The AI madness continues with IBM stocks taking a hit after MisAnthropic announce a COBOL Tool for Claude.
IBM have a lot of customers running old COBOL programs. Said customers happen to be some of the biggest banks and government organisations in the world and as such they’re very risk averse (hence still running COBOL programs written on the 1960s in 2026). IBM themselves have been trying to automate the migration of COBOL programs to cool/trendy languages like Java or JS for decades. The problem is it just isn’t that simple. Last year universally reviled Elon Musk made the headlines for trying to use DOGE to do exactly this with social security and royally messed up. Dan Hon’s write up for it in the MIT review was pretty spot on:
A long time ago I worked on a mainframe system in California—the kind that uses COBOL. It was as solid as a rock and worked day in, day out. Because if it didn’t, and reimbursements weren’t received for Medicaid, then the state might become temporarily insolvent.
That’s why many of the rules about technology in government make it hard to make changes: because sometimes the risk of things breaking is just too high. Sometimes what’s at stake is simply keeping money flowing; sometimes, as with 911, lives are on the line.
Move fast and break things may be a fine strategy for your website for perving on fellow college students (until you inadvertently break democracy oops). However, institutions underpinning the modern economy don’t have that luxury. The CIO or CTO that takes a risk and loses millions of dollars for their bank, or even worse the taxpayer, would never find another job as a technical leader.
These organisations, who have refused any kind of change to their system for 30 years due to the risk are not seriously thinking “if only we could have an AI vibe code a solution in react and nextjs.”
What exactly is a Claude Skill? It’s essentially a glorified prompt in a markdown file. Some of them bundle up supporting scripts and resources but it’s essentially the next iteration of “subscribe to my newsletter and I’ll share my foolproof prompts for ChatGPT”. This is not some magic new model with any solid proof that it works. Investors are once again demonstrating the irrationality of the rational market. Why? Because Mindless Business Automatons can’t understand the difference between flashy proofs of concept and the reality of complex systems. The AI hysteria continues apace…