Fewer Managers, More Managing
Most of what managers do all day is tracking, and an agent wired into Slack, email, and the CRM now does it better. The firms getting this right stop refilling management seats and spend the recovered hours on coaching, hard conversations, and new skills.
Most of what your managers do all day is tracking, and an agent wired into Slack, email, and the CRM now does it better. The shift is already in the payroll data: across 8,500 small companies, individual contributors per manager roughly doubled between 2019 and 2024.
The manager job has two parts. Tracking means status, reports, chasing updates, and relaying information between levels. Judgment means hard conversations, coaching, and deciding who is ready for what. AI takes the tracking work first. The firms getting this right spend the recovered hours on the judgment work.
A founder I know hit this last year. Revenue was healthy, clients were renewing, and the company had quietly become top-heavy. Before he reorganized anything, I told him to connect an agent to Slack, Gmail, the CRM, and GitHub and let it handle the tracking. He had automated half his client delivery but none of his management overhead. Most owners have.
How managers actually spend the day
Start by measuring the job. Asana's research across 10,000+ knowledge workers found that 60% of a person's time at work is spent on "work about work", including 352 hours a year just talking about work. Microsoft's data says workers are interrupted every two minutes and juggle 117 emails and 153 chat messages a day. Managers sit between other people's work, so they carry more of this load.
The load shows up in what managers can handle. Gartner data puts manager responsibilities at 51% more than they can effectively handle, and 1 in 5 managers would leave people management entirely if they had the option.
When AI shows up, the coordination work shrinks first. A Harvard Business School study of 50,032 developers using GitHub Copilot found project management fell 10% as a share of all activity, while core work rose. The tool absorbed the tracking around the work before it changed the work itself.
I have watched the same shift up close in service firms. Supervisors who spent 2 hours a day reading data and preparing reports now sometimes spend less than 10 minutes validating AI output. The report still exists. Nobody compiles it anymore.
Operational teams get to business fast, see the benefit quickly, and explore the tool heavily; managers and supervisors take considerably longer and often need dedicated training time. This is pulled from franchise delivery data, not an estimate. The cause is structural: managers are not less capable or more resistant, and their tracking work is what gets automated first, so they feel the change earlier and harder than anyone else in the firm.
When AI absorbs the tracking work, managers are left without a written process for the judgment half of the job: feedback, one-on-ones, mentoring, and motivating. The transition needs deliberate attention from leadership, and this pattern is observed across franchise delivery data, not a measured outcome. The firm documented the reporting and never documented the coaching. Once the reports disappear, managers have no shared process to run the judgment work.
| Tracking layer | Judgment layer | |
|---|---|---|
| What it covers | Status, reports, chasing updates, relaying between levels | Hard conversations, coaching, deciding who is ready for what |
| Who should own it | An agent wired into Slack, email, and the CRM | A manager with recovered hours |
| What happens if nobody does | Projects drift quietly | People drift quietly |
AI takes the tracking work
This is running today at ordinary companies, with ordinary tools. Off-the-shelf products already write status updates from your project data. An agent wired into the tools you already use can go further. It watches Slack, email, the CRM, and messages. It knows which projects have gone quiet, which tasks are overdue, which client has not been contacted in three weeks, and it reports all of it without asking anyone to fill in a form.
GitLab's CEO described the principle plainly in his Act 2 memo: "If an agent can do it, we automate it, and find things where our judgement or skill is essential." The same memo makes the org-chart point: "Every layer of management increases the number of places where priorities and communication gets filtered."
In firms I have watched, these systems go past reporting: the AI flags projects with no recent client contact, overdue tasks, and consultant interactions that look weak, then warns the manager before the situation becomes churn. The managers those systems serve did not disappear. They stopped compiling reports and started improving delivery processes, team performance, and client outcomes.
My own operation is the small-scale version. Compared with mid-2025, about 90% of my software and publishing work is automated: feature exploration, bug fixing, log analysis, and research. My agents have dozens of command-line tools to interact with my production systems. My rule is simple: any task I catch myself doing twice, the agent gets a script or a skill for it. What remains is writing, thinking, and understanding how systems should work and how to help more people in this AI-native transition.
That rule scales down to any firm. The starting point is a list of the repetitive questions your managers answer every week, no platform required.
Fewer managers, one unfilled seat at a time
The change in reporting lines is already measurable at small-business scale. Gusto's payroll data across 8,500 small companies shows individual contributors per manager roughly doubled, from about 3 in 2019 to about 6 in 2024. Manager hiring is down 40 to 50% since 2022, while hiring for individual contributors fell only about 11%. The share of workers in a people-manager role dropped 34%.
Gartner predicts that through 2026, 20% of organizations will use AI to flatten their organizational structure, eliminating more than half of current middle management positions. Deloitte counted 42% fewer middle-management job postings at the end of 2024 than in spring 2022.
Big tech got the headlines for doing this loudly. Amazon mandated a 15% increase in the ratio of individual contributors to managers. Google cut 35% of its managers of small teams. Coinbase replaced pure managers with "player-coaches" who manage and ship. Those are layoff stories. They are the wrong model for a firm your size.
