LLM-based AI coding tools are pretty seductive. As a seasoned developer I was sceptical but they’re part of my workflow too. Why? Because features appear fast, demos seem to work, progress feels real, and they seem so much cheaper than hiring “real” developers. It’s win-win. Isn’t it?
When you look past the hype, past those so bought in that they’ve lost all objectivity, or those with AI-coding tools to sell you, the picture is far less clear.
Some signs to watch out for:
- Things go wrong in ways nobody can explain.
- Bugs resist fixing or re-appear.
- Simple changes break unrelated features.
- Features appear more and more slowly.
- Your product becomes too complex.
- Users notice the product is getting slower.
- Experienced developers don’t want to work with your codebase.
By the time you’ve accumulated two or three of these, your problems will be well on the way to compounding and when you get to the last one it may be too late.
This is “vibe coding” - building software without deep product & technical judgment in the loop. The AI handles the coding fine but it makes poor software architecture decisions that will hurt you when you can least afford it: when you’re scaling, fundraising, or trying to exit.
I’m no ‘developer turned manager’. I’ve been building software since the 1980s and actively write code today. I’ve worked as a CTO, a strategic advisor to leadership teams, and run my own business. I know good architecture and bad architecture and how to tell the difference.
There’s never been more pressure to act in a hurry than there is today. If you’re making significant bets on software, employing AI tools as part of your process, and want more confidence that you’re not accumulating invisible risk, I can help with that.
A 45-minute scoping call will tell us whether there’s a problem and how serious it is. If we decide to work together, engagements start at £150/hour.
Email: self@mattmower.com or DM me on LinkedIn: https://www.linkedin.com/in/mattmower/