Evaluating AI Use in Our Company

This is probably the first thing I’ve written without prompting Claude or ChatGPT. That aided by the lack of overall reading outside of that app has left me pretty forgotten how-to-think-and-write-my-own-stuff.

It wasn’t always like this. 3 years ago all I was guilty of was copying StackOverflow solutions without reading them sometimes. But last year it came to a point where we stopped doing pull requests all together to avoid “friction” and tried to follow a more “iterative” process. I tried to go all in on AI and instead felt like a shell of a developer. Rework cycles were longer. Our QA was getting more and more frustrated. We were done with 50% of a project in the first two days and then it took more than a month to manually fix all the “small bugs” to take it to completion.

Dev morale sucked. Junior devs felt like they were going to be replaced. They were learning to not trust their own instincts and instead blindly follow Claude suggestions since “it was smarter than them”. Was it really? Even if we spend 200$ in credits we were unable to speed up projects without drastically losing quality. AI has the “move fast, break things” approach which doesn’t cut it with our designers.

I would call this the effect of the AI psychosis induced by all the marketing we see. This marketing aims to ruin our perception of our own abilities and instead blindly trust something much dumber than us. I get clients saying things like we can do this website/app in 20 days with AI. And then proceed to not launch for 6 months after I reject them.

In Hinduism there is a concept of Maya - roughly translating to illusion. I heard a Sadhu call AI “Maya within Maya”. And it really is. It’s an illusion of human intelligence which can never be the real thing. To discount our own ideas for something as bland and uninspired as AI would be a disservice to ourselves. This year I’ve decided to take the company in a different direction. We will prioritise human intelligence over any AI. We will build talent pools and processes iterating from the pre-AI era. This will allow us to truly improve as developers - thoughtfully write our code - and work with artists/designers who have the same views as us. I really value this craft I have spent years of my life honing. And I hate to see it be called “easy to replace” when it clearly isn’t. It’s a version of the same hatred backend developers had for frontend, calling our jobs easy when they couldn’t even center a div.

Read the original post on Substack: Click Here