On Monday, October 26, 2026, Microsoft's chief people officer, Amy Coleman, sent a memo offering a Voluntary Exit Offer to roughly 60,000 employees. That is about a quarter of the company. Fortune reported that the offer covers people at level 70 and below on participating teams in the US, the UK, Canada and India. Employees have until early December to decide, and those who accept leave on December 31. Coleman wrote that the program "gives eligible employees who may be considering a future outside the company a new choice to make that transition now with support."
I read that sentence a few times. It is kind and carefully written, and it describes something strange. The most capital-hungry software company on earth is in the middle of the largest labor cost transformation of its history, and it has handed the sorting of its workforce to the workforce.
The usual way to cut labor cost is a layoff. Someone draws up a list. The list can be wrong, cruel or political, but it belongs to the company, and somebody signed it. A buyout swaps the list for a door. The company says who may walk through and then waits to see who does. Everyone who chooses to leave is making a private bet about their own value, and they know far more about that value than the company does.
That gap in information sits under every AI and labor cost plan being drawn up this quarter. Executives keep treating it as an HR detail. It is a strategy problem.
The memo, read closely
The details matter, so here they are as reported.
KUOW reported that the heads of each organization decided who qualifies, and that the package is comparable to Microsoft's standard severance, which typically includes base pay for a period based on tenure, stock, health care and career planning. The sales organization, some corporate services such as marketing, and some engineering teams were left out. TechCrunch, citing Bloomberg, named Xbox, Windows, Copilot and Finance among the participating divisions, and said some data center, sales and marketing staff were excluded. Fortune, citing a Reuters source, added AI experiences and HR to the covered list.
So the door is open in the Copilot organization. It is open in Finance and HR, the functions every enterprise software vendor now pitches agents into. It is closed in the places that carry revenue and keep the data centers running.
Microsoft says AI is not driving the program. A spokesperson told Fortune, "We don't expect 60,000 people to walk out the door, and we would never want that," and said the company expects under 5 percent uptake, in line with the 3 to 5 percent typical of such programs. CFO Amy Hood told staff she expects "very few people" to take it.
Say uptake lands right at 5 percent. That is 3,000 people. Microsoft had 223,000 employees as of June 30, so in headcount terms this is a rounding error. In the composition of the workforce it may be the most consequential choice Microsoft makes this year, because the 3,000 will be a self-selected group, and self-selected groups are never random.
A buyout is an auction where only one side sees the bids
Economists have a name for what happens when the better-informed party gets to choose whether to trade. They call it adverse selection, and it is the reason used-car markets are full of lemons.
A buyout runs the used-car market in reverse. The employee knows their outside offer. They know whether a recruiter called last week, whether their skills move between employers, and whether the work they do here could be done anywhere. The company knows a performance rating, a level and a salary band. When the company posts a uniform price for leaving, the people who accept are the ones whose outside price beats it. The people with the best outside prices tend to be the ones the market values most.
This has been measured. Tsuyoshi Tsuru at Hitotsubashi University studied the personnel records of a Japanese machinery maker that ran buyouts in 1993 and 1996. In his 2005 paper in The Economic Review, he found that in the 1996 round, the lower-grade non-managers who took the buyout had significantly higher ratings than the people who stayed. His conclusion was blunt. Under forced separation the company decides who goes. Under a buyout it cannot control who accepts, even with careful incentive design.
Practitioners say the same thing in plainer English. John Frehse, global head of labor strategy at Ankura, told HR Brew on October 29: "Talent will be walking out the door and you'll be paying your highest-performing employees to leave in some instances." He also said something that deserves more attention than it got: "They're actually planning for extreme growth, and that growth does not involve the workforce."
Hold those two quotes next to each other. Companies are planning growth that runs on machines, and they are choosing the exit tool that is worst at deciding which humans stay to run the machines.
AI rewires the outside option
Here is the part that I think is new, and that the 1990s research could not have seen.
In a normal downturn, outside options fall for everyone. Hiring freezes spread, recruiters go quiet, and a buyout looks attractive mainly to people near retirement or people who were already halfway out. The adverse selection is real but muted, because nobody has a great offer.
An AI transition does not lower outside options evenly. It raises them for a narrow band of people and lowers them for everyone else. The person who has spent the past eighteen months learning to direct coding agents, write evaluation suites, wire models into a finance close or debug an agent that misread a ticket now has a skill that every company in October's layoff tables is trying to buy. The person whose work is routine document handling has watched their outside market shrink.
