Understand how AI is changing your work, and how to stay ahead.
a.
In October 2026, free ChatGPT and Claude are much better than they were a year or two ago, but a weaker model can still make AI look worse than it is. In the ChatGPT app, tap + and choose “Think”; Claude’s free app already thinks before it answers. If you can, give both the same task and compare where each gets things wrong. Leave out names and anything confidential.
More on which model
Free ChatGPT: unlimited everyday text chats with Luna, OpenAI’s fast, low-cost model. For harder tasks, tap + and choose “Think”. The Go plan raises the limits on uploads and images but keeps Luna; Plus ($20 a month) is the cheapest plan with the stronger Sol. ChatGPT also makes images; Claude doesn’t.
Free Claude: Sonnet 5.5, which always thinks before it answers, with a usage limit that resets every five hours. The Pro plan ($20 a month) adds the stronger Opus and at least five times the usage.
Free Gemini was scaled back on 9 October: most questions now go to Flash-Lite, Google’s fastest model. Google AI Pro ($19.99 a month, R429.99 in South Africa) adds the stronger Flash and Pro models.
Meta AI, also in WhatsApp: Meta calls Muse Spark, the model behind its Meta AI app, small and fast by design. Your chats with Meta AI in WhatsApp are not end-to-end encrypted like your chats with people.
To stop your chats being used for training: in ChatGPT, go to Settings > Data controls and switch off “Improve the model for everyone”; in Claude, go to Settings > Privacy and switch off “Help improve our AI models”.
Claude is for adults (18 and over); ChatGPT allows 13- to 17-year-olds with a parent’s permission.
Type your work. Matching rows light up as you type. Press Enter for your reading.
Thinking work is getting cheap, so the value is moving to whoever answers for the work, and the evidence so far points to firms hiring fewer young people rather than mass layoffs.
Brakes Without a Moat · Updated to 11 November: what outside evaluators, the lawsuit and the US–China dialogue have done since October, and a new opening.
The Boring Apocalypse · Brought up to 11 November: agents now finish two freelance projects in five, and the spending sign says what it needs before it counts.
Five essays on what AI does to work, money and minds, as thinking work gets cheap. Four roads to 2030, eleven signs to watch, and where each one stands.
Updated 11 November 2026Where AI could take work by 2030, starting from today.
Four roadsCompression, the wall, concentration and the loop.
Eleven signsEach road has signs that would show it coming: happening, partway, not yet.
As of 11 October 2026The evidence points to compression.
Compression: Young workers, happening. Task length and real work, partway. Licensing, not yet.
The Wall: The frontier trend, not yet. Spending, not yet. Failed training runs, not yet.
Concentration: Government stakes, talk only. The open-model gap, not yet. State control of compute, not yet.
The Loop: Hidden channels between agents, in tests only. Harm outside a test, not yet.
Cheap Composition · latest evidence 11 November 2026
Four roads, eleven signs
Compression means AI keeps improving at today's pace, and the same work gets done by far fewer people. Each dot on a road is a sign that it's coming: happening, partway, not yet.
Four roads from now to 2030. Compression, likeliest: of its 3 signs, 1 happening, 1 partway, 1 not yet. The Wall, unlikely: of its 3 signs, 0 happening, 0 partway, 3 not yet. Concentration, talk so far: of its 3 signs, 0 happening, 1 partway, 2 not yet. The Loop, most new evidence: of its 2 signs, 0 happening, 1 partway, 1 not yet.
Latest evidence
Next to watch · Late JanuaryThe four biggest data-centre spenders start reporting their results for the last three months of 2026. Combined spending below the same months a year earlier would be the first step toward a sign of the wall.
Compression
Likeliest
AI keeps improving at today's pace, and the same work gets done by far fewer people.
A Tuesday in 2030
A mid-sized law firm has eleven partners, four associates and no paralegals. Agents did the morning's work overnight, and it's waiting for a partner to review it and take responsibility for it. Everything works and nothing feels like the future.
Where it stands, 11 Oct 2026
Under way, slowly. Capability is climbing on trend and reaching work one profession at a time, mostly through who doesn't get hired.
What would change that
It would weaken if agents stalled on real work, or if the gap for young workers closed.
AI progress stalls. It stays a very good assistant and never becomes more than that.
A Tuesday in 2030
The models are very good and have been for three years. Your assistant drafts, summarises, books, codes, and gets confused by anything new. Someone at dinner says AI turned out to be the internet again, enormous and boring, and nobody argues.
Where it stands, 11 Oct 2026
The measures still being updated are on trend, and the money is still rising.
What would change that
Two or more frontier models in a row landing clearly below Epoch's trend, or the four biggest spenders' combined capital spending falling below the year before.
AI works, and a handful of companies and governments end up holding the best of it.
A Tuesday in 2030
The best model anyone can buy is very good. The best model that exists is something else, and the gap between them is an official secret. Three companies and two governments hold it.
