The question that has come up most these past few years was pointed the wrong way from the start. "Will AI take my job?" "Which positions will disappear?" Asked that way, the answer gets framed as a roll call, as if change happens one box at a time: keep this box, cross out that one, and leave the rest as they were.
What's actually moving is bigger than that. What's changing isn't the names in the boxes. It's how the chart itself gets drawn.
A company's org chart is really a division table. It records the last generation's answers: how many pieces a job should be split into, who does each piece, who reports to whom, and who has the final say when things get stuck. Those answers didn't fall from the sky. They grew out of the tools of their time: how long it took information to travel, how many people a document had to pass through, how many years of experience it took before someone dared make a call. The tools have changed, so the answers should be recalculated. Most companies' charts haven't moved.
So the question isn't "who gets replaced." It's "how should this work be divided now."
Our July pieces were about diagnosis: why your AI never got moving. August was about construction: once you've decided to move, how to actually build it. September's series was about operations: once it's built, how to run it, how to prove it works, how to keep it from becoming a liability. At the end of those three series, what you have to face is this chart.
01The "replacement" question is already out of date
Let's put the numbers on the table first, because anxiety usually comes from not looking at them.
Research from Indeed Hiring Lab reached a fairly restrained conclusion: only about 1% of skills are fully replaced by AI. But the same research also points out that the bar for entry-level positions has gone up, and the chances of getting in have actually narrowed. You only get the full picture when you put those two together. What gets taken first is the part of a position that "used to be where new hires learned the ropes." Once it's pulled out, the position still exists, but it's no longer the same position.
The layoff numbers look like a different story. The tech industry cut 78,557 people in the first quarter of 2026, with Oracle and Amazon accounting for half of them. But the same companies, while cutting individual contributors, were hiring AI engineers and product roles in large numbers, something the industry calls "cutting while hiring." Interestingly, very few companies in this wave explicitly blamed AI. Babak Hodjat, chief AI officer at RationalFX, put it plainly: most layoffs are "expectation-driven," not a sign that AI capacity has already paid off; it will take another six to twelve months before any real productivity gains show up.
Even in the cases that look most like "AI replacing people," what actually happened was a re-division: the old way of dividing the work was torn down, the new one was still being drawn, and the gap in between got recorded as a layoff number.
For a company of twenty or fifty people, this takes a different shape, but it's the same thing. You won't wake up one day to find "the sales assistant position has disappeared." You'll notice something small first: the weekly report that used to take an afternoon to put together now takes forty minutes. Then a second thing, and a third. Once enough of them add up, you're forced to ask a bigger question. If thirty percent of this position's time used to go to those things, what should the remaining seventy percent grow into?
At that moment you're not cutting people. You're redrawing a box.
02Three lines on the org chart are being redrawn
Everything so far has been about shape. To make it concrete, there are three specific lines being redrawn.
The first: department boundaries.
Traditional department boundaries were drawn by "who can do what." Marketing can write copy, design can make graphics, sales can close deals, so work gets passed back and forth among three departments, and every handoff loses something along the way. When the cost of producing copy and producing graphics both drop close to zero, the reason for that boundary weakens too. The question left isn't "who can do it" anymore. It's "who is responsible for how this turns out."
This extends what we discussed a while back: "roles shouldn't be cut per person anymore." That piece was about how to re-divide the inside of a single position; this one is about whether the walls between positions should stay. Walls have a cost: every boundary means a handoff, a round of back-and-forth confirmation, an argument over "whose job is this, anyway." Once the tools get cheap, keeping up a low-value wall becomes one of the most expensive line items in the whole company.
The second: reporting paths.
Most middle management does information relay: summarizing what's happening below and passing it up, translating decisions from above and passing them down. The role exists because information itself has friction: collecting it takes time, organizing it takes time, translating it takes time. When most of that friction gets squeezed out, the value of the relay station shifts from "passing things along" to "making judgment calls."
Some companies in Silicon Valley have already pushed this to the extreme. One AI-native company of twenty-five people, with fewer than ten engineers, has 99% of its code written by AI, and has compressed what used to be a six-week development cycle into a single day. Write a feature at 10 a.m., A/B test it at noon, cut part of it based on the data at 3 p.m., rewrite a better version by 5. The most radical thing they did was dismantle the product manager role entirely and spread its responsibilities back across every engineer and engineering manager. Their reasoning was blunt: a lot of the cost of back-and-forth confirmation was happening in that role to begin with; take it out, and that cost actually goes down.
Of course this case can't be copied as is. It's a startup that grew from zero, with no twenty-year-old legacy systems and no existing web of personal obligations, so its conditions are a long way from a trading company or design firm that's been around for fifteen years. What you can copy is the way it asks the question: open your org chart, find the layer whose reason for existing is "information travels slowly," then think about what that layer should be doing now.
The third: who decides what.
This is the one people talk about least, yet it matters most. Once a task is handed to AI, decision-making authority quietly moves: it used to be that a senior employee's experience had the final say, and now some rule that nobody wrote down has the final say. Most companies have never made a single explicit decision about this. It just happened, silently.
One review of where enterprise AI adoption gets stuck lays out five root causes, and one of them is called "a shortage of knowledge assets": many companies run on people and experience, lack complete documentation and SOPs, and leave AI with no raw material to work from. Another source puts it more precisely: what companies lack isn't more data, it's Business Context. Data records outcomes; Business Context records judgment, meaning the basis for the rules, the experience of the business, and the current state of the workflow.
If decision-making authority isn't written into the chart, it will flow to wherever nobody is watching.
