The traditional chart has a bottom row whose job is to execute tasks somebody above them thought up. If agents do that execution, the row changes, and so does every job above it. The short answer is that there are two different drawings of what happens. In the softer version, the analysts and associates stay and each one supervises agents. In the harder version, the row is removed and the directors and managers who were already there absorb the delegation and the checking. Large advisory firms have published versions of both. What almost nobody has published is the operating detail underneath either one: how many agents one person can actually be accountable for, and what they check the output against. This piece lays out the shape and shows who is saying what.
What does the traditional chart actually assume?
Most org charts still in use assume a specific division of labor. The top sets direction. The middle translates direction into tasks. The bottom executes those tasks. Analysts, associates, coordinators, and junior specialists occupy that bottom row, and the implicit deal is that they trade execution hours for training and a path upward.
The assumption worth naming is that there is nothing below the bottom row. Every task somebody thinks up has to be absorbed by a person, which is why the row exists and why it grows when the work grows. A director who needs more output asks for more headcount, because headcount is the only mechanism available.
What does a post-agent chart look like?
The alternative drawing puts a layer beneath the bottom row for the first time. Once that layer exists, the question becomes what happens to the row above it.
This is the harder of the two versions, and it is drawn deliberately. The individual contributor row is not repopulated with junior people supervising agents. It is removed, and the delegation, the quality checking, and the accountability move up to the directors and managers who were already employed. A director who needs more output no longer asks for headcount, because a second mechanism now exists.
Both drawings imply the same change to the surviving jobs, and it is a large one. Work you delegate is work you did not do yourself, which means you cannot verify it by recognizing your own reasoning. You have to check it against something external: the firm’s standards, its precedents, its prior decisions, its actual data. That requirement does not appear anywhere on an org chart, which is part of why it gets missed.
Who else is drawing this shape?
Several large firms have published versions of it, with meaningfully different emphasis.
Deloitte frames it as an operating model problem and reports that 84 percent of companies have not redesigned jobs to fit AI. Its formulation of the open question is the most precise one published: how many digital agents can a human manage and be accountable for? (Deloitte, 2026). Note that this is a question, not a finding. Nobody has answered it with data.
BCG and MIT Sloan Management Review put it most directly. In work on the agentic enterprise published in November 2025: “If agents coordinate workflows, traditional managerial spans of control increase and the number of hierarchical layers decreases” (MIT SMR and BCG report, November 2025). In the same research, 45 percent of agentic AI leaders said they anticipate a reduction in layers of middle management (BCG and MIT SMR, November 2025). Note where that cut falls. It is the middle of the chart, not the bottom.
Microsoft’s Work Trend Index describes what it calls the Frontier Firm, and its emphasis is on redesign rather than reduction: “The firms pulling ahead are focused on AI absorption rather than just AI adoption, redesigning how work gets done and turning output into insight” (Microsoft Work Trend Index, May 2026). That is a claim about changing the work, not about deleting a row of the chart.
McKinsey approaches it from the opposite direction, as the undoing of something that already happened. “Most companies have added at least one layer to their structure, from the CEO all the way to the front line, over the past decade,” senior partner Alexis Krivkovich said on The McKinsey Podcast, and in some organizations two or three, which she describes as both expensive and slow because more people weigh in before any decision can be made (McKinsey, April 2026). In that telling, agents do not invent a flatter org. They make it possible to reverse a decade of thickening.
Shopify is the clearest operational example rather than a framework. In an internal memo made public in April 2025, CEO Tobi Lütke told teams they must demonstrate why a task cannot be done by AI before requesting additional headcount (CNBC, April 2025). That is the post-agent chart expressed as a budgeting rule rather than a diagram.
Aneesh Raman, LinkedIn’s chief economic opportunity officer, disputes the object rather than the rows in it. “The org chart was built in the industrial age to bring order, predictability, and stability to rapidly growing organizations,” he told Fortune in March 2026. “Companies need to let that go, as it’s going to hold back innovation.” In the same piece he argues executives will have to get comfortable with workers figuring out AI on their own, even when those experiments cut across departments and job descriptions (Fortune, March 2026). Deloitte, BCG, and McKinsey argue about which row changes. Raman’s position is that the chart itself is the constraint.
What do the numbers say about the bottom row?
Less than the diagrams imply, and the honest summary is that the evidence is early and contested.
Two measurements are worth putting side by side, because they describe the bottom row changing in two different ways.
The first is about what entry-level jobs now ask for. PwC’s 2026 Global AI Jobs Barometer found that in the most AI-exposed quarter of occupations, 52 percent of the new skills appearing in US entry-level job postings since 2019 were skills traditionally associated with experienced workers. In the least exposed quarter, the figure was 7 percent (PwC, June 2026). The row is not simply shrinking in that data. It is being asked to start further along.
