Programming with AI


In the past three months I have been programming a lot with AI. A lot.

Among the results are the releases of rcurses and RTFM.

Let me share a few conclusions. The positives:

  • When I program with the help of AI my efficiency is 3X-4X
  • While I have programmed in Ruby since 2003, the past three months has made me an 80% better Ruby programmer (check my code for evidence)
  • The coding is much faster, the code is far better and I learn a lot more in a shorter time

The “beware”s:

  • I have almost lost control of my code along the way
  • AI tend to over-engineer almost everything
  • AI can get stuck into a narrowing spiral of knuckleheadedness
  • Good prompting is key, just like communication is key when picking up girls

The future:

  • Programmers are doomed, in a few years there will be no professional programmers
  • But designers, communicators, empathy and understanding will prevail

When you talk to grandma, you communicate. You prompt. And she prompts. Prompting is communication. When you prompt an AI, explain what you want as output, how you would like the output, what role should create the output (a senior engineer, a software designer, an application tester,…). And just like with people, focus on one thing at a time. In the future, good developers will be those who are excellent communicators. Those who can pick up girls. Or boys.

Being polite and courteous to the AI seems to work.

For a current set of tips on how to program with the help of AI, check out this post by Andrej Karpathy.

Another tip is how to stop a narrowing spiral of the AI banging its non-existing head against a virtual wall desperately trying the 20th time to fix something in an even more convoluted way: Give up that context window and start anew. And before you throw in the towel, ask that context window for a really good prompt for the next context window. I learned this one after trying to fix an rather complex issue in rcurses and after at least 200 screen heights of context window until Firefox gave up.

This blog post is probably outdated before summer. The development of AI is silly fast, and you’ll be smart to try to keep up. Those who go “nah, AI’s not my thing” will soon be left behind. AI will soon be so ingrained in everything we do that a question like “Did you ask?” will literally (yes, a correct use for that word) mean “Did you ask AI?”.

PS: The image used for this post was created by ChatGPT 4o with the prompt “Create a realistic image that I can use as an illustration for this blog post:” and then I simply added the above blog post text to the prompt.


Link to this post: https://isene.org/2025/05/AI.html