When your team is agents, you stop assigning and start editing

A manager opens their week and there is no standup to run. No one to nudge about the ticket that’s been “in progress” since Monday. No one’s calendar to defend, no one’s blocker to escalate. Instead there are six drafts waiting in a queue: a pricing page, a customer reply, a migration plan, a competitor teardown, two bug fixes. All six were done overnight. The job for the day is not to assign any of it. The job is to read it and decide what’s good enough to ship.

That’s not a lighter version of management. It’s a different job wearing the same title.

When your team is agents, you stop assigning tasks and start reviewing drafts. The core management skill flips from delegation-and-tracking to judgment-and-feedback, which is, almost exactly, the work of an editor.

This post is about that flip: what dies, what survives, and why the skill that’s left is harder than the one it replaces.

The job management actually was

Strip a manager’s week down to its verbs and most of them are logistics. You break a goal into tasks. You match tasks to people. You track who’s doing what, chase the ones that slipped, unblock the ones that stuck, and re-plan when reality moved. Then, somewhere in the margins, you look at the output and form an opinion about whether it’s any good.

Notice the proportion. The breaking-down, the matching, the tracking, the chasing, that’s the bulk of the calendar. The judgment is the sliver at the end. Most management, honestly counted, is the overhead of coordinating slow humans who can only hold one thing at a time and forget the rest by Friday.

The naive picture of “AI for managers” keeps that whole structure and just adds a co-pilot. You still assign, you still track, you still chase, now with a chatbot in the corner that summarizes the standup you still had to run. It’s the same job, plus a tab.

That picture misses the actual change. The coordination overhead doesn’t get assisted. It gets deleted. An agent doesn’t need its task broken into a ticket, it can hold the whole goal. It doesn’t slip and need chasing, it either finishes or it doesn’t, overnight, and tells you which. It doesn’t forget the date or the context by Friday, because remembering is free for it. The breaking-down, the matching, the tracking, the chasing: those were taxes on human bandwidth. Remove the human-bandwidth bottleneck and the taxes have nothing to collect.

What’s left when you subtract the logistics is the sliver. The judgment. And the sliver was never the small part of the job, it was the part we never had time to do well.

The bottleneck moves from assigning to reviewing

Here’s the mechanism, stated plainly. The work didn’t get easier; the constraint moved.

In the old shape, the constraint was production. You had more ideas than hands. The scarce thing was getting work done at all, so management optimized for throughput: keep everyone busy, minimize idle hands, don’t let a person sit blocked. The manager was a scheduler for a resource that was always oversubscribed.

In the new shape, production is cheap. Suppose an agent team can draft, overnight, ten times what your old team shipped in a week, say a dozen finished things instead of a couple. (Take the exact number as illustration, not a promise; the point is the direction.) Now the scarce thing isn’t making the drafts. It’s deciding which ones are right. The pile of work-that-looks-done grows faster than any human can read it, and every item in that pile is a claim, “this is correct, this is on-brand, this ships”, that someone has to either trust or interrogate.

The bottleneck never disappears. It just moves. It moved off the keyboard and onto the desk where things get read.

And the instant the bottleneck is review, the job has a name. The person whose scarce resource is judgment, who sits in front of a queue of finished drafts and decides what’s good, what needs another pass, and what to kill, that’s an editor. Not a metaphor for one. The same job.

What an editor actually does that a delegator doesn’t

It’s tempting to think reviewing is the easy half, that approving is lighter than assigning. Anyone who has actually edited knows the opposite. So let me be concrete about the skill, because “judgment” is a word people nod at and then ignore.

A delegator’s core question is who and when: who’s free, who’s good at this, when will it land. Those are scheduling questions, and agents answer them for you.

An editor’s core questions are different, and no schedule answers them:

Is this actually right, or does it just look right? A draft can be fluent and wrong. The agent’s reply to the customer is well-written and quietly misstates the refund policy. The migration plan is clean and skips the one table that’s load-bearing. Fluency is not correctness, and the editor’s first job is to refuse to confuse them. A green checkmark and a confident tone are exactly the disguise that a real error wears.

Is this on-message, or just on-topic? Ten drafts can all be defensible and only three can sound like you. The editor holds the standard for voice, for taste, for what the company would and wouldn’t say, and that standard lives nowhere in the draft itself. It lives in the editor.

What’s the one change that lifts this from fine to good? This is the craft. Not rewriting it yourself, that’s regression to being the producer, but seeing the single note that turns a mediocre draft into a strong one and sending it back with that note attached. “Cut the second paragraph.” “You buried the lede.” “This is technically correct and completely misses why they’re asking.” The note is the leverage.

