Software developers are among the first professionals to feel the full force of generative AI at work, and the experience is proving more complicated than a simple productivity gain. Coding assistants such as Anthropic’s Claude Code can now finish in minutes tasks that once took days, but a growing number of experienced engineers say the speed has brought an unexpected sense of loss along with it.
The mood was captured in a January 8 report by Fortune, which gathered accounts from engineers and founders about what happens to professional identity when hard-won skills become cheap to reproduce.
Experienced engineers describe a shift in their work
Gergely Orosz, a former Uber engineer who writes the widely read Pragmatic Engineer newsletter, has said publicly that he expects most of his code in the future to be written by AI, a shift he ties to the arrival of agentic tools such as Claude Code.
Andrew Duca, the founder of a tax software company, told Fortune that the technology is remarkably powerful, yet it also produces a kind of sadness: expertise that took thousands of hours to build is quickly becoming commonplace.
Erik Meijer, a former Meta engineer known for his work on programming languages, told the outlet that Claude Code has, in a matter of months, pushed the state of the art in software engineering further than decades of academic research. He also described losing the “flow state” that made programming absorbing, and said he has taken up music to recover that feeling elsewhere.
Agents still fall short on real jobs
The emotional shift is running ahead of what AI systems can actually do on their own. The Remote Labor Index, a benchmark published in October 2025 by Scale AI and the Center for AI Safety, tested AI agents on real freelance projects drawn from online labor marketplaces, with actual briefs and deliverables.
According to the researchers, the best-performing agent completed only about 2.5 percent of the projects to a standard a paying client would accept. Many outputs looked plausible at first glance but contained internal flaws, missing pieces or basic technical problems that a human would have had to fix.
The finding suggests that, for now, the tools are reshaping how developers work rather than removing the need for them. The judgment required to specify a task, review the result and catch subtle errors remains a human job.
Wider effects on skills and junior roles
Fortune also pointed to research warning that heavy reliance on AI without proper training can erode skills and weaken decision-making over time. Low-quality AI-generated material, sometimes called “workslop,” has been reported to affect roughly 40 percent of employees who have to deal with it, and several analyses have noted declines in junior hiring in fields with high AI exposure. Those figures come from the studies cited in the report, not from independent measurement.
Taken together, the accounts point to a profession in transition. Output is rising and the range of what a small team can build has widened. At the same time, the satisfaction that came from mastering a difficult craft is being renegotiated, and engineers are working out what their expertise is worth when the mechanical part of the job can be delegated to a machine.
