Will AI decrease employment?

The May jobs report looked great.

The BLS report for May showed 172,000 new nonfarm payroll jobs, unemployment holding at 4.3%, and March/April revisions that added another 93,000 jobs. Axios called out that the number was more than double the 80,000 jobs economists expected.

Lots of people on Twitter were debating the merits of AI on Friday as the market sold off broadly. If jobs are up, does that mean we invested too much into GPUs, memory, and data centers? Shouldn't AI cause a decrease in employment? This is what common knowledge suggests.

And I think they're wrong.

The job is the task, and region skews the data

So let's start where AI is destroying jobs outright.

Benedict Evans had the right frame on Lenny's Podcast: ask whether AI is replacing a task, or the job itself. Sometimes the task is the job. Elevator attendants were the example from the episode. Once the button existed, the job disappeared.

Today this means Fiverr "Photoshop this image"-type gigs are over. Low-priced logo creation, cheap web development, and quick illustrations are all easy to do with AI. If I had to guess, many freelancers aimed at lower price points in these categories are struggling.

But if you've ever browsed Upwork, you'll notice quite a few of these folks are outside the United States. Their employment changes would not show up in US payroll data.

Since the US is leading in the data center rollout race, everyone looks to our economy for how AI will impact jobs, but this is the wrong mental model.

Business operations and supply chains are global now. Upwork is not bounded by borders. Jobs data for a single country won't tell you the whole story.

The sell-off on the back of last week's US jobs data is more of a sign that the Fed might raise rates and has nothing to do with AI ROI yet.

Bottom-up AI

Most companies are still in a bottom-up AI deployment pattern. This means your company gives you access to a tool, and you remember to use it of your own accord. This is like "my company bought everyone Cursor/Copilot/Claude/etc.".

The company does not know what to expect as far as productivity gains go, and employees are not honest about how much time they're saving (why should they be?).

I wrote about this previously in The economics of AI-powered dev efficiency. It mostly speaks about developers and writing code, but the lessons apply to any bottom-up productivity gain.

The real productivity gains come when entire workflows and systems are redone to be "native" for the new technology. An individual worker in a company normally does not have the power to change the workflow. They can usually only become a more efficient step in that workflow.

Change like this takes business leaders actually understanding the new technology and its limits — and the courage to apply it.

Most businesses are lacking both the understanding and the courage.

So we're stuck in the Bottom-Up Local Maximum and the gains will roll out incrementally.

The big efficiency jump will happen when businesses rethink their operations from the ground up.

We're still in the early innings of ROI — most gains are going to workers who lean into the tools. Businesses won't see the full gain until they revamp their processes to be AI-first.

At the limit, advantage is competed away

Okay, but even after that big efficiency jump from AI-native operations, what then? Won't there be massive decreases in job counts and massive unemployment?

I don't think so.

I believe humans will always provide additional judgment on top of any AI-powered systems that run a business.

If Firm A (all AI) and Firm B (AI + one human) are competing and both have the same flawless AI execution, but Firm B has a human show up at your door with a thank-you note from the company, which firm do you buy from?

So firms will continue to hire humans to improve things in ways humans best can.

The shape of jobs will likely change, but these dynamic systems will adapt and humans will remain employed and competitive.

Companies are competitive; humans help companies win.

More, smaller firms

Okay, Drew, what if each firm needs way fewer humans? What then?

I think it's simple: people will start their own firms. Or join smaller firms.

The world is not short of problems, and people will always trade something of value to solve them.

Smaller companies can exist in a world with more AI because the startup cost is much smaller. Employees are expensive.

And when you do hire employees, they become indispensable. They know everything about your niche customers, the weird inventory system, and where all the skeletons are buried.

And because of this, you're much more likely to cut them in on ownership. Maybe a full founder's share is too rich. 1%? 5%? I think so.

And small focused teams have always out-executed large lazy teams.

By the way, this is already showing up in Stripe's data, which shows solo founding is at an all-time high.

AI makes starting new businesses cheaper and easier.

Maybe I'm just an optimist

You might read all of this and say "Drew, this time is actually different. I understand you're pulling from every technological change from the past, but this time is actually different."

I think you're right. I do think this time is different. They're always a little different. But consider this:

One of the major changes in agriculture that the exhibit documents is technological advancement, with exhibits and examples from every era of American history. In the 1700s nearly 80% of the population was farmers but by the 1900s the number was halved to just 40% of the population. Today, the percent of American farmers in our country has decreased to less than 2%, indicating the increasing efficiency of the agricultural industry. usda.gov src

The system adapted, people found new ways to provide value, and I still know farmers who take a lot of pride in the work they do. All of these things can be true.