In my last post, I talked about a developer who chose to stay with Delphi because scarcity paid well. A rational short-term decision that I believe became a long-term trap.
Earlier this year, Anthropic announced that Claude Code can automate the most labour-intensive phases of COBOL modernisation, the kind of work that used to require armies of consultants spending years mapping legacy workflows.
IBM's stock dropped 13% in a single day. $31 billion in market cap, gone. By the end of February, shares were down 27% for the month. The worst monthly decline since 1968. (A US government quantum award in May 2026 drove a partial recovery, though the stock remains below its pre-announcement level and the recovery was unrelated to the COBOL thesis.)
Not because a product failed. Not because revenue dropped. Because AI demonstrated it could do the work that an entire consulting ecosystem was built around.
COBOL developers, like Delphi developers, commanded premium pay because of scarcity. The market rewarded what few people could do. Until the moment it didn't.
But the story that should really keep you up at night isn't IBM. It's a software engineer at a large San Francisco tech company, speaking anonymously to the SF Standard earlier this year.
He described himself as "basically a proxy to Claude Code." His manager tells him what to do, and he tells Claude to do it. His predominant feeling? Grief. The skill he spent years developing has been commoditised.
His backup plan? Move to Yosemite and become a park ranger.
That's someone who followed the playbook, degree, skills, career, and watched the ground shift beneath him.
This is the part people get wrong about technological disruption. It doesn't punish the incompetent. It punishes the inflexible. The capable people who built real skills and then assumed those skills would remain valuable forever.
The developer I interviewed years ago wasn't bad at his job. The COBOL consultants weren't bad at theirs. The engineer in San Francisco isn't bad at his. They were, and are, skilled professionals caught in a pattern as old as technology itself: the premium for scarcity flipping into a penalty for obsolescence.
The only variable that's changed is the speed.
So what do you do about it?
You don't panic. You don't throw out everything you know. You start treating AI as a skill to develop, not a threat to resist.
Pick one workflow this week. Code generation. Test scaffolding. Spec drafting. Pipeline creation. Force yourself to do it with AI, start to finish. If the output isn't perfect, iterate. If it frustrates you, ask yourself honestly whether that's a real limitation of the tool or your own resistance to change.
The engineers who will lead in five years aren't abandoning their craft. They're elevating it, moving from writing code to directing systems, from manual execution to architectural thinking.
The window is still open. But every week, it gets a little narrower.