The org chart with entries that aren't people

Mara runs talent at a 900-person software company. In March, preparing for her annual planning offsite, she did something she described to me as half exercise, half provocation: she drew her org chart as it actually operates, rather than as the HR system renders it.

The HR system shows eleven people. Her drawing showed eleven people and nine other entries. A sourcing agent that reads every open req and builds ranked shortlists overnight. A screening agent that scores candidates against requirements and writes its reasoning down. An engagement agent that runs candidate communication and scheduling across the whole pipeline. A scorecard agent, an analytics agent that posts the Monday digest, an onboarding assistant, and a few narrower ones. None of them appear in any system of record as workers. All of them do work that, three years ago, was done by people she managed, and the work they do is reviewed, corrected, and directed by the people she manages now.

Her provocation to her leadership team was a single question written under the drawing: if these things do work, receive feedback, improve or fail based on how we manage them, and consume budget, in what sense are they not on the org chart?

I want to take her question seriously, because I think it's the question every talent leader will be living inside within a couple of years, and most of the management canon has nothing to say about it.

The delegation you already made

Start with an uncomfortable observation: if your recruiting function has adopted any of the current generation of agents, you have already delegated judgment, whether or not you framed it that way.

A ranked shortlist is a judgment. So is a drafted scorecard, a personalized outreach message, and a decision about which candidate gets a nudge today versus tomorrow. The comforting story leaders tell themselves is that agents handle "the busywork" while humans keep "the decisions," and the story is directionally right but lazy at the boundary. A shortlist that determines which forty of two thousand people a human ever looks at is not busywork. It's the most consequential filter in your funnel, and something is exercising discretion inside it.

The leaders getting this right are not the ones who deny the delegation. They're the ones who treat it the way they'd treat delegation to a new senior hire: deliberately, with explicit boundaries, and with review proportional to stakes. Mara's team maintains what is effectively a delegation charter, a living document that says, plainly, what each agent may do on its own, what it may do subject to human sign-off, and what it may never do. Draft outreach: freely. Send outreach: only after a named recruiter approves. Advance a candidate: never; humans advance, agents recommend. Answer a candidate's factual question at 11pm: yes. Answer a question about visa sponsorship: no, escalate to the human who owns it, and say so honestly to the candidate.

Nothing in that charter is technically sophisticated. It's management, applied to colleagues who happen to be software. The failure mode she guards against isn't rogue automation. It's drift, the slow slide where a recommendation starts getting rubber-stamped because it's been right forty times in a row, and the human sign-off decays into ceremony. Which is precisely how she'd worry about any strong senior hire under a checked-out manager.

Feedback is the job now

Here's the shift Mara says took her team longest to internalize: these colleagues improve or stagnate based on the quality of feedback humans give them, which makes giving that feedback a core productivity lever of the department, not an act of tidiness.

When a recruiter overrides a ranking and writes two sentences about why, that reasoning shapes future shortlists. When a hiring manager corrects a scorecard draft's confusion of confidence with competence, the correction propagates. Multiply across a year and you get compounding: the sourcing that gets steadily better at your company's actual bar, the outreach that stops making the mistake it made in February. Teams that write thoughtful overrides get pipelines tuned to their judgment. Teams that silently fix outputs and move on get, permanently, the median.

Mara made this concrete in a way I haven't seen elsewhere: override quality shows up in her recruiters' performance reviews. Not override volume, which would be gameable and stupid, but whether a recruiter's corrections teach. Her best recruiter this year is not her fastest closer. It's the woman whose written pushbacks on rankings were so consistently sharp that the shortlists for her job families became measurably better than everyone else's, an improvement the whole team then inherited. In the old world, that woman's judgment lived in her head and would have left in her notice period. Now it's load-bearing infrastructure.

The team's cadence changed to fit. Alongside the usual pipeline reviews, Mara's staff run a short weekly ritual they've taken to calling the correction review: fifteen minutes where the recruiters surface the overrides that mattered that week, argue briefly about whether the underlying pattern is one-off or systemic, and decide which corrections should become standing guidance. It looks, from the outside, exactly like a manager reviewing a junior employee's week and deciding what to coach. That resemblance isn't a coincidence and isn't a metaphor. It's the same activity. The difference is that coaching a human improves one colleague, while a systemic correction here improves every shortlist the department will see from that day forward, which changes the return on the fifteen minutes rather dramatically.