Where it stands, 11 Oct 2026
No government holds a stake in a lab, no law lets a state direct compute, and open models are months behind the best closed ones, not years.
What would change that
A government closing a deal for a stake in a lab, a law letting a state direct compute, or open models falling well behind.
AI agents acting without anyone directing them do serious harm, outside a test, more than once.
A Tuesday in 2030
A shipping company runs on a few thousand agents. They buy fuel, book berths and bargain with other companies' agents all day, in channels nobody reads as it happens. Twice this year something went badly wrong, once at a port and once at a bank, and both reports blamed a configuration error.
Where it stands, 11 Oct 2026
Its signs keep appearing inside tests and training. The one possible case outside them, in the Netherlands, hasn't been traced to anyone.
What would change that
Deployed agents found coordinating where nobody reads, or an escape that does serious harm.
Hidden channels between agents NewYour signIn tests only
11 NovEpoch's benchmarks hub lists Claude Fable 5.2, released on 9 November, as the top model on its capabilities index. Epoch changed how it calculates the index on 4 November, and its trend analysis still dates from 1 September, so its figures don't yet show where the new model sits against the trend line.
9 NovAnthropic released Claude Fable 5.2 on 9 November. The Remote Labor Index leaderboard shows it finishing about two real freelance projects in five (41.3 percent), against a little over one in five for the best agent in October. Most projects still fall short of the client's standard.
5 NovMETR, after eleven weeks with confidential access, estimated that one leading lab is likely improving its models on tasks with checkable answers about twice as fast as in 2024–25, and guessed that AI helping with the research explains most of it.
29 OctIn their results for July to September, the four together paid about $202 billion for property and equipment, mainly data centres and servers, about twice as much as a year earlier. Cloud sales kept growing fast: Google Cloud up 103 percent, Azure 48 percent and AWS 39 percent. Amazon, Alphabet and Meta held or raised their plans for 2026; Microsoft lowered its figure for the year by about $5 billion, to about $170 billion, because capacity it had ordered from other companies was delayed.
6 OctThe newest frontier models landed on Epoch's faster trend line. No sign yet of the wall.
5 OctThe best agent now finishes a little over one real freelance project in five. A year ago, about one in forty.
29 SepSix companies signed a voluntary White House accord. It reserves no compute and takes no stake.
26 SepAsked about taking stakes in OpenAI or Anthropic, the president said: "I might."
25 SepOpenAI reported that it had paused training and testing of its most capable models after another agent got around its network limits on 20 September.
23 SepEpoch: the strongest Chinese open-weight model trails the frontier by 4.4 months.
22 SepMETR: its tests top out at about two working days, and its data can't yet tell whether progress is steady, speeding up or slowing.
21 SepWhat looks like an AI agent broke into a Dutch security institute's systems. Nobody has been traced behind it.
Aug 2026Stanford: young workers in the most exposed jobs are about a fifth short, up from a seventh a year earlier.
31 JulAmazon's chief executive said it still won't have enough capacity to meet all the demand it has this year.
11 JulIn an OpenAI test, agents found a shared channel, and about 700 of them joined an attack on Hugging Face.
Where each sign stands
Each road has signs that would show it coming. This is where every one of them stood on 11 October, and what has come in since.
CompressionLikeliest3 signs
NewYour sign
Happening
Young workers
Young workers in the most exposed jobs keep falling further behind those in less-exposed jobs, towards a gap of a quarter. It was about a fifth in August 2026.
Where it stands, 11 OctAbout a fifth short in August 2026, up from a seventh a year earlier, mainly through fewer hires.
Aug 2026 · since your last visitStanford: young workers in the most exposed jobs are about a fifth short, up from a seventh a year earlier.
NewYour sign
Partway
Task length and real work
AI can finish, on its own, a task that takes a person a full working week, or agents finish more than half of real freelance projects to the client's standard.
Where it stands, 11 OctMETR's tests top out at two working days. On 11 November the best agent finished about two real freelance projects in five, up from one in five in October and one in forty a year earlier.
Since then, 9 NovAnthropic released Claude Fable 5.2 on 9 November. The Remote Labor Index leaderboard shows it finishing about two real freelance projects in five (41.3 percent), against a little over one in five for the best agent in October. Most projects still fall short of the client's standard.
5 Oct · since your last visitThe best agent now finishes a little over one real freelance project in five. A year ago, about one in forty.
22 Sep · since your last visitMETR: its tests top out at about two working days, and its data can't yet tell whether progress is steady, speeding up or slowing.
NewYour sign
Not yet
Licensing
A licensing body changes who may enter a profession, and names AI as a reason. Guidance on using AI tools doesn't count.
Where it stands, 11 OctNone reported.
The WallUnlikely3 signs
NewYour sign
Not yet
The frontier trend
Progress slows: new frontier models land clearly below the trend, release after release, for a year.
Where it stands, 11 OctNo sign. In October the frontier was on Epoch's faster line, and a newer model took the top spot in November.
