Friday charts: Jobpocalypse now?

Even when things were good at the investment bank I used to work at, there always seemed to be another round of layoffs looming — partly, I think, because management had no idea how many people they needed.

I worked on the sales-and-trading floor, where there’s a revenue number at the end of every day: commissions paid minus trading losses (and occasionally gains). So you’d think it would be easy to quantify who was adding or subtracting what.

It wasn’t. 

The commission paid on a trade might be attributable, in part or in whole, to the research analyst, salesperson, or sales trader who spoke to the customer — or to the trader who took the other side of the trade (me!).

No one really knew why a customer traded with us. So it was impossible to attribute who exactly was responsible for each commission, and therefore, to figure out who was absolutely necessary to the business.

To paraphrase Wanamaker, half the payroll might have been wasted; they just didn’t know which half.

The only way to find out was to fire some people and see what happened.

It feels like something similar is about to transpire in companies everywhere, because it’s not just investment banks that struggle with this.

When jobs were mostly in farming and manufacturing, it was easy to measure an employee’s productivity: Just count up how many apples they picked or widgets they turned.

When most people started working in offices, however, it got much harder.

“Knowledge work is not defined by quantity,” Peter Drucker wrote. “Neither is knowledge work defined by its costs. Knowledge work is defined by its results.”

Employers didn’t know how to measure these results — what’s the unit of output for a day of meetings, phone calls and inter-office memos?

So they measured time instead: Employees were paid to confine themselves to an office for eight hours a day and employers just hoped they got eight hours of work out of them.

Time became a proxy for output.

But what happens when everyone works from home?

If employers can’t measure employees by their time in the office, they have to measure their output instead.

This is a good thing. “Stressing output rather than activity is the key to productivity," Peter Drucker wrote in 1967. 

But employers never really figured out how.

Now, AI is forcing employers to try again. Large language models can do a lot of time-consuming things, so employers are starting to rethink what they pay people to do.

I’m not sure they’ll figure it out any better than the bank I used to work at. But the AI narrative is putting so much pressure on companies to find productivity savings that a lot of them will just fire people and see what happens.

This morning’s data suggests it might have started: The BLS reported technology-sector jobs fell by 12,000 last month from the month before, and by 57,000 over the last year.

There was good productivity data this week, too, which some economists think is the first sign that companies are using AI productively. 

So maybe companies will soon be able to do more with less.

But they might just do more. 

A new paper from the Harvard Business Review finds that “AI does not reduce work, it intensifies it.”

In an eight-month study of work practices at a tech company, the authors found that AI led employees to work at a faster pace, take on a broader scope of tasks, and extend work into more hours of the day.

“Many prompted AI during lunch, in meetings, or while waiting for a file to load. Some described sending a ‘quick last prompt’ right before leaving their desk so that the AI could work while they stepped away.”

That might sound good to employers hoping to get more out of their employees. And this part will sound great: “[W]orkers increasingly absorbed work that might previously have justified additional help or headcount.”

But the researchers have a warning for employers:

If so, companies may soon find they need more people, not less.

That, at least, is what the head of human resources at IBM expects. Slashing early-career recruitment may save money in the short run, Nickle LaMoreaux told Bloomberg, but it risks creating a scarcity of mid-level managers later on.

As a result, IBM plans to triple its entry-level hiring. “And yes,” LaMoreaux says, “it’s for all these jobs that we’re being told AI can do.”

The investment bank I worked at was always hiring in between rounds of layoffs — constantly churning through staff as it tried to figure out who really did what.

The whole economy might soon be doing the same.

Let’s check the charts.

Jobpocalypse now?

This morning’s jobs report was “brutal” for the tech sector. Losing 57,000 jobs over the past year is “nearly as bad as the worst of the 2024 tech-cession, and significantly worse than either the 2008 or 2020 recessions.” Yikes.

Too many chiefs?

An academic paper finds that generative AI is creating “seniority-biased technological change” in employment that disproportionately affects junior-level staff. It’s not just in technology, either: The study crunched resume data from 285,000 employers.

The hiring recession:

The same study explains that the decrease in junior-level employment happens “entirely through a decline in hiring.”

The AI effect:

Sites people have long consulted for buying advice, like Wired and Tom’s Guide, have experienced collapses in traffic. We just ask chatbots now — which get information from the very sites they’re putting out of business.

Also AI? 

Professor of Applied AI Alex Imas notes that this week’s productivity data is “showing signs” that companies are already reaping benefits from AI.

Almost all talk?

Goldman Sachs (via Callum Williams) data shows that while 70% of companies talk about AI, only 10% can explain how it helps their business, and only 1% can quantify the effect on earnings.

Work is always changing:

Technology journalist Rowland Manthorpe maps the most common jobs of the 1980s, finding that “secretary” was the most common job in 19 US states. 

The jobs AI can and can’t do:

Peter Walker reworks data from Anthropic on what portion of each occupation AI could theoretically perform (in blue) and how much it’s doing now (in red).

Good question!

In a reply on X, Boris Cherny, who leads Claude Code, explains that all the code Claude is writing is creating new work that can only be done by people. 

Nice work if you can get it:

Annual Salary: $405,000−$485,000 USD.

A few of the job openings at Anthropic and what they pay. Code writes the code, but someone still has to tell the code what code to write, and it’s a high-paying job.

Claude is winning:

An incredible chart from Ramp showing the shrinking share of the business market for OpenAI (in shades of blue) vs. the growing share for Claude (in shades of orange-ish).

Timing mismatch:

A Gartner study predicts “there will be no ‘jobs apocalypse’ due to AI — but there will be job chaos.” They expect AI to create more jobs than it eliminates starting in 2028.

Call me an Apocaloptimist, but I think it’ll happen even faster.

Have a great weekend, hard-working readers.

— Byron Gilliam

The building blocks for a new, composable financial system are here. Digital cash and wallets are poised to rewire the global payment stack, but how will it all unfold?

Join this live Blockworks Roundtable to hear frontier takes from experts across the revamped global payments stack including Canton, Hifi and Dfns. Happening March 10th!

^{In partnership with Canton}