Software engineering was one of the best-paying professions in the US in 2022. Now, according to the account at the center of this story, the arrival of AI is disrupting the work itself, contributing to layoffs and underemployment, and forcing at least some engineers to change how they spend their days.

Matt, a software engineer who asked not to use his real name to protect his employment, looks forward to a very specific ritual: a four-hour train commute every weekday from his life to Pawling, New York. That commute is not downtime. It is time to work on a browser-based video game he builds on his own, writing every line of code himself. His stated goal is blunt: keep his “axe sharp,” because he believes the job’s direction is weakening key skills.

In the last six months, Matt says his work has increasingly shifted away from coding, problem solving, and software architecture. Instead, he spends more time reviewing code generated by artificial intelligence. That change sounds subtle until you zoom out. Software engineering has always been more than typing. It is designing systems, reasoning about tradeoffs, and building problem-solving muscles that show up when requirements change, bugs appear, and production incidents hit.

Matt is trying to counter the skill drift by doing what he can to keep development fundamentals active. He says he is actively trying to keep his axe sharp. He also says, “I am trying not to leverage AI where I can.” In other words, he is not just learning a new tool. He is deliberately choosing where AI stays out of the loop, in an attempt to preserve the kind of hands-on experience that his day job is slowly taking away.

For executives and boards, the second-order effect is uncomfortable: AI does not only change output, it changes role design. When AI can generate code, companies can often speed up certain tasks. But when work shifts to code review, the competency mix changes too. Review is valuable, but it is a different skill set than building from scratch, and it tends to elevate questions about quality, security, and correctness that must be handled by the team.

There is also a workforce reality baked into Matt’s experience. The article frames the broader labor market shift as disruption that has led to several layoffs and underemployment. That matters because underemployment tends to create volatility inside organizations. Skills get stranded, morale takes a hit, and recruiting becomes trickier: if engineers believe their craft is eroding, retention and engagement are harder, and the cost of replacing talent rises even if AI reduces the marginal cost of certain deliverables.

Now layer in the regulatory backdrop, because AI in software is increasingly not just a technical question but a governance question. Even without getting into new policy claims beyond the scope of this account, the direction is clear across the industry: organizations are under pressure to document how systems are built, how decisions are made, and how risks are managed. When code is generated by AI and then reviewed, governance needs to track provenance and accountability. That pushes engineering teams toward processes, audits, and controls, which again reshapes jobs. Matt’s description is personal, but the pattern aligns with a governance-heavy world where it is not enough to ship; you also need to show your work.

Collective action is also part of the broader theme signaled by the original piece. When engineering roles shift quickly, the consequences are rarely confined to one person’s desk. Teams adjust hiring and training, managers renegotiate what “good” looks like, and leaders decide whether the organization is building depth or just accelerating throughput. Matt’s commute game is a small, almost stubborn counter-move: keep coding to avoid becoming only a reviewer.

The strategic stakes for leaders are straightforward. If AI changes what engineers do, your talent strategy has to change too, or you risk hollowing out core capabilities at the exact moment you need them most. Matt’s experience is a reminder that the human skill layer does not disappear. It either gets maintained or it degrades. And in a labor market already showing layoffs and underemployment pressures, that difference can determine whether a company stays resilient when the next product requirement, incident, or architecture decision lands.