Local AI as Personal Stack Overflow

EricMesa

October 25, 2026 - 4 minutes read - 783 words

If you’ve been programming long enough, you know the old joke: the programmer needs to do a task and they aren’t sure how to code it up. Surely at this point someone has already asked how to do it on Stack Overflow. The programmer goes to Stack Overflow and may not even understand how the code works (depending on the joke teller’s opinion of this programmer) and copies and pastes that code into the production code. It’s common enough to become a meme:

Shoot, that meme even became so huge that Stack Overflow actually made it into a real product that people could buy. I know over the years that I have found myself to Stack Overflow when I needed to use some obscure library. More often than not, I needed a bit more customization than I could get by simply copying and pasting. So I really cherished when the answer to the question contained an explanation and not simply “here’s some code”. I mean, besides needing to understand how the thing worked, it’s definitely not best practice to copy random code off the Internet. That’s how

rm -rf */*

Tuns into the reddit post “Help! I’ve deleted my whole computer!” But, it’s true that, sometimes, when needing some really advanced feature of a programming language (I’m an electrical engineer who just so happens to love programming), I’ve had to copy some code in and go into a modify, compile, test loop.

So how does this all tie into AI?

Well, there’s a huge debate now about how much AI to use in programming. (Putting aside the data center issue - which is why this post is about local AIs) There’s everything from vibe-coding to carefully managing AI-created PRs to having AI find the bugs, but having humans do the actual coding. Outside of the “No AI” absolutists, there are pros and cons to each use of AI along that spectrum. But as I was working on improving the code I use to generate my end of year video game blog post (it calculates time played to give me a list of most played to least played games - see last year’s post) I came up with what should probably be the least controversial use of AI with programming.

I use a local MongoDB database to store the data on how long I spent playing each video game. Some time 5 or more years ago I learned about MongoDB and how to use it with Python. Since then it’s just been working fine so I haven’t had to touch that part of the code. To be honest, I haven’t touched most of the code except for little bug fixes or feature enhancements at the edges. So I forgot most of what I used to know about interfacing with MongoDB. Frankly, even though MongoDB queries are simpler than MySQL queries, most database engines are annoying to interface with if you don’t do it all the time. So I used my local AI, not to do the code edits for me, but to ask it - “how do I do a query for such-and-such a field?” And it gave me an answer, but I needed to get a more detailed answer based on the schema I was using. So I was able to do a bit more of a back and forth. Once I knew the right query to run, I was able to move on with the refactor I wanted to make.

In my mind, this is no different than using Stack Overflow to remind myself of how to use a feature I barely use. Only instead of wasting time searching and hoping someone gave an answer that is either exactly what I wanted to similar enough - the local AI could be that answer and, if I needed clarification it could provide it now, not tomorrow or days later. It didn’t do the coding for me or anything else. Once I had that answer I still had to apply it to the work I was doing, leaving me without the AI brain rot that some researchers are claiming occurs to when we leave too much up to the AI and stop remembering how to do things.

My thoughts on AI continue to evolve as I learn more and listen to some trusted sources on how they are using it and learning from their experiences. I imagine I will continue to change my mind as I gain more experience, but this seems to be one of the most benign uses of AI and something that everyone can benefit from if they have the resources to run a local instance of an LLM-style AI.