Confessions of an AI Skeptic

When I first experimented with ChatGPT I was impressed, but convinced we were at least a decade away from AI making a meaningful impact on my daily life.

That changed when I retired in 2024. Navigating healthcare options, IRA-to-Roth conversions, Social Security timing, and retirement tax strategy is genuinely complex, and I found myself increasingly turning to AI for answers. By my count I ran well over 100 prompts working through what was optimal for my situation. AI became my on-demand financial advisor.

Then I heard about "vibe coding" and Claude Code. As a learning experiment, and admittedly because retirement left me with time to tinker, I decided to try building a retirement planning web app from scratch. What happened next genuinely blew my mind.

Within a few hours I had a working framework. Within a week I had a fully functional prototype. That prototype became WhenIm64.

Using Claude Code I am at least an order of magnitude more productive than hand-coding. It's like having a highly competent development team I can direct with plain English. Claude is particularly strong at the things I was weakest at, UI/UX and CSS in particular. In seconds it can restructure hundreds of lines of HTML, make sweeping layout changes, and refactor components that would take a human developer hours of tedious work.

That said, AI is not infallible. It makes mistakes, introduces bugs, and sometimes misses obvious issues. I had to sanity-check every step, occasionally guide the debugging process myself, and keep prompting it to log issues and update GitHub. The developer doesn't disappear. They shift from writing code to directing, reviewing, and quality-controlling.

Where Claude genuinely struggles is with large codebases. Without explicit instruction it doesn't retain how an app is organized, and tends to re-scan and reverse-engineer the existing code each time you tackle a new feature. Context management becomes a real skill.

One thing I'm confident won't change anytime soon: non-technical people won't be building sophisticated applications with today's tools. Without years of development experience behind me, I couldn't have achieved what I did. What I do see is that strong software engineers will become dramatically more productive, and the nature of the work will shift in genuinely exciting ways. The drudge work, typing thousands of lines of boilerplate, contentless code reviews, manual Git housekeeping, largely disappears. What remains is the interesting part: architecture, judgment, and problem-solving.

The future is bright for great developers. There may be some short-term disruption, but the ones who embrace these tools will be formidable.

Originally published on LinkedIn