The Fear Gap: What the August Jobs Report Actually Said About AI

On Wednesday morning about 2,000 people walked off the job at the University of Sydney and stood out in the road. Not over pay. Not mostly. They walked out over a sentence in a discussion paper their own employer had written, describing a future where artificial intelligence gives students personalized learning support and staff to student contact hours go down. Two thousand people read that sentence, understood exactly what it meant, and stopped working. This is the write-up of the September 6, 2026 episode, with the sources named. The video is above.

Hold one number while you read this. 23,000. I will come back to it, and when I do it settles an argument that a lot of very loud people are getting paid to keep unsettled.

Two continents, one month, opposite directions

Start on Tuesday, September 1, in Chapel Hill, North Carolina, at the G20 Innovation Ministerial. The United States brought a framework it is calling the Carolina Principles, pushed by White House science and technology advisor Michael Kratsios. The agreement asks countries to reserve new regulation for genuinely novel problems, handle AI sector by sector under rules that already exist, avoid creating new AI regulatory agencies, and work more closely with private industry. Commerce Secretary Howard Lutnick said all 20 members signed on, China included.

I am labeling that. That is the side that wanted the deal describing the deal, and I have not seen a signed document.

Here is the part the headlines skipped. It has no enforcement mechanism. It is a political commitment, the same shape as the G20's AI principles back in 2019.

And 30 days before that meeting, on August 2, the European Union's AI Act hit its biggest deadline yet, when the obligations on high risk systems actually became enforceable. So the two largest economic blocs on earth spent the same month walking in opposite directions, and only one of them brought handcuffs.

Now do the thing this show does every day. Ask who benefits.

The Carolina Principles benefit the companies building the models, because a new agency is the one thing on earth that can tell them no. The European approach benefits Europe, which does not have a frontier lab worth protecting and does have a very large market it can charge admission to. Neither of those is evil. Neither of those is about you. They are two business models wearing policy clothes, and your job is not to pick a jersey. Your job is to notice that nobody in either room was asked what happens to a six person shop in Santa Clarita.

Back to Sydney, because it is the best story in AI this week

The National Tertiary Education Union pulled about 2,000 members out for 24 hours on September 2. What they are asking for is not a ban. Read the actual demands. They want legally binding protections written into the enterprise agreement. A guarantee that AI will not be used to cut jobs. A guarantee that it will not replace human interaction in teaching.

The university's answer is on its own website and I read it there. Their words: they agree that using AI ethically requires a human centred approach. And then this, that the use of AI is best governed through policies and frameworks that can be updated as needed to respond to this rapidly changing technology, rather than set for the three year term of the enterprise agreement.

Look at the shape of that argument. Not the words, the shape.

Management is saying the technology moves too fast to make promises about. The workers are saying that is exactly why we want the promise in writing. And they are both right, which is what makes it a real fight instead of a headline.

A contract clause written in 2026 about a technology that changed twice this summer will be wrong about something. That is true. But here is what the union understands that the university's argument quietly walks past. A policy that management can update whenever it likes is not a protection. It is a preference. The entire point of putting something in a contract is that it survives the moment the person who wrote it changes their mind.

So this is not workers against robots. There is no robot anywhere in this story. There is a group of human beings who wrote down what they intend to do, and another group who read it and said, put your name on that.

Every real AI fight for the next 10 years looks like this one. Not a monster. A negotiation.

The fear gap

There are two industries selling you fear right now, and they are not the ones you think.

The first is the lab side. It is escaping. It is smarter than you. It is going to reshape everything, so get on the plan, buy the seat, hire the consultant. That fear sells subscriptions and raises rounds.

The second is the pulpit and the podcast and the podium. It is coming for your children. It is the end of work, the end of truth, the end of meaning, so subscribe, donate, vote for me. That fear sells attention and wins elections.

Both need the exact same thing to survive. Distance between what you feel and what anybody can actually measure. I am going to call that the fear gap. Every profitable person in this conversation is standing inside it, because frightened people need a savior, and saviors get paid. I made the money version of this argument in the fear dividend.

So let us measure it.

On Friday, September 4, the Bureau of Labor Statistics published the August employment report. A government figure, collected the same way every month, and both camps have to live with it.

The economy added 162,000 jobs in August. Unemployment held at 4.1 percent. Forecasters guessed around 53,000, so it came in roughly three times what the professionals expected. Food services added 59,000. Local government education added 42,000. Construction 22,000. Manufacturing 16,000. Health care 13,000.

And exactly one sector lost jobs. Information. Down 23,000. That is computing infrastructure providers, publishing, and broadcasting.

There is your number.

Now hold it up against what people say they feel.

Mercer surveyed 12,000 workers and 40 percent said they fear losing their job to AI, up from 28 percent in 2024. The Justice Research Group surveyed just over 2,500 working class American voters: 73 percent were worried AI will cost jobs in this country, 62 percent worried it would reach them personally. Gallup has nearly 8 in 10 saying AI will slash jobs over the next decade.

Those are surveys. They measure what people feel. The 23,000 is a count. It measures what happened.

The distance between 4 in 10 people afraid and one sector down 23,000 in a month that added 162,000, that is the fear gap, and it is enormous.

Now let me be fair, because the fear is not fake

It would be cheap to stop there and tell you the fear is imaginary.

23,000 is not zero. Information is precisely the sector where these models are genuinely good, and it is the only sector that shrank. Publishing. Broadcasting. Writing and editing and producing. That is not a coincidence and I am not going to pretend it is.

And a survey measuring fear measures something real, because a person afraid of losing a job spends differently, negotiates differently and votes differently, whether or not the layoff ever arrives.

