At 8:41 on a Monday morning, your replacement finishes your week.
What’s real here, and what’s imagined Open
This story is fiction. The lines marked in it lean on the real world, and here is where each one stands. When reality catches up with a line, it moves to “Real already,” with the date. Checked September 23, 2026.
Real already
In 2025, Salesforce’s chief executive said AI agents had let the company cut its support staff from 9,000 to about 5,000. The company said it mostly stopped refilling those roles, and moved hundreds of people to other jobs.
- Salesforce sacrifices 4,000 support jobs on the altar of AI · The Register, September 2025
A 2024 study in Science estimated that about 80% of US workers are in jobs where at least a tenth of the tasks are exposed to language models, and that higher-paid jobs are more exposed. The authors’ earlier working paper found trades like mechanics and cement masons had no exposed tasks. Exposed means AI could speed a task up, not that the job disappears.
- GPTs are GPTs: Labor market impact potential of LLMs · Science, June 2024
- “Scientists” and “researchers” top the list of jobs where tasks are most exposed to LLMs · EurekAlert!, June 2024
- GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models · arXiv, March 2023
In a 2022 UK pilot, 61 companies cut working hours without cutting pay. Afterward, 56 kept the four-day week and 18 made it permanent. The companies volunteered, and there was no control group.
- The results are in: the UK’s four-day week pilot · Autonomy, February 2023
The EU’s AI Act classifies AI that evaluates students’ learning as high-risk, which brings extra obligations for whoever builds or uses it. An amendment adopted in July 2026 delayed those obligations to December 2, 2027.
- Regulation (EU) 2024/1689, the AI Act (Annex III, point 3) · Official Journal of the European Union, July 2024
- Digital omnibus on AI · European Parliament, Legislative Train, July 2026
When AI companies have asked the public how their models should behave, it has been about a thousand people at a time: about 1,000 Americans for Anthropic and the Collective Intelligence Project in 2023, and about 1,000 people for OpenAI in 2025. Bigger studies have since asked tens of thousands what they want from AI, but they gather opinions. They don’t set the rules.
- Collective Constitutional AI: Aligning a Language Model with Public Input · Anthropic, October 2023
- Collective alignment: public input on our Model Spec · OpenAI, August 2025
- CoVal: Public Input on Model Behavior (dataset) · OpenAI on Hugging Face, August 2025
- What 81,000 people want from AI · Anthropic, March 2026
Not yet
METR has found the length of software tasks AI agents can finish on their own doubling roughly every seven months since 2019, and about every four months since 2023. In OpenAI’s GDPval test, experts rated the best model’s work as good as or better than professionals’ on 47.6% of well-defined tasks from 44 occupations. That’s tasks, not whole jobs.
- Measuring AI Ability to Complete Long Software Tasks · METR, March 2025
- Time Horizon 1.1 · METR, January 2026
- GDPval: Evaluating AI Model Performance on Real-World Economically Valuable Tasks · OpenAI, October 2025
No country pays its people a share of what AI earns. The nearest real thing is Alaska, which has paid every eligible resident a yearly dividend from its oil fund since 1982; it was $1,000 in 2025. In 2020, researchers proposed a “Windfall Clause” asking AI companies to pledge part of any extreme profits. We know of no company that has signed one.
- Summary of dividend applications and payments · Alaska Department of Revenue, September 2025
- The Windfall Clause: Distributing the Benefits of AI for the Common Good · Centre for the Governance of AI, January 2020
Imagined
Many people use AI for parts of their work. But owning an agent that does your whole job, and selling its output to your old employer, isn’t an established arrangement anywhere we know of.
Nobody knows yet
- Which of these arrangements, if any, we end up with.
- Whether the work that’s left is work people want.
- Who, in the end, grades the machine.
Other Tomorrows is published by Ctrl AI at ctrlai.com.