Run a uniform severance offer through that split and you get a predictable sort. The AI-fluent person in the Copilot organization reads the offer, sees three or six months of pay plus a job offer elsewhere at a higher salary, and leaves with what is effectively a signing bonus that you paid. The person whose job is most exposed to automation reads the same offer, sees a shrinking market outside, and stays. The buyout removes the people who could have built your automation and keeps the people your automation was meant to absorb.
I want to be careful here. Nobody outside Microsoft knows who will accept by early December, and Microsoft may have designed eligibility precisely to blunt this effect. Leaving sales and some engineering teams out suggests someone there thought hard about where the door should not open. My claim is structural, about the mechanism, and it applies with much more force to companies copying the tool without Microsoft's cash, data or HR analytics.
The spring cohort is a different warning
Microsoft has run this experiment once already. Earlier in 2026 it offered a voluntary retirement program to long-tenured US employees. TechCrunch, citing Bloomberg, reported that more than 30 percent of the roughly 8,750 eligible workers accepted. Fortune noted that the terms were richer than the October offer.
Thirty percent is a big number for a voluntary program, about six to ten times the typical range Microsoft's own spokesperson cited for the October offer. The people it drew were selected on tenure, and tenure is where the undocumented knowledge lives: why the billing system has that odd exception, which customer gets a call before a price change, which integration fails when a certain vendor ships an update. Every agent deployment I have watched depends on that knowledge being written down before the people who hold it leave. A tenure-based exit program, run before the knowledge is captured, sells it at severance prices.
The capital side of the ledger
Labor cost transformation would be a strange phrase if labor were the only thing changing. It isn't. The labor line is being reshaped to make room for a capital line that is growing faster than anyone planned.
Fortune laid out Microsoft's numbers. In the fiscal year ended June 30, the company spent $115.9 billion on property and equipment, up from $64.6 billion the year before, while its workforce fell by about 5,000. In the quarter ended September 30, capital expenditures including finance leases rose about 30 percent to $45.2 billion, and roughly three quarters went to short-lived assets such as chips and networking gear. Hood expects spending to exceed $50 billion this quarter.
Jay Zagorsky of Boston University put it simply: "Microsoft's announcement is a clear indication they want fewer workers." Joseph Phillips of Seattle University offered a more textured read, describing a company becoming more capital-intensive and noting that "they have been hiring while they have been laying people off."
Phillips's line is the one that matters for anyone running a business. Microsoft is changing its mix of people, cutting in some places and hiring in others, and the hiring side is competing in the same market where its buyout recipients will be looking for jobs. That makes the sorting problem worse. When the whole industry is trying to buy AI-fluent talent at once, the outside offer for that talent rises, and every uniform exit package in the industry becomes a subsidy for poaching.
Harvard Business Review's editor Daniel McGinn framed the Microsoft offer on October 30 as part of a pattern of profitable companies cutting labor costs as they invest heavily in AI. The pattern is real. The question for each of those companies is whether the people they cut are the people they meant to cut.
October was the month the pattern became a number
The Microsoft memo landed in a month that broke records. Challenger, Gray & Christmas counted 153,074 announced US job cuts in October 2026, according to Startup Fortune, the highest October total since 2003 and up 175 percent from October 2025. Cost-cutting was the top stated reason, at 50,437 cuts. AI restructuring accounted for 31,039, the largest AI-attributed monthly total Challenger has recorded this year.
The Associated Press listed the names on October 23. Barclays plans to cut 5,000 jobs, about 5 percent of its workforce, citing "organisational simplification and technology change efficiency driven by AI," with about 60 percent coming from support functions, developers included. Kroger will cut 3,000 corporate roles over three years, and CEO Greg Foran told staff that "Technology is reshaping what's possible." Coinbase's Brian Armstrong said engineers now "ship in days what used to take a team weeks." Jack Dorsey's letter at Block said "Intelligence tools have changed what it means to build and run a company."
Most of these are layoffs, with lists. Microsoft chose the door. HR Brew quoted Frehse on why more companies will follow: "Companies, with reputations to protect, are looking at incentives for employees to exit instead of doing mass layoffs." He added, "It's expensive, but it's the kinder way to reduce headcount."
I agree that it is kinder. Kindness is a fine reason to pick a tool. It is a poor reason to pick a tool without understanding what it selects for.
AI and labor cost: design the sort before you open the door
Every executive I talk to this quarter has a version of the same plan. Agents will absorb some share of routine work in finance, HR, support and internal engineering. Headcount in those functions should fall. The board wants the savings in the next fiscal year. Layoffs are ugly and risky for the brand, so a voluntary program looks attractive.