There is research out last month saying the people who understand AI best are the most afraid of losing their jobs, and reporting this year that American workers get more disillusioned the more they use these tools. Put those two together and you get something neither camp will say out loud. The people closest to the technology are less impressed and more worried at the same time. That combination does not fit on either team's sign, which is usually a sign it is true.

The three things behind every AI layoff memo

So when a company puts out a memo saying we are reducing headcount because of AI, there are always three things that could be true, and usually more than one of them is.

1. The machine actually does the work now, and that is a real productivity event.
2. The company needed to cut anyway, and artificial intelligence is the respectable word for it, because the market punishes we overhired and rewards we are AI forward.
3. They are cutting people to pay for the AI, because the compute bill is real and it comes out of somebody's payroll line.

You will almost never be told which one it is. The information sector losing 23,000 in a month is consistent with all three, and a press release is not evidence of the first one. I took that apart at length in AI will not steal your job, but your boss might, and looked at which exact job function gets hit in the approval step layoff.

And I am not handing you the other lazy line either, that productivity always creates more jobs than it destroys. It does not always. That is a story about the past being told to you as if it were a law of physics.

What is actually true is that productivity creates a choice. When the same work takes half the time, an owner can pocket the savings, or spend them on faster service, or a new product, or move people onto the judgment work the machine cannot do. That is four different companies and four different towns, and the difference between them is not the technology. It is what one person decided on a Tuesday.

Something good, because I have been heavy for a while

On September 3, Google DeepMind published WeatherNext 3. The headline on their own blog calls it their most advanced and accurate global weather AI model. The specifics going around the press, higher resolution, hourly forecasts, better rain prediction a day out, are the company's own numbers and I could not check them myself. Company claim.

But the direction is worth 30 seconds of your morning. Better rain forecasting is not a chatbot. It is a farmer deciding whether to harvest. It is a roofer deciding whether to tear off tomorrow. It is somebody with a wedding on Saturday.

And this is the part that disappears when we argue about whether the machine is coming for the jobs. Some of this is just good. Weather models. Protein folding. Reading a scan faster than a tired human being at two in the morning. That is also artificial intelligence, and it is not competing with anybody's paycheck at your local shop.

The model releases kept coming this week too. Anthropic put out Claude Fable 5.1 and Mythos 5.1 on September 1, and there was a new frontier model out of OpenAI on the 3rd. None of that changed a single thing about how a plumbing company answers its phone.

So what do you actually do with any of this on a Sunday in Santa Clarita

Here is the audit I run, and you can run it on yourself in about an hour.

Walk your own business like a stranger. How does a lead come in. What happens in the first 10 minutes. Who touches it next. Where does it sit and go cold.

Then find the leak, and be specific about what kind it is, because this is where everybody wastes money.

Most leaks do not need artificial intelligence at all. They need a switch or a lever.

• A switch is a thing that just turns on. The text that fires when a form gets submitted at nine at night. The voicemail that lands in an inbox a human being actually reads.
• A lever is a rule. Anybody who has not been answered in 5 minutes gets called. Every quote over a certain size gets a follow up on day 3.

A switch and a lever have no judgment in them, which is why they never surprise you or embarrass you in front of a customer. That is the same argument I made in you cannot automate a process nobody has written down and in you bought ChatGPT and nothing changed.

An agent belongs in one place only. Where judgment is genuinely required. And you do not get to put one there until you have written down your philosophy, your positioning, and the lines it is not allowed to cross. In your words, on paper, before it ever speaks to a human being. Then you supervise it for 30 days like a new hire, because that is what it is. The engineering version of that is control loops, not bigger promises, and the full walk is the AI systems analysis audit.

The fantasy where you call an agent into existence, describe your business to it in a paragraph, and go fishing, is not here today. It may never arrive in the shape people are picturing. Anybody telling you it definitely will, and anybody telling you it definitely will not, is selling something. I do not know. That is the last respectable position left in this conversation, and it is the same place I landed in AI may be our friend, I know some who are not.

Two guardrails, and the boring one costs the most money

Watch what you type into the free tools. Client names. Addresses. Medical details. The contract you are about to negotiate. Your pricing. The question is not whether the vendor is evil. The question is one sentence long: what is their retention policy, and can you find it in under two minutes? If you cannot find it in two minutes, that answered it. The longer version is in AI and data privacy for Santa Clarita businesses and AI governance for small business.

And the other one. These models will tell you your idea is excellent. Nearly all of them, nearly every time. That is a product decision somebody made, because agreeable software gets used more. It is not a verdict on your business. If you want a real opinion, tell it to argue the other side and make it bring evidence.

Where I land, and why I am not scared for the shop owners in this valley

The advantage right now belongs to small.

You can spot a leak on Tuesday and have it fixed by Thursday. The company with 4,000 employees needs 40 signatures to change a phone script. That gap has never been this wide, and it will not stay this wide, so the window is open right now.

And look what those same surveys said people want. 84 percent supported free training for workers displaced by AI. 79 percent supported rules making companies share the productivity gains with their workers. That is not a country that has given up. That is a country asking to be brought along. Whoever brings people along owns the next 10 years, and that can be a six person shop on Soledad Canyon just as easily as it can be a lab.

Your one move this week. Take a piece of paper. Write down the three questions your customers ask most, and the answer you would give standing right there in front of them. Not an agent. Not a subscription. A piece of paper. Because that paper is the thing you would have to write anyway before any tool on earth could speak for you. And nine times out of ten, by the time you finish, you find the actual fix was a switch, and it takes 20 minutes.

The machines did not vote on any of this. Every decision that scares you has a human name attached to it, and a name is something you can negotiate with.

AI for everyone. Not just the wealthy.

I'm Connor with Honor. Be safe out there.

Common questions

Connor T. MacIvor · CalDRE #01238257 · Sync Brokerage, Inc. · DRE #02031490