What I see at service firms is quieter. A manager leaves, and the position never gets refilled because AI absorbed enough of the role that nobody needs the seat. A supervisor gets promoted into what remains, usually a younger one, more AI-fluent, and takes over the part of the manager's job that is left. The number of people stays the same while reporting levels shrink. The change happens without a single layoff, one unfilled seat at a time.
The number of people did not shrink. The constraint moved to skills instead.
More time for people
Would fewer managers and an agent handling coordination make a company colder? In the firms I watched, the opposite happened. I did not expect it either.
In the firms where I watched the 2-hours-to-10-minutes shift, recovered time went to training, developing, and motivating teams. AI is changing company morale and helping teams operate in a more connected way. Other owners tell me similar stories: less fetching data, less copy-pasting, and more time making sure people are working on what moves the business.
The management research explains why. Gallup's long-run finding is that managers account for about 70% of the variance in team engagement. And in the AI era the effect compounds: employees whose manager actively supports the team's use of AI are 8.7 times as likely to say AI transformed how work gets done in their organization.
Managers have a large effect on how a team feels and performs, but tracking work consumed the hours they needed for that job. Managers spent their coaching time on spreadsheets. The agent gives that time back.
I wrote before about delegating the inputs and owning the outputs. The same idea applies to management. An agent handles the tracking inputs. A manager owns the judgment about people, which remains human. On the Delivery Model Ladder, this is the Stage 2 to Stage 3 move: agents take the tracking layer, and the org chart changes shape instead of just moving faster.
Where it goes wrong
Freed time disappears without a plan. BCG found that 42% of frontline AI users save at least a workday per week, and 66% receive limited or no guidance on what to do with the saved time. A manager with a free afternoon and no instruction fills it with more tracking or with nothing. I tell owners to have the AI watch this too: recovered hours are an asset, and the agent should report where they went, just as it reports overdue tasks. The extra time AI buys you is already gone unless you decide where it goes.
Companies sometimes remove the management layer while leaving the human work in place. Korn Ferry found 41% of employees say their organization slashed management layers, and 37% say the lack of managers leaves them feeling directionless. Zappos ran the extreme version of this experiment before AI existed as an excuse, deleting its hierarchy outright, and about 18% of the workforce left by January 2016. Removing a level from the org chart does not remove the coaching, escalation, and development that came with it. Someone still has to do that work, with real hours attached.
AI is added without removing the tracking duty. Harvard Business Review documented consulting-firm managers drowning in AI adoption: expected to validate AI output, coach the team on AI skills, and keep quality up, on top of an unchanged job. The order matters. The agent takes the tracking first. After the tracking goes, managers have time for the new responsibilities.
People learn with other people. We want interaction, motivation, and validation. Remove that contact and work becomes a chore, and growth stops.
There is a firm size where none of this applies. If nobody in your company is a full-time manager, because you are still holding the coordination yourself, there is no tracking layer to hand over. An agent will still save you hours. It will not change your org chart, because you do not have one to change. This argument starts to bite once you have people whose main output is other people's status.
Use the hours to build new skills
Use the recovered time to prepare your team to run an AI-native firm. That is a management job now. Supervisors need to learn to run agents, validate output, and build the systems I described above. The managers I have watched make this transition stopped producing reports and started producing operators.
The firms I see flattening do not need more people. They need people with skills that barely existed in 2024, and those people are hard to hire. The old structure produced those skills as a side effect: juniors watched seniors, learned the trade, and moved up. Gartner's own flattening prediction carries a warning about exactly this: mentoring pathways break and junior development suffers. When the structure no longer grows your people by accident, mentoring becomes a line item, and the freed management hours are where it comes from.
Before you hire the next manager or coordinator, wire in the agent and see what work remains. That remaining work is the real role. It is smaller than the job description you were about to post, and it is a people job. Sometimes it justifies a hire. Often it justifies a promotion instead.
What to do on Monday
Have someone list every report a manager in your firm compiles by hand, along with every status question that gets asked in Slack more than once a week. That list is an agent's job description. The hours it frees are your mentoring budget. The agent should track where they go because unrouted time disappears.
Then write down the judgment half. Include what a good one-on-one covers, how feedback gets given, and what a manager should notice about a person and when. That document does not exist in most firms. It is what managers need once the tracking is gone.
Your org chart has a tracking layer and a judgment layer. Only one of them ever needed a person.
The questions owners ask me
Will AI replace middle managers? It replaces the tracking part of the job: status, reports, chasing updates. The judgment part, hard conversations, coaching, development, gets more room, and that is the part Gallup ties to about 70% of the variance in team engagement.
What does an AI-first org chart look like in a service firm? Fewer layers and wider spans. An agent handles coordination, supervisors work in the delivery while developing people, and management seats that empty out stay empty. Headcount stays; layers go.
Where should the freed management time go? Into training, mentorship, and the skills an AI-native firm needs: running agents, validating output, building delivery systems. Route it on purpose. In BCG's June 2026 survey, 66% of AI users got no guidance on what to do with the time they saved, and unrouted time disappears.
If you run a service firm and you are working through this shift, I write about it every week. The newsletter is free, and replies land in my inbox.
