How I Use AI in My Work

I use AI all the time.

I wrote about using AI for programming last year, and my position has not really changed. I use it to help me work. I do not use it as a replacement for understanding what I am doing.

I still do not like the phrase “vibe coding.” I understand what people mean by it, but I do not want to accept code just because it looks like it works. If something breaks, I need to be able to debug it. If a database query is slow, I need to understand why the change might help.

At the same time, I keep seeing people vibe code games in JavaScript and get something functional and releasable in a day. It is not polished, and it probably needs a lot more work, but it exists and people can play it. That is genuinely impressive. The distance between having an idea and having something that works has become incredibly small.

I use AI for optimizing queries, documenting methods, writing boilerplate, translating files, and exploring possible implementations. It is good at giving me a first pass or a second opinion. Sometimes it gives me something useful immediately. Sometimes it gives me something confidently wrong. Either way, I still have to review it.

Using AI to explain an implementation is great too. I can ask it questions, ask for an ELI5 explanation, or keep asking it to explain the bit I do not understand until it makes sense. Having something that will patiently answer those questions is incredibly useful for engineers. It is much easier to learn when you can ask the next question immediately instead of getting stuck searching for the right wording.

Unreal is different

I tried using Claude with Unreal Engine about a year ago. I kept having to correct it on things that someone with real Unreal experience would already know. It would produce code that looked reasonable, but then I would find a pile of errors, performance problems, or decisions that were technically possible but completely wrong for the engine.

That is what makes the AI push around Unreal feel dangerous to me. Epic is adding more AI tooling around Unreal Engine 5.8, and its plans for Unreal Engine 6 point toward an even bigger AI focus. That could be useful, but it also means people may be able to generate a lot of engine code before they understand what that code is doing.

Unreal is too complicated for that to be harmless. You can create something that works in a test level and still have terrible memory usage, bad frame-time spikes, unnecessary network traffic, or a system that falls apart at scale. Those are the kinds of problems that require experience to spot, not just a compiler that says the code is valid.

Financial forecasting

I use AI for financial forecasting, and it is really good at that. I can give it assumptions, ask it to model different scenarios, and have it point out patterns I might otherwise miss.

That does not mean I blindly trust the result. A forecast can be mathematically consistent and still be completely wrong because the assumptions were bad. I still have to decide which inputs are reasonable and whether the output makes sense in the real world.

Localization

For localization, I use AI to get an initial pass done. It lets us get the data into shape quickly before sending it to paid localization professionals or companies later.

That first pass saves time, but it is not the finished product. People still need to review the language, understand the context, and make sure it actually sounds right. I am not going to trust a machine to decide whether a line of dialogue feels natural in another language.

This is how I use AI most of the time. It helps me get from nothing to something. Then I review it, change it, or throw it away.

I do not think AI is going away, and I do not want it to. It is already useful to me. I just do not confuse useful with magical.

I still need to know what I am building, why I am building it, and whether the result is any good.

until next time