The naive version of agent management skips all three. It reads the green tests, sees the draft is plausible, and merges. That’s not editing, that’s rubber-stamping, and a rubber stamp is just delegation with the supervision removed. You’ve added a signature to a blind spot. The work looks done, so it ships, and the error reaches the customer wearing a checkmark.

The editor’s version is the opposite reflex. A claim of “done” is the start of the read, not the end of it. The editor pulls the draft, ignores the fact that it compiled, and asks the questions the producer structurally couldn’t ask about its own work, because the mind that wrote a thing is the worst-placed mind to catch what it missed.

That reflex, interrogate the claim, don’t forward it, is the entire job. Everything else is logistics that agents now handle.

The feedback loop is the new management loop

There’s a second half to editing that delegation never had, and it’s the part that compounds.

When you assign a task to a person and they do it poorly, your options are slow and awkward. You coach, you wait, you hope the lesson takes by the next sprint. Feedback to humans is high-latency and emotionally expensive, which is why so much bad work gets quietly tolerated, correcting it costs more than living with it.

Feedback to an agent has neither cost. The note you send back isn’t a hard conversation; it’s an input. “You missed the edge case where two things happen at once.” “This violates the brand voice, too breezy for a security advisory.” “Cut it in half.” The agent doesn’t get defensive, doesn’t need a week, and, this is the part that changes the math, the note can become permanent. A correction you give once can be folded into how the work gets done from then on, so the same mistake doesn’t come back next Tuesday.

That’s the loop that replaces the standup. Not assign-track-chase, but review, send the note, watch the next draft get better. The editor’s feedback isn’t the end of one task; it’s the calibration of all the next ones.

This is why “judgment-and-feedback” beats “delegation-and-tracking” as the core skill. Delegation scales linearly, more work means more assigning, more tracking, more chasing, until you hit the ceiling of your own attention. Feedback scales differently. A good note doesn’t just fix one draft; it raises the floor on every draft after it. The editor who gives the right correction once is buying down a whole category of future errors. That’s leverage a delegator never had, because a human you corrected on Monday might still forget by Friday.

Why this is harder, not easier

If the agents do the producing and you do the reviewing, it sounds like the manager’s job got smaller. It didn’t. It got narrower and heavier.

Here’s the trap. When production was the bottleneck, a manager could hide. You could be busy, genuinely, exhaustingly busy, assigning, tracking, chasing, re-planning, and never once render a hard judgment about whether the work was actually good. The logistics were a place to put your effort that felt like the job and let you skip the part that was actually hard. Plenty of careers were built entirely in that hiding place.

Editing has no hiding place. When the only thing left on your desk is the judgment, you can’t be busy instead of right. Every draft in the queue is a decision only you can make, and the quality of the company becomes, very directly, the quality of your read. There’s no standup to run out the clock. There’s a draft, and an opinion you owe, and it’s yours.

That’s the uncomfortable truth under the flip. We told ourselves judgment was the small part of management because we never had to do much of it, the logistics ate the day and the taste got the leftovers. Take the logistics away and what’s exposed is how thin our taste actually was. A lot of managers were good schedulers who never had to be good editors. The agents just removed the excuse.

The naive read of “AI does the work now” is that it deskills the manager. The real story is the opposite. It strips out the part of the job anyone could do, the coordination, and leaves the part that was always the actual job and the actual differentiator: knowing good from almost-good, and being able to say why.

The turn: taste was always the job

Strip away the agents and what’s left is something every great editor, every great creative director, every founder with real taste already knew. The work was never the hard part. Deciding what’s worth shipping was.

For most of management history we couldn’t act on that, because we were drowning in coordination. The person with the best judgment in the building spent their mornings chasing status updates and their afternoons defending calendars, and got to the actual judgment tired, late, and in the margins. We called that management. It was mostly logistics with a sliver of the real thing on top.

When your team is agents, the sliver becomes the whole job. You stop assigning tasks and start reviewing drafts, and the skill that’s left, judgment-and-feedback, is the editor’s skill: read what came back, see the one note that lifts it, decide what’s good enough to ship. It’s harder than the job it replaces, because there’s nowhere left to hide behind being busy. But it’s also the only part of management that was ever really yours.

The manager who’s worried they’ll have nothing to do has it backwards. For the first time, they’ll get to do the part they were always best at, and never had time for.

---

That’s the shift we’re building Apollo around: an operating system where agents do the producing and the most capable person in the company finally gets to spend their day on judgment instead of coordination. If you’ve ever wished you could fire the logistics and keep the part where you actually decide what’s good, that’s the job that’s coming. It looks a lot like editing.