XING

The dirty secret of the software engineering profession is that we have gotten away with barely working software for years, decades, even generations. Amazing people have contributed in amazingly altruistic ways to improve that situation! They fight against the fact that software is malleable, flexible, fixable, cheap (at least compared to hardware) post production. Shipping deadlines are met and maintenance cost is a fact of life that’s ignored. Enter: AI. Does anyone think this helps software engineering teams ship a higher quality end-user product in contrast to shipping more product? The code was already on the brink, quality control on the brink of failure and unable to cope. In the few good hands, AI is a power tool. But we never got to the point where code was unmistakeablely easy to write in the first place. So the so called democratization of software development is just a euphemism for more junk, faster, cheaper software at scale. What we are producing now is the ultimate manifestation of fast-food software. Correct me where am wrong.

Hej everyone, I wrote this little reminder article - mainly for myself, but you might be interested as well when you are working with AI Coding Agents and their setups. So here are the principles that I follow: https://lnkd.in/dbinevsR If you have additions or see things differently, let me know and let's discuss. I'm all in for continuous improvements ...

What I’m learning while building software with AI After spending the last 10 years in management roles, I found myself wanting to get my hands dirty again — at least enough to test something quite practical. After a few weeks building with AI, I’ve noticed something I didn’t expect. The time is not where people think it is. Not in writing prompts. Not in generating code. Not even in debugging. The real time sink is somewhere else entirely: figuring out what you actually want. I’ve had moments where the model gave me something “good enough” in seconds… and I still spent hours going back and forth. Not because the model was wrong. Because I wasn’t clear. What I thought was a coding problem was actually a thinking problem. And AI makes that painfully visible. Before, you could hide behind implementation: writing code, refactoring, “making progress”. Now, if your intent is vague, it shows immediately. The model will happily give you a solution. Just not necessarily the right one. So my fifth practical takeaway is this: AI doesn’t remove the work. It removes the illusion. You don’t spend less time. You just spend it somewhere else: defining, deciding, correcting, aligning. And if you’re not ready for that shift, it feels like the tool is failing you. But it isn’t. It’s just exposing where the real work was all along.