• Regardless of AI hype and a real slowdown in tech hiring, the variety of working software program builders in the USA elevated between April 2025 and April 2026.
• Software program job openings elevated 14% year-over-year. The tech unemployment spike of early 2025 seems to have been a correction, not a collapse, although progress got here partly from non-tech sectors like building and healthcare.
• The query isn’t whether or not AI will get rid of builders, it’s what we’ll name the work. “Pc programmer” has already declined 30% since 2018. The identical cultural drift may retire “software program developer” no matter how a lot coding work people nonetheless do.
I received a cheesesteak this month.
A 12 months in the past, Ohio-based investor Brian Brackeen and I made a guess: Would there be extra software program builders working in the USA in April 2026 than there have been in April 2025? Brian, clear-eyed about AI’s capacity to put in writing code, guess no. I guess sure. The loser owed the winner a cheesesteak from Angelo’s in South Philadelphia.
Earlier than our Builders Convention drew 350 entrepreneurs and financial builders from 35 states to Philadelphia, we settled up. Image Brian and me standing within the rain with an economist on speaker cellphone.
Opposite to the headlines, software program developer openings are up roughly 14% year-over-year, Ashley Putnam from the Federal Reserve Financial institution of Philadelphia shared, serving as our official arbiter with job posting knowledge from FRED. A Federal Reserve paper printed earlier this 12 months confirmed the development, displaying that more people held software developer jobs in 2026 than in 2025, with progress pushed partly by non-tech sectors like building and healthcare adopting extra in-house growth capability.
Brian purchased me the cheesesteak. I took a giant chunk on ninth Avenue. I’m instructed it was arduous to observe.
Then on stage on the convention he supplied double-or-nothing: Would there be extra software program builders in 5 years than at the moment? I took the guess once more.
Right here’s the factor: I don’t assume I’ll win or lose primarily based on whether or not AI displaces human coders. I believe I’ll win or lose primarily based on whether or not we nonetheless use the time period “software program developer.”
Job titles are cultural artifacts. They drift. A decade in the past, “pc programmer” was nonetheless a standard function. Now it’s declining to zero. In Baltimore-area faculties, for instance, pc programming diploma graduates dropped from over 100 in 2018 to precisely one in 2021. None since.
The work didn’t change as a lot because the naming conventions did.
5 years from now, I guess individuals will nonetheless be doing work that we might acknowledge as software program growth — designing programs, writing logic, debugging, transport merchandise. However they could be doing it with a complete new era of instruments, like AI pair programmers, natural-language interfaces, automated testing and code era that feels extra like prompting than typing.
Two issues can occur. We are able to slowly and even dramatically redefine what we imply by “software program developer” to incorporate this new toolkit, and extra individuals will maintain the title. Or, for cultural causes I can’t predict, “software program developer” will begin feeling as dated as “pc programmer” does now. In that case, fewer individuals will declare the title whilst extra individuals do this type of work.
If I lose Brian’s 5-year guess, I don’t assume it is going to be as a result of AI destroyed the sector. I believe it is going to be as a result of we modified what we name it.
The Jevons paradox is the historic sample I’m betting on
Within the nineteenth century, economist William Stanley Jevons noticed that as coal engines grew to become extra environment friendly, coal consumption went up, not down. Effectivity made vitality cheaper, which elevated demand. The identical sample exhibits up in gasoline economic system: higher mileage requirements correlated with extra driving, not much less. Any environmentalist wrestles with the phenomenon.
Apply that to software program. If AI instruments make growth quicker and cheaper, corporations might not shrink their engineering groups — they could increase their ambitions.
Extra merchandise. Extra options. Extra inner instruments. Extra automation. The urge for food for software program isn’t fastened. The constraint has been the price and time required to construct it. Decrease these constraints and also you may get extra demand for the people who direct the work, not much less.
The American economic system was as soon as once more extra productive in the first quarter of 2026, in line with this month’s newest launch from the US Bureau of Labor Statistics. In comparison with the 2010s and the primary half of the 2020s, industries as far-flung as retail and mining, building and property administration all noticed significant features in labor productiveness per hour labored (see chart above). Large administration consulting firms like PwC and the legal industry have reported steady gains in recent years too.
The deployment of software program and knowledge science, which pertains to how AI instruments work, accounts for a few of these features. Productiveness can justify extra employees, not fewer.
That’s the optimistic case. The Fed paper that confirmed my guess win additionally included a warning: Even with total progress, the researchers noticed “softening” within the function kind. The roles are there…for now. On the Builders Convention, Fed economist Adam Scavette shared an evaluation of Swedish employees displaying that AI deployment seems to be rewarding skilled employees, and forsaking junior ones.
For many years, latest faculty graduates (these aged 22-27) almost all the time had decrease unemployment charges than the general economic system. That began altering in 2019 and, in line with Federal Reserve Financial institution knowledge, the discrepancy has grown since. Now, the graduates who booed AI-cheering commencement speakers are becoming a member of an unemployment fee (5.8%) near double that of older faculty graduates (3.1%). I acquired tons of of shocked feedback after I posted final 12 months that the unemployment fee for latest pc science graduates was larger than for philosophy grads.
Their financial outcomes rely much less on AI than on what we collectively select to do about its disruption.
An evaluation by financial growth membership org IEDC earlier this 12 months put it bluntly: AI is exposing a jobs system that was already broken. Solely 40% of Individuals at present maintain what researchers name “high quality jobs” with truthful pay and advantages. Fewer than 10% of states have robust AI-to-jobs methods. Tech has by no means had very many apprenticeships, so the already networked get a lot of the entry.
As Marek Gootman from Brookings put it at our convention: “We essentially mishandled transitions like this previously in conventional industries.”
The query isn’t whether or not AI will change work. I’ll. The query is whether or not we’ll construct programs that assist individuals adapt, or whether or not we’ll go away them to determine it out alone.
So sure, I took the 5-year guess. I believe there shall be extra software program builders in 2031 than there are at the moment — IF we hold calling them that. Who shall be doing these jobs, and what they seem like, is anyone’s guess.
This week, Brian texted me two screenshots: Certainly knowledge displaying software developer listings climbing again and the Github COO confirming that the code-tracking platform hit a billion commits in 2025 and is now on tempo to beat 14 billion in 2026.
His playful message: “Appears like I’m going to lose the following guess too!”