Originally written in summer 2025

Experiences with AI in Software Engineering

1.

I wrote some code that converted specifically formatted user inputs into different fragments of that input. I had some sample inputs, and I used ChatGPT to generate several credible variations for me. I then used those inputs to create a a battery of test cases, which in turn caught some cases that my code hadn’t accounted for. This feels like a good outcome.

2.

It is 2:30 am. I am working, as I have been for all waking moments of the last 48 hours, to meet a project deadline. I’m trying to query a database table in a slightly different way than something else we’ve written elsewhere is querying it. My attempts consistently error, because GROUP and ORDER statements were designed explicitly to toy with human ken. In a fit of desperation, I turn to an LLM with the error and describe the results I want to achieve.

After lexically fellatiating me, it spits out an alternative query that doesn’t work, with a different error. Another. . .seven rounds of this ensue: description, fellatiation, slightly wrong answer, with slightly more diverse emoji each time (we started with just checkmarks and ended with :brain: and :soap:). Eventually, we arrive at something that works.

It is now 3:30 am. I move on to the next task. Despite having spent now the majority of my life comfortably working in the small hours to meet project deadlines, I’m struck with an intense feeling of loneliness, greater than anything I’ve previously felt professionally.

3.

We’re having a monthly planning meeting with the whole engineering department, and my boss has just unveiled a plan to have AI tools write our implementation plans for us. I ask him how and why we expect this to improve the team’s delivery. He answers, in front of the entire engineering team, “I don’t know.”

Nobody questions this answer, and the meeting moves on.