03The new boxes are already on other people's charts
If redrawing sounds too abstract, look at the shapes others have already drawn.
Meta's developer tools division launched an AI Builder program at the scale of a thousand people, and three new roles grew out of it internally:
| Role | What it does |
|---|---|
| AI Builder | An individual who uses AI tools to prototype features quickly |
| Pod Lead | Coordinates a small group of two to five people plus AI tools, and integrates the output |
| Org Lead | Designs AI workflows, sets the boundary between people and machines, and is accountable for organizational results |
What's worth noticing isn't the three names; big companies invent job titles all the time. It's that the axis this table is divided along has changed. It isn't divided by specialty (front end, back end, design). It's divided by "how large a combination of people and machines you're responsible for." The first tier is responsible for itself, the second for a small group plus tools, and the third for "where the line between people and machines gets drawn."
The third role deserves a three-second pause. In plain terms, the Org Lead's job is "deciding which work goes to people and which goes to machines, then taking responsibility for the consequences of that decision." In most small and mid-sized businesses, nobody is doing this job right now. It's scattered across each manager's instincts, with no name and no one held accountable.
And that is exactly why "we adopted AI but it didn't work" keeps happening. One study summed up the five biggest roadblocks on the enterprise side: vague requirements, poor data quality, compute costs, system integration, and no one driving it. Among its points: "the hard part is integration, not the interface," and "real adoption isn't going live, it's continued use." Translate "no one driving it" into the language of the org chart, and it means there is no box on that chart that says whose job this is.
04How fast should the chart be redrawn
This next part does nothing for our business, but leaving it out would be dishonest. On how fast the org chart should be redrawn, the different camps don't agree.
The aggressive camp · AI-first organizations. That approach is to dismantle roles outright, let the system lead, and move people back to the position of "reviewing whether the results were good or bad." They believe the hardest step to cross isn't technical, it's getting every employee to trust AI-led decisions. The same interview also points out that junior engineers transition more easily than senior ones, because the ten to twenty years of expertise senior engineers hold is losing value fast.
The conservative camp · capacity hasn't paid off yet. Babak Hodjat argues the opposite: most organizational changes made in the name of AI are "expectation-driven," and real productivity gains won't show up for another six to twelve months. Redrawing the org chart based on expectations that haven't paid off is betting real staffing costs on a bill that hasn't arrived.
The middle ground · the data. Indeed's numbers side with the conservatives: only about 1% of skills are actually fully replaced. But the higher bar for entry-level positions shows that re-division really is happening, just inside positions, not between them.
The three views look like they're arguing, but they're describing the same thing at different speeds. For a small or mid-sized business, the conclusion you can draw is actually easier to handle than you'd think: redraw how the work is divided first, then redraw the boxes for people.
The order can't be reversed. Changing staffing first and thinking about workflows later is the most expensive way to do it: you pay severance and get back a new workflow nobody thought through, and six months later you have to change it again. Work out the division first, and people's places often surface on their own. Which boxes should merge, which should split, and which one was never drawn in the first place.
There's also some good news, which we've covered before: this is easier for small companies than for large ones. Large enterprises have more compliance, more legacy systems, more people factors, and the transition is hard; smaller companies are actually standing at the very start of the payoff window. Fewer boxes on the org chart means you can actually move them when it's time to redraw.
05The extreme case: a one-desk company
There's a thought experiment that pushes this to the point where you can see it most clearly.
Shrink a company down as far as it goes, until there's only one desk left. Now the org chart has a single box, but the work hasn't shrunk. It's all piled into that one box. There's a whole field that studies how this state should work, and its core formula is roughly: capability × tools × systems. Positioning, content, customer acquisition, delivery, review: every stage needs a workflow that can actually run, rather than relying on the person to remember.
This isn't telling you to downsize. The point of the extreme case is that it forces the variables into the open: in that state, how much revenue it can carry depends entirely on how many judgment calls have already been written into a system, so they don't have to be rethought every time. Every box on an org chart faces the same question. It's just diluted by headcount, so you can't see it.
Put back into a twenty-person company, the question gets very concrete: of these twenty boxes, how many have work that is "already written down, so someone else could take it over"? For the boxes where you can't answer, the problem isn't the people in them. It's that the chart was never asked to account for this.
Three series in, what's left is this chart
Back to where we started.
Before An org chart showed "who we have, and who manages whom"
Now It has to answer three more questions: how should this work be divided, which pieces go to machines, and who takes responsibility for the results
Diagnosis in July, construction in August, operations in September. Looking back, every piece was preparing one corner of this chart. Asking "which step are you on right now" was about confirming how far the chart can be redrawn; if the foundation isn't laid, redrawing is wasted effort. Talking about "roles shouldn't be cut per person anymore" was about taking one box apart. Talking about "the model is rented, so what do you actually own" was about the four things that stay after the redraw: permission boundaries, acceptance criteria, memory that doesn't walk out the door, and the orchestration of who does what first.
All four of those are part of the org chart. They just aren't drawn on paper.
So an AI workflow diagnostic, by the end, usually turns into a rereading of the org chart: how many years ago was this chart drawn, do the reasons behind those divisions still hold, which box has sat empty for a long time with no one filling it. The tool list actually comes later.
We're fortunate to live in this era of computers, the internet, and AI. What this generation has been handed isn't a new list of job titles. It's a chance to divide the work all over again. The last time a chance like this came along was in the years when the internet redrew the channels.
One team, from strategy to launch. An org chart won't redraw itself. Someone has to sit down and question it again, box by box.
Let AI move fast. Let people hold the anchor.
Rooted in craft. Built for the new wild.