Lately I have been working with a lot of AI tools for my job @ KernBouw. Mainly using Claude and Codex for making internal tooling. But I have been less sattisfied with building software than I did before. Before AI tools programming seemed very thought trough. You had time to think about problems and time to artistically program the solution. What I mean with artistically? You write the code a certain way, which seems very personal. Just like writing a poem.
What does it look like now?
Well. I try to write some sort of Product Requirements Document or product specification based on a meeting we had with some stakeholders.
Try to define that up to a degree that feels ok. Haha, yess feeling the vibe. Then converting it to an implementation plan and when that plan is ready, I let the AI tooling do the rest.
And surprisingly, almost all of the time this just works!
What does this mean for me? Well…
- I am constantly designing and writing spec documents, which is mentally straining.
- Prompting
claude pls fixon bugs that claude found for me. - Constantly trying new tooling and methods to keep up, which means sometimes focussing more on fixing my own setup and not actually doing any work on the product we are building.
What do I think about this?
This whole AI thing is ultra-utilitarian. We just focus on the output and the way we get there does not matter. This means I get to make the decisions that a product-owner makes. I enjoy this A LOT! It feels like actually solving problems, while not worrying about the technical details. But on the other hand I do not have trust in what I build and do not feel as connected to what I am building as I used to. I do not like slop.
Big note! I have never worked in Software Engineering before AI tools. I do not know what it means to ship code without and never wil. That makes me a bit sad, because a lot of software patterns and technical details i’ll never learn, because AI simply knows and is not ‘worth’ my time. This also raises an internal dillema if it’s worth to become a GOD with AI tools, or that I should still continue in learning this technical details.
People like Kun Chen and Matt Pocock sharing their ideas in using these tools gives me hope.
ALSO: these agents are very good at finding bugs and also fixing these bugs themselves. What do I bring to the table at this point? I want to quickly highlight the existential ‘crisis’ part of being a software engineer.
My perspective for the future is that I’ll be having wayyy more conversations with stakeholders and loosely defining the product, while AI makes almost all technical decisions. How I deal with slop? I do not know yet. What I do know is that it irritates me. That’s why this blogpost will not be run through AI to fix my mistakes. I am also to lazy to fix the mistakes I wrote in this blogpost.