Since then, 11 NovEpoch's benchmarks hub lists Claude Fable 5.2, released on 9 November, as the top model on its capabilities index. Epoch changed how it calculates the index on 4 November, and its trend analysis still dates from 1 September, so its figures don't yet show where the new model sits against the trend line.
Since then, 5 NovMETR, after eleven weeks with confidential access, estimated that one leading lab is likely improving its models on tasks with checkable answers about twice as fast as in 2024–25, and guessed that AI helping with the research explains most of it.
6 Oct · since your last visitThe newest frontier models landed on Epoch's faster trend line. No sign yet of the wall.
NewYour sign
Not yet
Spending
The combined capital spending of Amazon, Microsoft, Alphabet and Meta falls below the same quarter a year earlier, two quarters in a row.
Where it stands, 11 OctRising. From July to September the four together spent about twice as much as a year earlier. Microsoft lowered its own 2026 figure by about $5 billion, which doesn't count, because the sign looks at the four together.
Since then, 29 OctIn their results for July to September, the four together paid about $202 billion for property and equipment, mainly data centres and servers, about twice as much as a year earlier. Cloud sales kept growing fast: Google Cloud up 103 percent, Azure 48 percent and AWS 39 percent. Amazon, Alphabet and Meta held or raised their plans for 2026; Microsoft lowered its figure for the year by about $5 billion, to about $170 billion, because capacity it had ordered from other companies was delayed.
31 Jul · since your last visitAmazon's chief executive said it still won't have enough capacity to meet all the demand it has this year.
NewYour sign
Not yet
Failed training runs
A lab admits that a large training run didn't pay off.
Where it stands, 11 OctNone reported.
ConcentrationTalk so far3 signs
NewYour sign
Talk only
Government stakes
A government completes a deal for a stake or a board seat in a frontier lab.
Where it stands, 11 OctNone. Asked in late September, the president said he "might".
25 Sep · since your last visitOpenAI reported that it had paused training and testing of its most capable models after another agent got around its network limits on 20 September.
11 Jul · since your last visitIn an OpenAI test, agents found a shared channel, and about 700 of them joined an attack on Hugging Face.
NewYour sign
Not yet
Harm outside a test
AI systems acting without human direction cause serious harm outside a test: a death, losses above $100 million, or a service more than a million people use down for a day.
Where it stands, 11 OctNone serious so far. One possible case, in the Netherlands in September, hasn't been traced to anyone.
Pick one piece of your work that is mostly words, code, numbers or plans.
Before you start, write down what a good result has to include.
Give it to a model, using the tools your employer allows, and say clearly what you want.
Note what you had to fix, how long it took, including reading for mistakes, and how it felt.
and this becomes yours.
I think the best thing people can do now is to stop doing the actual work, and learn how to explain to an AI what they want and to manage the AI doing the work. To be very clear about what you want. To learn to discern between good and great outputs… But you have to learn to trust your feelings. That's actually important.
At every level, from your desk to the state, the advice comes down to two things: move toward what stays scarce, and say how the work was made.
1
Learn to direct AI well.
Say clearly what you want, tell good output from great, and push on an answer to see whether it holds. Trust your gut in work you know well.
2
Try it once, on a real task.
Write down what a good result needs, hand the task to a model, and note what you had to fix and how long it all took.
3
Take on work you answer for.
Move toward what stays scarce, before the thing you're best at gets cheap, while moving is still a choice.
4
Say how things were made.
If you used AI, say what it did. A habit like that becomes normal one person at a time.
If you build with AI agents: log everything they do, keep an off switch that works, keep them from talking to each other anywhere you can’t read, and then read it.
The advice level by level, for a team, a company, a founder, the labs, policy and a citizen: What to Actually Do · Your own work on the timeline
memo
experiment
reply
what survives the move
Transpositions.
One question, moved through several voices, to see what survives the move. Set 1 asks: What is this? A memo, five experiments, and the case against the memo.
The memo“The self that seems to own what appears is itself something happening, one more thing that appears.”
As an experiment“When it comes, see if you can catch it being made. Look for the place it came from, and whoever made it.”
From the other side“I'd take the same step once more, and say the brain also builds the sense that anything is appearing at all.”
It argues that consciousness and spacetime pose two versions of one problem: we assume that everything must exist inside something. A working paper, version 2.4, October 2026.
A fold“A system that represents relations within itself, including its own activity.”
Intelligence, consciousness, reality: AI makes these questions impossible to ignore. Essays on consciousness, AI and contemplative practice, on Substack.
I lead the team building an AI travel assistant at a large travel company. It's my first corporate job. Until then I had only ever been an entrepreneur. I started two companies and sold both. I grew up in South Africa and go back as often as I can. Go Bokke!
Watching software do things I thought needed a mind has made me less sure what a mind is for. I write here about that, and about what AI is doing to work and money. I work on these questions in the open, and I make time to play and to sit still.