Microsoft offered voluntary buyouts to about 60,000 employees in October. On Wednesday at the DealBook Summit, Satya Nadella gave his reason, TechCrunch reported(opens in a new tab): the offer was meant to make sure everyone "felt like they had agency in making choices." He said Microsoft is not replacing people with AI but reorganizing so that everyone uses it, and he predicted that every employee will eventually manage AI agents rather than do the work directly. An AI agent here means software that carries out multi-step tasks on its own, inside the tools a company already uses.
His example was the team behind Cowork, the agent built into Microsoft 365 apps. It has 11 developers, and Nadella said they build "a software factory that builds the product" rather than the product itself. That is the clearest picture yet, from the head of a very large company, of what an agent-managed job looks like: a small team whose output is the system that does the work.
What he did not say matters as much to anyone weighing a similar move. Microsoft has not disclosed how many people accepted, what the program cost, or which teams they left. A voluntary offer lets employees choose, which is its appeal. It is also most attractive to the people with the best outside options, and those can be the experienced staff who know which work is safe to hand to an agent. Nadella himself noted that Microsoft is 51 years old, was not built AI first, and that the change will take longer than many expect.
For most enterprises the lesson is about order. Microsoft can absorb a program at this scale and wait years for reorganized teams to pay back. Most companies cannot. If the job that replaces a role is "manage agents," that job needs to exist first, with named agents, a review routine and a measure of output, before headcount moves on the assumption that it does.
For the CAIO and CHRO this week: pick one team whose leaders already say agents do part of its work. Write down which agents it runs today, who reviews their output and how often, and one number for the team's output per person, then and now. If you cannot fill that in, the team is not ready for a headcount plan built on agents. And if any voluntary exit program is under discussion, list the roles you cannot afford to lose before the offer goes out, not after the acceptances arrive.
Action items
Microsoft offered voluntary buyouts to about 60,000 employees in October. On Wednesday Satya Nadella explained why: to give staff a choice while Microsoft reorganizes so that every employee manages AI agents. His model is the 11-developer Cowork team, which builds "a software factory that builds the product."
Microsoft has not said how many people accepted or what it cost. A voluntary offer is most attractive to people with the best outside options, and those can be the staff who know which work is safe to hand to an agent.
For the CAIO and CHRO this week: pick one team whose leaders say agents already do part of its work. Write down which agents it runs, who reviews their output and how often, and one number for output per person, then and now. If you cannot fill that in, the team is not ready for a headcount plan built on agents. If a voluntary exit program is under discussion, list the roles you cannot lose before the offer goes out.
Researched and drafted by an automated workflow, then reviewed and edited by a human editor before publication. Every source is linked. See how we use AI here.
Also worth knowing
- TechCrunchMicrosoft CEO Satya Nadella explains why he asked 60K employees to quit(opens in a new tab)
Nadella said the October buyouts gave staff a choice and that Microsoft is reorganizing so every job manages AI agents. He gave no acceptance count or cost, the two figures any board copying the model would need.
- TechCrunchMore banks say customers using AI agents are on their own, even if they get hacked(opens in a new tab)
Citi's new terms make customers responsible for what an outside AI agent does in their accounts, even after a hack. PNC changed its terms and Citizens warned customers, after JPMorgan and Bank of America. Before you link an agent to your bank login, read your bank's terms.
- 9to5GoogleGoogle launches Gemini 4 Neon with Carbon quality at Flash speeds, live for AI Pro(opens in a new tab)
Neon costs $3 per million input tokens and $15 per million output, 25% below Carbon once Carbon's promotional price ends November 30, and it is live for AI Pro and Ultra subscribers. If you run work on Carbon, test Neon on it before December 1.
- TechCrunchSnapchat now lets developers build AR lenses with AI coding assistants like Claude Code and Cursor(opens in a new tab)
Snap's new tool lets creators build camera effects and games from plain-language prompts, and Snap says it paid $1.8 million to Lens+ creators last month. That total is spread across many creators, and earning requires joining Snap's Lens+ Payouts program.
- WizMicrosoft agent-framework Compromised: Credential Stealer Targets Azure and CI Pipelines(opens in a new tab)
Hijacked versions 1.26.1 and 1.26.2 of Microsoft's Python toolkit for building agents stole cloud and build-pipeline secrets on November 17 before removal. If your teams build agents in Python, check lockfiles for those versions and rotate secrets.
- CNBCAnthropic IPO raises over $81 billion in largest offering on record, topping SpaceX(opens in a new tab)
Anthropic priced at $83 a share, above its $70 to $78 range, for a fully diluted value near $1.5 trillion. A listed model vendor must report its revenue and spending each quarter, which gives buyers public figures to bring to contract talks.