Engineering Manager Span Of Control In The Age Of AI

Engineering manager span of control is becoming one of the most important and least discussed constraints in software delivery. Gallup reports that the average number of direct reports per manager rose from 10.9 in 2024 to 12.1 in 2025, while the median team size stayed around five to six people. In plain English, most managers are still operating at a human scale, but a growing minority are being asked to manage teams that are simply too large for real leadership. (Gallup.com)

See my article: The Role of Technical Management and Mentorship in Your Career. This is critically important!

A lot of companies looked at AI coding gains, decided they could compress management overhead, and called it efficiency. I think that is the wrong lesson. AI can help people produce more code. It does not eliminate the need for coaching, prioritization, judgment, feedback, and trust. Those are exactly the things that start to break when span gets too wide.

Engineering Manager Span of Control Is Not Just a Cost Metric

Too many leaders treat span of control like a budgeting lever. That is a mistake. Span of control is really a complexity lever.

McKinsey’s research is useful here because it avoids the fake certainty of one universal “right number.” Their ranges depend on the type of managerial work. A player-coach role typically supports 3 to 5 direct reports. A coach role supports 6 to 7. A supervisor role supports 8 to 10. Only highly standardized coordinator roles naturally stretch to 15 or more. Most engineering managers are not coordinators. They are usually operating much closer to the player-coach or coach model.

That matters because engineering management is not mostly about status collection. In healthy teams, engineering managers are developing people, handling ambiguity, reducing friction across functions, and making judgment calls under uncertainty. Once you push that role past a reasonable span, management turns into triage.

Engineering Manager Span of Control and Communication Channels

There is also a simple coordination reality here. Project management literature commonly uses the communication-links formula n(n-1)/2 to show how coordination complexity grows as teams expand. At 6 people, that is 15 potential communication links. At 7, it becomes 21. At 8, it becomes 28. The point is not that every person talks to every other person all day. The point is that coordination overhead rises nonlinearly as teams grow. (Project Management Institute)

Coordination overhead rises nonlinearly as teams grow.

This is one reason smaller teams often feel dramatically more coherent. Around six people, you can still maintain a real shared conversation. Past that point, informal alignment starts getting harder, side channels multiply, and misunderstandings become normal rather than exceptional.

Engineering Manager Span of Control in the Age of AI

AI is making this problem more visible, not less visible.

CircleCI’s 2026 software delivery research, based on analysis of more than 28 million CI workflows, says teams are writing dramatically more code while fewer changes are reaching production. Their data shows main branch success rates fell to 70.8%, the lowest in more than five years, and median recovery time climbed to 72 minutes, up 13% from the prior year. CircleCI’s own framing is blunt: writing code is no longer the constraint; review, validation, integration, and recovery are.

That is the real issue. AI increases change volume. It does not automatically increase a team’s capacity to review that change, integrate it safely, coach less experienced engineers through it, or keep the system understandable. When leaders widen engineering manager span of control because “AI makes everyone more productive,” they are usually optimizing the wrong bottleneck.

Engineering Manager Span of Control and the Coaching Gap

Deloitte makes the more practical point. AI can help managers become better coaches by surfacing performance and collaboration insights, but it is not a replacement for the manager role. Their research explicitly calls out empathy, support, connectivity, and psychological safety as things AI does not replace.

That lines up with what many engineering organizations are about to learn the hard way. Once an engineering manager gets too many direct reports, one-on-ones become rushed, feedback gets delayed, career development goes shallow, and weaker performers drift longer than they should. Your retention numbers usually feel that pain before your velocity dashboard does.

What Good Engineering Manager Span of Control Looks Like

For most engineering organizations, I would be skeptical of any claim that 12+ direct reports is a healthy default. There are exceptions. A very senior team of highly autonomous staff engineers working in a stable environment can tolerate more width. A new product area, a mixed-seniority team, or a team navigating architectural change usually cannot.

A better rule is to stop arguing about management layers in the abstract and start asking harder questions:

Is the engineering manager a player-coach or a coordinator?

If the manager still owns meaningful technical, architectural, hiring, or cross-functional work, wider spans are a bad bet. McKinsey’s research points toward materially smaller spans for those roles.

The Real Lesson

The real lesson is simple: AI can compress some administrative overhead, but it does not repeal human limits.

If anything, engineering manager span of control matters more in the AI era because code generation is getting cheaper while human attention is not. As software organizations accelerate output, the scarce resource is no longer keystrokes. It is management attention, review capacity, architectural judgment, and the ability to keep a team aligned without turning every week into chaos.

That is the span problem nobody wants to talk about. AI may let you run leaner. It does not let you ignore the basic math of communication, coaching, and trust.

Principle: When AI increases code volume, do not assume you can widen management span. Assume you need to protect the human bottlenecks more aggressively.