Span of Control Is Dead

The lie

The number was never about people. It was about decisions.

Seven. That was the number. A manager could oversee roughly seven people before the org chart started lying to you.

It was never magic. It was a proxy. Seven humans meant seven streams of work you could still hold in your head — the bugs, the designs, the politics, the quiet person who stopped talking in standup. Span of control was a span of attention, dressed up as headcount.

Then agents showed up, and every one of those humans started supervising a small factory. Same seven faces in the 1:1 calendar. Ten times the output behind each face. The org chart looks unchanged. The load is not.

You didn’t get more hours. You got more judgment surfaces than you can see.

The trap

You’re still counting heads. The work stopped living in heads.

The comfortable move is to keep the ratio and celebrate the leverage. Seven engineers, each running agents, shipping like a team of forty. Leadership loves that slide. The dashboard is green. The PRs keep coming.

What collapses first is not velocity. It’s the quality of the yes.

An engineer shipping agent output they half-understand is one failure mode. A manager who can no longer tell which of their seven is half-understanding is the failure mode that scales. You stop catching the soft no, the skipped alternative, the design that only works until Tuesday. You start signing off on summaries of summaries. Span of control still says you’re fine. Span of judgment already left the building.

The old number assumed the work was mostly human, mostly visible, and mostly paced by how fast people type. None of those assumptions survived.

The shift

Count what you actually ration: judgment, not reports.

Span of judgment is how many concurrent streams of consequential decisions you can still coach, challenge, and catch before they harden into production.

Not how many people report to you. How many judgment surfaces you can still see clearly.

A judgment surface is anywhere a human is deciding what an agent’s output means — ship it, cut it, rethink the premise, escalate the risk. Every engineer running agents is a bundle of those surfaces. Some bundles are quiet and high-trust. Some are a firehose of plausible garbage. Treating them as equal “direct reports” is how you get blindsided by the polite ones.

Four things shrink your span faster than headcount ever did:

Opacity. Agent work hides inside sessions you’ll never open. If your only signal is the PR description, you’re managing press releases.

Novelty. Greenfield and ambiguous problem spaces burn judgment calories. Familiar CRUD with a good test harness barely does. Same person, wildly different load.

Blast radius. A bad call on a toy feature is cheap. A bad call on auth, money, or the migration everyone is depending on is not. Weight the surfaces, don’t average them.

Coaching debt. New managers, new domains, people still learning to supervise agents — they need more of your judgment per unit work. Counting them as “one” is fiction.

If you ignore those and cling to seven, you will eventually manage a team that looks fully staffed and somehow still can’t tell good work from fast work.

The test

You’ll know the number when the answers get worse.

You don’t need a consultant to find your span of judgment. You need an honest week.

Watch for the tells:

• You’re approving things you couldn’t explain to a skeptical peer.
• Your 1:1s are status broadcasts. Nobody brings you a real fork in the road anymore — or they do, and you’re too underwater to sit in it.
• Bad news arrives late, wrapped in a narrative that already picked a winner.
• You’re surprised by production more than once a month, and each surprise had a review you “signed off on.”
• Two of your people could swap projects and you wouldn’t know whose judgment you trust more.

When three of those show up, you’re past the number. Not because you’re weak. Because the unit of management changed and your calendar didn’t.

The fix is rarely “hire another manager” as the first move. Sometimes it’s fewer concurrent bets. Sometimes it’s making judgment visible — ADRs, real review standards, agents that have to show their work. Sometimes it’s admitting one of your seven is actually three jobs wearing one Slack handle.

One last thing

Stop asking how many people. Ask how many decisions you can still see.

Span of control was a useful fiction for a world where humans did the work at human speed. Span of judgment is the metric for a world where humans supervise machines that don’t get tired and don’t know when they’re wrong.

The org that keeps optimizing headcount ratios will look efficient right up until the quality cliff. The org that rations managerial attention against real decision load will look slightly overstaffed on paper and weirdly calm in production.

Seven was never sacred. It was a guess about attention. Update the guess for the actual job: how many humans-supervising-agents can you oversee before your yes stops meaning anything?

Find that number. Protect it. Everything else is headcount cosplay.