Friday charts: AI makes work (for now)

The people who know the most about AI keep telling us it’s coming for our jobs.

Dario Amodei has told us to expect 20% unemployment any time now. Sam Altman has said AI will be able to do the jobs of all digital workers. Geoffrey Hinton — the godfather of AI — says it will cause “massive unemployment.” Elon Musk thinks there will eventually be no jobs at all.

So far, though, the opposite seems to be happening.

One study found that AI is making our workdays longer. Satellite imagery of Chinese cities showed that the lights were staying on longer at the office buildings of companies most exposed to AI. “AI exposure significantly increases firm work intensity,” the study concluded. 

A separate study of US workers quantified the effect: “An increase from the 25th to the 75th percentile in AI exposure corresponds to an additional 2.2 hours of work per week.”

One problem is that humans keep finding themselves cleaning up the messes that AI makes. 

Meta has reportedly found that code-writing autonomous agents have caused a 40% increase in “major technical and security incidents.” The amount of time human programmers have had to spend “firefighting” problems in the code has increased by 70%.

Even when the AIs are not making mistakes, they can dramatically increase people’s workload.

The Wall Street Journal quotes a startup founder who says he no longer sleeps through the night because he feels compelled to be available to the AI agents he employs. “The cost of the agents’ being blocked for eight hours is way too high,” the founder says. “They can be done with their work at any point of time, in the middle of the night.”

Or AI might just have a question. “They just demand your attention,” he added. “Does it need anything? Can I help it in any way?”

“The more work AI agents do, the more founders find themselves working,” the Journal concluded.

Another problem is that executives often have unrealistic expectations for AI. Management researcher Malte Jauch says employees are currently caught in a “rat race” of working longer hours to meet inflated management targets when AI tools fail to deliver promised efficiency gains.

There’s data to support the idea: A survey found that 96% of C-suite executives expect AI to raise output. But 77% of their employees say it increases their workload.

The good news is that corporations are realizing that AI can’t yet do everything.

Meta, for example, abandoned plans to shrink the size of some of its teams by as much as 60% by replacing engineers with autonomous AI agents.

Others are hiring again. The Wall Street Journal reports that companies like Alphabet, CSX, and ServiceNow have recognized a need for humans to work alongside AI.

“Just because you have coding agents doesn’t mean you’re not hiring engineers," one CEO said.

Similarly, Ford Motor Company recently said it hired 350 veteran engineers — some of them re-hired — to address quality and recall issues that AI was unable to fix.

"AI is a fantastic tool,” a Ford VP told reporters, “but it’s only as good as the information you use to train it.”

Right. But what happens when the engineers are done training the AIs?

The news on jobs — including this morning’s data — is good. 

But the Chinese workers keeping the office lights burning a little longer probably have the right idea — because the competition is only going to get better from here.

Let’s check the charts.

AGI now?

OpenAI says that, in testing, its new Astra model was able to abstract reasoning puzzles 99.9% of the time. Claude passed the same tests only 30.2% of the time. Human testers scored 48%. (Note: You can’t fix a Ford pickup truck with an abstract reasoning puzzle.)

The one yet — by a little:

Astra finished tops in Epoch AI’s evaluation of models, with a particularly good result in the category of “Mystery Game Puzzles.” But the improvement is roughly on trend.

Learning machines:

Epoch AI also finds that humans are better at learning than AIs. “We ran a human baseline on Earthborne Rangers, the complex, long-horizon board game behind our benchmark EBR-bench. Humans started low, but the best mastered the game after five play-throughs. No AI we have evaluated ever got there.” I don’t think Astra has had a go at it yet, however.

Humans still make the movies:

Despite all the AI-generated video you see on social media, AI has not yet reduced the number of crew employed on movie sets.

Getting bigger, faster:

In 2025, Anthropic and OpenAI were already among the fastest-growing companies in history. Somehow, they're growing even faster in 2026. I’d guess that’s a one-time acceleration from the advent of agentic AI. But maybe not! It might also just be the agents getting more and more useful.

I get it:

Jon Harley finds that the average US worker believes there is a 20% chance that AI will cause them to lose their job within two years. (Not shown: The average newsletter writer thinks there’s a 100% chance.)

Experience matters:

An a16z chart shows that tech companies have been hiring for experience over skill. As someone with a lot of experience (and not many skills), I welcome this development.

Humans are still dialing and smiling:

The Economist reports that outsourced call centers in the Philippines — perhaps the first job you’d think AI would replace — are still growing.

It’s a great time to be a software engineer:

The median new engineering hire at Anthropic earns $1.3 million a year. The median!

It’s a great time to have real skills:

The demand for skilled construction workers is booming, thanks to the race to build data centers.

Will AIs ever be able to install an HVAC system?

Not any time soon, I don’t think.

Have a great weekend, hardworking readers.

— Byron Gilliam

_{Brought to you by:}

Avalanche Summit NYC returns September 16–17, bringing together the institutions, enterprises, investors, and builders turning blockchain technology into real business outcomes.

From tokenized markets and institutional finance to payments and consumer applications, the Summit will explore how production-ready infrastructure is enabling faster settlement, lower costs, and entirely new products and revenue streams.

Use promo code BLOCKWORKS15 for 15% off!