Who does the work, and who pays: Apple puts AI on the desk, OpenAI halves the agent's price, Microsoft offers 60,000 buyouts

Three signals this Tuesday, and one question: who does the work now, and who pays for it? Apple puts AI on your desk rather than in the cloud. OpenAI halves the price of the agent that does office work. And Microsoft offers sixty thousand people a paid exit, to help fund the machines.

Signal one: Apple puts AI on the desk

On Tuesday, Apple unveiled a new 14-inch MacBook Pro and a new iMac, both powered by its latest M6 chip, touting faster performance for AI tasks run directly on the device.

• M6 MacBook Pro: from 1,999 dollars. Up to 30% faster graphics and up to 25% more AI performance than M5, up to 24 hours of battery, the longest ever on a Mac.
• M6 iMac: from 1,499 dollars. Up to five times the on-device AI performance of the M4 model.

Pre-orders open in 35 countries, with availability from November 6. The launches come the day before an Apple event in Los Angeles on October 28, where Bloomberg expects the first touch-screen MacBook Pro.

Why it matters

Apple's bet runs opposite to everyone else's. While rivals spend hundreds of billions on data centers, Apple argues the most valuable AI runs on the machine you already own.

On-device AI means no round trip to a server, no meter running on every request, and documents that never leave the desk. For a company that sells hardware, that technical choice is the business model itself: the value of AI becomes a reason to buy a machine.

I connect this bet to privacy, which this series has crossed since day one. Muse reading messages, Full Disk Access tightened by Apple on October 2. AI that never leaves the device solves part of those problems by design.

What changes

If the best AI runs locally, the advantage shifts from whoever owns the biggest model to whoever owns the device in front of you. The computer stops being a window onto someone else's cloud and becomes the place where work happens again.

For every software company, the question becomes simple: does your product still need the cloud, or just a faster machine? Apple just made the second answer far more credible. For a brand handling sensitive data, customer or health data for example, "your data never leaves your device" becomes a selling point in its own right.

Signal two: the office agent at half price

On Monday, OpenAI launched GPT-6.1 Terra, its mid-range model built for visual reasoning and computer use: reading screens, filling in forms, entering invoices into accounting software, drafting proposals.

The price is the real news. Through the API, Terra costs 1 dollar per million input tokens and 5 dollars per million output tokens (0.05 dollars for cached input), 50% less than GPT-5.6 Terra and the larger GPT-6.1 Sol, according to OpenAI. It is available in ChatGPT Work and Codex on paid plans, with a 1.05 million token context window. Amazon Web Services made it available on Bedrock the same day. I note the detail: on October 21, OpenAI arrived on Google Cloud. Models are now on every shelf the day they launch.

The vendor's numbers

OpenAI's tests show the factual error rate falling from 53.6% with GPT-5.6 Terra to 16.7% at low reasoning effort. On Terminal-Bench, a coding test, Terra solves 50.3% of tasks at maximum effort, versus 21.5% before. These are vendor numbers. After the ARC-AGI-3 episode on October 25, I read them as such.

What can be verified is the price, and the price has been halved. An independent measure, Agents' Last Exam, puts Terra 15 points above its predecessor at about a third of the cost per task.

Why it matters

The agent economy just found a workhorse it can run at volume. Whether the model could read an invoice was never the real blocker. The blocker was whether running it a hundred thousand times a day made financial sense. That math just changed.

What changes

For every company with administrative work, the question flips from "can AI do this?" to "why are we still paying an hourly rate for this copy and paste?"

The cheap agent leaves judgment to humans and takes the repetitive middle of the work. Companies that redesign their processes around that fact will bank the savings. Those that bolt a cheap model onto a broken process will get a faster broken process. I said it on October 24 about small businesses: integration is the real work. The price cut makes that redesign more urgent, and the redesign is what counts.

Signal three: Microsoft offers an exit to a quarter of its workforce

On Monday, Microsoft offered paid voluntary buyouts to about 60,000 employees, more than a quarter of its global workforce of 223,000.

The offer, announced in a memo from chief people officer Amy Coleman, covers eligible employees at level 70 and below, up to vice presidents, in participating teams in the US, Canada, the UK and India. Participating divisions include AI experiences, Copilot, finance, human resources, Xbox and Windows and Devices. Marketing, commercial business and security teams are excluded.

Coleman writes that the offer gives eligible employees considering a future outside the company a new choice to make that transition now, with support, and gives organizations another way to adapt their workforce to changing needs.

The context

Microsoft has forecast 175 billion dollars of capital expenditure for 2026, mostly data centers and chips. Earlier this year it offered voluntary buyouts to about 9,000 US employees. This round is nearly seven times larger. According to reports, about 30% took up the previous program.

Why it matters

I read the three signals together. The same week Apple puts AI on the desk and OpenAI halves the price of agent work, Microsoft asks a quarter of its workforce to consider leaving, to help fund the machines.

This is the human ledger of the AI buildout, written plainly. The buyouts are voluntary, and nobody knows how many will accept. But the direction is set: headcount has become a source of funding for compute.

I connect this to Anthropic's October 18 study on disappearing junior roles. Both describe the same transformation: fewer people, more agents, and an open question about how anyone still gets trained.

What changes

For every company building an AI strategy, headcount has left the footnotes and become the core of the strategy. Microsoft is reorganizing around a smaller, more AI-amplified workforce, and doing it openly, through an internal memo it stands behind.

Every leader has to answer the question Microsoft just settled for itself: when machines get cheaper and people get more expensive, what do you keep people for? My answer, for the companies I work with, fits in three words: judgment, relationship, accountability. Those are what no agent can carry for you in front of a customer. An honest answer to that question is the whole AI strategy. The rest is procurement.

The thread of the day

Three signals, one thread. Work is changing hands: from the cloud to your desk, from expensive models to cheap ones, from headcount to compute.

The companies that win this shift will be the ones that decide what humans are really for, before the budget decides for them.

The noise filtered. The signal kept.