This inverts a piece of conventional wisdom worth naming. The fear was that working alongside agents would deskill recruiters. In the well-run version, it does the opposite: it takes the senior judgment that was always the scarce ingredient and gives it distribution. The junior recruiter's shortlists arrive pre-shaped by the senior recruiter's accumulated corrections. Apprenticeship used to require adjacency. Now it partially ships with the tools.

Someone has to own confidently wrong

Every talent leader running a blended team eventually collides with the same incident, in one costume or another. Mara's version: for about three weeks, the screening agent's reasoning on one job family drifted, overweighting a certification that the team had stopped caring about a year earlier but had never removed from an old requirements doc. Forty-odd candidates were ranked lower than they should have been. The drift was caught by a recruiter who noticed the pattern, not by any process designed to catch it.

The lesson she took wasn't "the agents can't be trusted." Humans misjudge candidates continuously; that's not news. The lesson was that agent error has a different shape than human error, and her org had no muscle for that shape. A human screener having a bad month is erratic, visibly tired, complained about. An agent miscalibrated by a stale document is consistent, articulate, and wrong in the same direction hundreds of times, with a well-written paragraph justifying every instance. Confidence and correctness fully decoupled, at scale.

So she created a role. One senior person on her team now spends a meaningful slice of each week auditing agent judgment: sampling rankings and re-evaluating them cold, reading rejected-candidate reasoning for drift, checking that escalation rules are actually firing, and running the quarterly review where the team argues about what the agents have been getting subtly wrong. When she describes this role to peers, about half of them hear it as overhead. She's blunt in response: you already pay this cost for humans, it's called management, and the fact that nobody assigned it for the agents doesn't mean the need went away. It means the need was going unmet.

If you take one operational idea from this essay, take that one. The question "who manages the agents" has to have a name attached, or the answer is no one, and no one is a terrible manager.

The question of who manages the agents has to have a name attached, or the answer is no one, and no one is a terrible manager.

Capacity is not headcount, and your budget knows it before you do

The planning conversation is where the org-chart question stops being philosophical. Mara's finance partner asked her the standard question in the standard form: how many recruiters per hundred open reqs? And she found the ratio had quietly stopped meaning anything. Her team of eleven carries what fourteen carried two years ago, but the honest unit of capacity is no longer a person. It's a person plus the agents around them, and that multiple varies with how good the person is at directing the agents.

This has second-order effects leaders should get ahead of rather than discover. Seniority mix inverts: she needs fewer hands and more judgment per hand, which changes who she hires into talent teams and what the junior pathway even looks like. That last one worries her, and I think the worry is legitimate and mostly unanswered across the industry: the traditional apprenticeship, where juniors learned the craft by doing the mechanical volume, is evaporating, because the mechanical volume is what got delegated. Someone has to design what replaces it. The companies that don't will look up in five years and find no bench.

And the honest sentence about the arithmetic, which vendor keynotes omit and leaders shouldn't: the capacity gain is real, and some of it arrives as fewer human seats. Mara didn't lay anyone off; she grew into the capacity. Plenty of companies will make the other choice. A leader who pretends the question isn't on the table forfeits the chance to shape how their own company answers it.

What refuses to be delegated

It would be easy to read all this as a story about a shrinking human remit, so let me be precise about what Mara's people spend their reclaimed hours on, because the list is the actual point.

They close. They sit with hiring managers and argue about what a role really requires, which is the conversation that determines everything downstream and was always the first thing sacrificed to busywork. They call rejected finalists and tell them the truth kindly. They notice the thing no transcript contains, the hesitation in a candidate's voice when relocation comes up, and they act on it. They exercise the judgment about people that the entire apparatus exists to serve.

None of that got smaller. All of it got a larger share of the week. The apparatus took the reading, the remembering, the reconciling, the relentless follow-through, and handed back time, which the humans spent on humans.

Mara never did resolve the org-chart question at her offsite, not formally. But she told me the exercise changed one thing permanently: she stopped asking her leadership team "should we adopt these tools," which was the question of 2024, and started asking the question that actually determines outcomes now. Not whether the agents join the team. They already have. The question on the table at her next offsite, and it belongs on yours, is simpler and harder: are we managing them, or merely running them?