Share of new entry-level skills that were traditionally experienced-worker skills
The second is about how many of those jobs exist. In “Canaries in the Coal Mine?”, Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen at the Stanford Digital Economy Lab report that employment of workers aged 22 to 25 in AI-exposed occupations “now stands 19% below where it would be had it kept pace with that of their less-exposed peers,” measured on ADP payroll data (Stanford Digital Economy Lab, August 2026). Earlier versions of the paper headlined a model-adjusted estimate instead (13 percent, then 16 percent). By the simpler measure the authors now use, the shortfall was 15 percent on July 2025 data and has widened to 19 percent.
Neither finding settles which drawing is correct. Both are consistent with a bottom row that is getting smaller and harder to enter, and neither tells you whether the firms doing it are producing more.
Who thinks the harder version is wrong?
There is a real counterweight, and it is not a strawman.
The most pointed disagreement comes from the same researcher who produced the most-cited evidence that the bottom row is shrinking. Writing in TIME in January 2026, Erik Brynjolfsson, director of the Stanford Digital Economy Lab, described the endpoint as expansion rather than subtraction: “By 2050, most people will command workforces larger than the biggest multinational corporations of today” (TIME, January 2026). The workforces he means are fleets of agents, not people. Read against his own payroll finding, the two are not in conflict. They describe a transition in which entry-level hiring falls first and supervision spreads to nearly everyone later. That is the softer drawing carried to its limit: not a row of supervisors, but an organization in which almost every role includes delegation.
Gartner surveyed 350 executives at companies with at least $1 billion in revenue that were piloting or using AI agents, automation, or autonomous technology. About 80 percent reported workforce reductions, and, more pointedly, reduction rates were nearly equal among those reporting higher returns and those reporting only modest gains or negative outcomes. Gartner’s own summary is that reductions may create budget room but do not create return (Gartner, May 2026). The survey work was conducted in 2025 and published in May 2026.
Klarna is the most-cited reversal. The company leaned on an AI assistant it said could do the work of 700 customer service agents, then in May 2025 announced a drive to hire human agents again so customers would always have the option of a real person. CEO Sebastian Siemiatkowski told Bloomberg that cost had been “a too predominant evaluation factor” and that “what you end up having is lower quality” (Bloomberg, May 2025).
Read together, the Gartner and Klarna findings describe the same failure mode from different angles. The chart is a claim about capability. The savings are a claim about the present. When a firm books the second before proving the first, it ends up rehiring.
What has nobody settled?
Three things, and they are the reason this is a drawing rather than a plan.
Span of control. Deloitte asks how many agents one human can be accountable for and does not answer it. Neither does anyone else with data. A director supervising three agents and a director supervising thirty are not doing the same job.
What accountable means. In the traditional chart, a manager is accountable for people who can explain their own work. An agent cannot be questioned the same way. The accountability stays with the human and the explanation does not.
What the checking is done against. This is the one with the most practical weight and the least published attention. Verifying delegated work requires a reference: how this firm does it, what it decided before, what it flagged last time. In most organizations that reference is not written down anywhere. It lives in the people in the row being removed.
Frequently asked questions
Does a post-agent org chart mean fewer employees?
Not necessarily, and the published evidence does not support treating it as automatic. Gartner’s survey of 350 executives, fielded in 2025 and published in 2026, found workforce reductions common among large AI deployers but no corresponding improvement in financial returns (Gartner, 2026). The chart describes a structure, not a headcount outcome.
Do the analysts and associates become agent supervisors, or do the roles go away?
That is precisely the unsettled part. The softer version keeps them and changes their work. The harder version removes the row and pushes the work up. Published frameworks exist for both, and no dataset yet distinguishes which is happening more often.
Is the entry-level job disappearing?
The measured picture is narrower than that. Stanford’s payroll analysis finds employment for 22 to 25 year olds in AI-exposed occupations running 19 percent below the trend of less-exposed peers (Stanford Digital Economy Lab, August 2026), and PwC finds that 52 percent of the new skills in entry-level postings for the most AI-exposed occupations are skills traditionally associated with experienced workers, against 7 percent in the least exposed (PwC, June 2026). Fewer openings, and a higher bar on the ones that remain.
What actually changes for a manager?
The job shifts from reviewing work done by people who can explain it toward verifying work produced by systems that cannot. That raises the value of written standards, recorded decisions, and accessible firm history, because those become the reference the checking is done against.
Is any of this settled enough to reorganize around?
No. Deloitte’s finding that 84 percent of companies have not redesigned jobs for AI (Deloitte, 2026) is better read as a description of appropriate caution than as a backlog. The firms moving first on structure are largely doing so without evidence that the structure produces results.