how you avoid being wrong on the record. The conductors who win will be the ones who can specify.
I'll say something personal, because it's relevant and because I've stopped pretending otherwise. I was diagnosed with ADHD late, long after I'd built a working theory of myself without it. For most of my working life, the single-track corporate path felt like swimming through molasses — handed one repetitive task, told to stay in my lane, watching my attention leave the building. What I was actually good at, it turned out, was running six things at once and holding the shape of all of them. For decades that was a defect, because the work was built for the violinist, not the conductor. The first time I sat in a meeting of six where five were agents, the thing I felt wasn't novelty. It was recognition. The work had finally bent into the shape my head was always in.
I don't offer that as inspiration. I offer it as data. A mind that scatters across many threads was a liability in a world where the unit of work was one person doing one thing well. It is an asset in a world where the unit of work is one person directing seven things at once. The wiring didn't change. The work did. And if the work is genuinely changing for everyone, then a lot of people currently filed under "can't focus" are about to discover they were built to conduct.
What this asks of everyone else
The uncomfortable implication is that the middle of the org chart — the layer whose job was coordination, status, and translation — is the layer most exposed. Not because those people lack value, but because the specific scarcity they were solving for is the one collapsing fastest. The honest move isn't to reassure them that nothing will change. It's to tell them the truth: the skill is migrating from coordinating people to conducting machines, and the two are closer than they look. Anyone who was good at keeping a team aligned already has the core instinct. They just have to learn that the players don't have feelings to manage, only instructions to clarify.
For the people who do the making — the writers, analysts, engineers, designers — the message is gentler and stranger. You're about to be handed a section. Nobody is going to teach you to conduct, the same way nobody taught most managers to manage. You'll learn it the way everyone learns it: by standing in front of the thing and discovering that the sound is yours to shape, that the silences matter as much as the notes, and that the players will do exactly what you say, which is precisely the problem.
There's a version of the next decade where we get this wrong and spend it building bigger pyramids — central AI platforms, agent tiers, a new bureaucracy of software stacked on the old bureaucracy of people. It's the path of least imagination, and large organizations are very good at taking it. But I don't think it holds. The pull of the personal is too strong, and it always has been. We end up wanting the tool that's ours, that knows us, that fits. We end up, each of us, in front of our own small orchestra.
The score, in the end, is the same as it ever was — the work the world needs done. What's new is who gets to stand in front of it. For most of history, that was the rare few who'd climbed high enough to be handed a baton. Now the baton is cheap, and the instruments are software, and the only question left is whether you can hear the music well enough to shape it. A great many people are about to find out they can.