That plan treats the labor cost transformation as a quantity problem: how many people, how fast, at what price. The harder problem is about composition. The AI-era company needs a different mix of people, and the voluntary mechanism hands the mix to whoever has the best alternatives. Here is how I would design around that.
Map the work before you map the people
Eligibility at Microsoft was set by team. Teams are the wrong unit. Inside a single Finance organization you will find someone whose week is spent reconciling exports by hand and someone who has quietly built the agent pipeline that will replace that reconciliation. Both are in the same team, at the same level, and both are eligible for the same package.
Before any exit program, break the work into tasks and tag each one by its future: absorbed by agents, supervised by humans, or newly created. The people doing tomorrow's tasks today should not be eligible. That is a harder conversation than a team-level cut, and it is the only one that protects the build.
Price the exit by the work, not the tenure
Standard severance pays by tenure. Tenure is close to uncorrelated with exposure to automation, and in a transition it often runs backwards, since long-tenured people hold the undocumented context agents need.
Suppose a company paid exit incentives in proportion to how much of a person's current task mix is slated for automation, and paid nothing extra to people whose skills are in the "newly created" column. The offer then becomes attractive to the people whose roles are going away and unattractive to the people whose roles are being invented. The research on Japanese buyouts suggests design can soften adverse selection without eliminating it. You will still lose some people you wanted. You will lose fewer of them.
Capture before release
No tenure-based program should open until the knowledge in that cohort has been extracted into something an agent can read: runbooks, exception logs, decision histories, recorded walkthroughs of the weird cases. Say you give the cohort a paid ninety-day window to document their work before the exit date, with the final tranche of the package tied to completing it. That window costs less than the first major agent failure caused by a missing exception rule.
Keep a veto, and use it honestly
A management veto on acceptances is the bluntest tool. It also tells the people you veto that they are valued, which is information most companies are bad at giving. Used narrowly, with clear criteria tied to the task map, it closes the worst leaks. Used broadly, it turns a voluntary program into a layoff with extra paperwork, and employees will notice.
Make a counteroffer to the people the market wants
The most direct answer to adverse selection is to raise the inside price for the people whose outside price is rising. If your AI-fluent operators are worth more to Microsoft's hiring side than your severance is worth to them, pay them to stay before the door opens. Companies routinely find money for severance and rarely find it for retention, even though retention is usually cheaper.
What the door teaches
There is an older lesson buried in all this. Every labor cost decision is a decision about who gets to decide. A layoff concentrates that decision in management, with all its blind spots. A buyout spreads it to employees, and each of them uses private information to make a choice that is good for them and often bad for the firm. Neither is neutral. The skill is in knowing which decisions to keep and which to give away.
AI raises the stakes because it splits the workforce into people whose market value is climbing and people whose market value is falling, and it does so within teams, within levels, sometimes within a single desk. A uniform exit price on top of that split is a machine for losing the climbers. The companies that come through this transition with their capability intact will be the ones that refused to let a severance formula design their org chart.
Microsoft can afford to learn this the expensive way. It has $45.2 billion of quarterly capital spending, 223,000 people and decades of HR data. Most companies copying the playbook have none of those cushions, and they are copying the visible part, the memo and the door, without the invisible part, the judgment about where the door goes.
Do not run a voluntary exit program to fund your AI plan until you have mapped the work, priced the exit by exposure, captured the knowledge and retained the people building the replacement. This is architecture. It touches compensation design, workforce analytics, knowledge capture and agent deployment in one sequence, and getting the order wrong costs you the very people who would have made the savings real. Buying another tool will not do this for you. It takes a plan drawn by people who have seen agent deployments fail for want of a single departed employee's knowledge, and who know which hands you want raised.
That is the work Agor AI Advisory does. We map the tasks, design the transition, and build the agent systems that make the new cost structure hold up after the exit date. Schedule a strategic consultation with us today.
Sources
- Fortune, Microsoft's 60,000-person buyout offer hints at a future with fewer workers and more equipment, October 30, 2026
- KUOW, Microsoft offers voluntary buyouts to a quarter of its workforce, October 26, 2026
- TechCrunch, Microsoft offers buyouts to around 60,000 employees, October 26, 2026
- Harvard Business Review, Are Microsoft's Voluntary Buyouts a Sign of What's to Come?, October 30, 2026
- HR Brew, HR risks losing top performers with voluntary buyout programs, October 29, 2026
- Startup Fortune, US employers announced 153,074 job cuts in October, the most since 2003, November 3, 2026
- Associated Press via Spectrum News NY1, Layoffs linked to changes in corporate AI spending, October 23, 2026
- Tsuyoshi Tsuru, Buyout and Adverse Selection in Employment Adjustment, The Economic Review, 2005