Full image ↗
Programming
Thirty years of code, and then the AI arrived. What actually changes when you have an assistant that never sleeps and never judges your absurd variable names.
Full image ↗
I wrote my first program in 1990. It was called something like CALC.BAS and it added two numbers together. Since then I have been through BASIC, Pascal, Delphi, VB6, VB.NET, SQL Server, PHP. Thirty years of building software, of fixing bugs at eleven at night, of searching Google for answers that did not exist yet.
Then Claude came along.
When I first heard about AI for programming, my reaction was sceptical. I had tried GitHub Copilot: useful for trivial completions, useless for real problems. I assumed Claude would be the same thing with a wordier interface.
I was wrong.
The difference is not in how fast the code gets typed. It is in the quality of the conversation. Claude reasons with you. It asks questions. It takes in the context. If I say "there is a bug in the blog login, session_destroy is being called when it should not be", it does not answer with a generic template about PHP sessions. It reads the code, finds the operator precedence problem (?? against ===), and explains it with a clarity that a good many human colleagues do not manage.
After thirty years, I had learned to program alone. I solve the problems, I decide the structure, the architectural calls are mine. It is a habitual trade, in the best sense: you have a method, and you trust it.
Claude broke that solitude, and for the better. Now, before writing code, I think it through out loud, in a chat window. I describe the problem, I listen to what comes back, and often I change my mind. Not because Claude knows more than I do about my own application domain — plainly it does not. But because putting a problem into words for an intelligent listener always makes it clearer.
MyShop, the retail management system I have been developing since 1995, runs to millions of lines of VB.NET. I am not rewriting them with an AI. But when I have to add something new, or refactor a module I have not touched in years, having Claude as a second pair of eyes is worth a great deal.
I do not delegate the thinking. I delegate the tedium.
Experience cannot be delegated. Thirty years of debugging, of wrong architectures put right after the fact, of clients with impossible requirements: that knowledge is mine, and no AI can stand in for it. Claude is a remarkable tool, not a replacement.
The gap between an experienced programmer using AI and a beginner using it is enormous. The first knows when to trust it, when to correct it, when to ignore it. The second takes everything on faith.
I correct it often. And I am still learning.