Some thoughts on AI and coding

Holy cow, things are moving fast in the software business—ideas are coming like an oil gusher, and keeping up is a challenge. Here are a few observations, amazements, puzzlements, and prognostications regarding AI and software development that have occurred to me recently. 

Vibe coding for the win

Vibe coding has come a long way in the six months since I first tried it. I recently picked up the same project that I (or rather Claude Code) had built, and I was shocked at how much better the agent was. In the first go round, I had to keep a close eye on things to make sure that the agent didn’t go off the rails. But this time? It pretty much did everything right the first time. I’m still stunned. I suspect it’s going to be a while before the amazement wears off.

One thing that is absolutely fantastic about vibe coding is debugging. Those cryptic error messages that take human programmers minutes or sometimes hours to run down can be deciphered and debugged by AI in seconds. I’m now at the point where I don’t even ask the agent about the error message; I just enter it, and it automatically identifies the problem. A great example: package dependency hell. No human can decipher the deep dependency chains created by our applications today, but AI can untangle—and fix!—these issues without missing a beat.

I expect we will see an explosion of what might be called “boutique software” as a result of vibe coding. There are endless ideas for websites and mobile apps that never got written or created because the cost to produce them outweighed the benefits they promised. But if the cost of producing them is drastically reduced, then that cost/benefit ratio becomes viable, and those small but great ideas will come to fruition. Prepare for “short form software,” similar to what TikTok did for content producers.

Software development is uniquely positioned to take advantage of AI agents. Large language models (LLMs) are—no surprise—based on text. They take text as input and produce text as output. Given that code is all text, LLMs are particularly good at producing code. And because computer code isn’t particularly nuanced compared to spoken language, AI easily learns from existing code and thus excels at producing code. It’s a virtuous cycle.

Software development futures

The previous point creates a dilemma of sorts. Up until now, humans have written all the code that LLMs train on. As humans write less and less code, what will the LLMs be trained on? Will they learn from their own code? I’m guessing what will happen is that humans will continue to design the building blocks—components, libraries, and frameworks—and LLMs will “riff” off of the scaffolding that humans create. Of course, it may be that at some point AI will be able to learn from itself, and we humans will merely describe what we want and get it without worrying about the code at all.

It seems kind of nuts to put limits on what AI can do in coding and software engineering. “We’ll always need software developers” is easy to say, but frankly, I’m not so sure it is true. I suppose it was easy to say “We’ll always need farmers” or “We’ll always need autoworkers.” Although both of those statements are still true, there are a lot fewer farmers and autoworkers today than there were decades ago. I suppose there will always be a need for software developers—the question is how many.

Hidden Figures is a beautiful movie about a group of Black women who were instrumental in getting the early US space program off the ground. They were called “computers” because they literally computed trajectories, landing coordinates, and all the precise calculations needed to safely conduct space flight. They did heroic and admirable work. But today, all of those calculations can be done with a Google spreadsheet. I think that AI is going to do to software developers what HP calculators did to human computers.

At this point, the only thing I can predict is that no one has a clue where software development is headed. AI is such a strong catalytic force that no one knows what will happen next week, much less next month or next year. Whatever does happen, it is going to be amazing.