Microsoft offered 60,000 people buyouts. Redesign the work before the team.

On 26 October Microsoft offered voluntary buyouts to around 60,000 employees in the United States, Canada, the United Kingdom and India — roughly a quarter of a workforce that stood at about 223,000 at the end of June. It is the second buyout this year; the first, in the spring, went to about 7% of its US staff. Microsoft has been careful to say this is not a layoff: eligible people in participating teams can take a package similar to standard severance, and which teams participate depends on each business unit’s priorities and customer commitments. But nobody reading the coverage missed the context. The company is pouring money into AI infrastructure, and it is reshaping the workforce around that bet.

Most of our clients are not Microsoft. They do not have 60,000 people to offer anything to. But every leadership team we speak to is asking a smaller version of the same question: if AI changes how the work gets done, what does the team look like afterwards? And the way the largest software company in the world is answering it — headcount first, voluntary exits, unit by unit — is about to become the template a lot of boards reach for. We think, for most companies, it is the wrong order of operations.

Cutting the team is not the same as changing the work

A buyout reduces the number of people. It does not, by itself, change a single workflow. The work those people did is still there on Monday — the reports still need writing, the tickets still need triaging, the reconciliations still need doing — and it lands on whoever stayed. If AI has not genuinely taken a share of that work, the result is not a leaner organisation. It is the same organisation with fewer people and more overtime.

Voluntary programmes add a second problem: you do not choose who leaves. The people most likely to accept a package are often the ones with the most options — experienced, well-networked, confident they will land well. They are also, frequently, the people who know where the exceptions live: which client gets special terms, why the month-end process has that odd manual step, what to do when the system says one thing and the contract says another. That knowledge is exactly what you need to automate a process safely, and it walks out the door first.

Microsoft can absorb that. It has the scale, the tooling and the internal AI expertise to rebuild processes after the fact. A 200-person company that copies the move without the redesign underneath it tends to discover, about three months later, that it has cut capacity and kept all the work.

Redesign the work first, then size the team

The order we recommend is the reverse. Start with the work, not the org chart. Map what each team actually does in a week, task by task, and sort it honestly:

• Work an agent can do end to end. High volume, well defined, recoverable when wrong — first-line triage, data entry from documents, routine drafting, reconciliation against clear rules.
• Work an agent can do with a human reviewing. The output is drafted by AI and approved by a person who owns the result. This is usually the biggest bucket, and the one that changes roles most.
• Work that stays human. Judgement calls, relationships, negotiation, anything where being wrong is expensive and hard to undo.

Then build and run the first two buckets for real — not as a pilot on a slide, but in production, with the team using it every day. Only once you can see how much time has genuinely moved do you have an honest basis for any conversation about headcount. Often the answer is not “fewer people”. It is “the same people doing different, more valuable work”, or “we can grow revenue without the hiring plan we had pencilled in”. Sometimes it is a smaller team. But that decision should come from evidence about the work, not from a target set before anyone checked.

Capture the know-how before anyone leaves

The single most useful thing a company can do before any restructuring is to write down how the work is actually done. Not the process document from four years ago — the real one, exceptions included. That written-down know-how does double duty. It is the onboarding material for whoever inherits the work, and it is the instruction set an AI agent needs to do the work properly. Agents do not fail because the model is weak; they fail because nobody told them the rules your best operator carries in their head.

This is most of what an AI Enablement engagement with us looks like in practice. We sit with each team, capture its workflows and edge cases as runbooks, turn the right ones into AI-assisted processes, and train the people who stay to supervise, correct and improve them. The aim is that the team ends up more capable, not just smaller — and that the knowledge stays in the company whoever comes and goes.

Reskill the people you want to keep

When AI takes a share of a role, the role does not vanish; it changes shape. The analyst who used to build the report now checks the report an agent built and spends the saved time on the question the report raised. The support lead who used to answer tickets now manages the agent that answers them and handles the cases it escalates. Those are better jobs, but they require skills most people have not been trained in: writing clear instructions, reviewing machine output critically, spotting when an agent is confidently wrong.

Reskilling is cheaper than rehiring and far cheaper than losing the person who knew where the bodies were buried. It also changes the mood of the whole exercise. A team that is being trained to work with AI hears a very different message from a team that is being offered packages.

The honest counter-argument

There is a reasonable case for moving on headcount first. Redesigning work takes time, and in a business under real cost pressure, time is the scarce thing. Voluntary programmes are also among the most humane ways to reduce a workforce: people choose to go, on known terms, rather than being told. And there is an argument that some organisations will never redesign their work until they are forced to — that a smaller team is the forcing function that makes people adopt the tools. Microsoft may well be making exactly that calculation, with far better data about its own workflows than any outsider has.

We accept all of that. But a forcing function only works if the tools and the know-how are there to be adopted when the pressure arrives. For most mid-sized companies they are not yet. The redesign does not need to take a year — a focused team can map, automate and retrain around two or three core workflows in a quarter — and doing it first means any people decision that follows is made with eyes open.

What to do in the next 30 days

• Pick two teams and map a real week of their work. Task by task, sorted into the three buckets above.
• Write down the exceptions. Ask your most experienced people what goes wrong and how they fix it. Record it before anything else changes.
• Put one AI-assisted workflow into daily use. Measure the hours it actually moves, not the hours a vendor estimated.
• Start the training before the reorganisation. Teach the people you want to keep how to review, instruct and supervise AI output.

Microsoft’s buyout will be read as a signal that AI means smaller teams. The more useful lesson is about sequence. Companies that change the work first and size the team second end up with organisations that actually run better. Companies that do it the other way round end up with fewer people, the same workload, and the know-how they needed already gone.

Under pressure to restructure around AI and not sure the work is ready? Talk to Cravings about redesigning the work first — we map your teams’ real workflows, capture the know-how as runbooks, put AI-assisted processes into daily use, and train your people to run them.