Hey Boss, look here!
I made a magical X-ray app that displays our code as wiring.
Here is the shiny, neat program that Bob wrote 15 years ago, right before he left us:
And here is what it is today, after Tom took it from Bob, and Jane from Tom, and Mick from Jane, and me from him, porting it to roughly four new platforms, and adding about 240 features from customer requests (150 of which were urgent):
Or, ya know, run open source models internally if you actually find them helpful.
Sure, just need $3000 for an extra GPU with at least 24GB ram to load any reliable models. Tested Qwen 3 with my 3060 and it hallucinated a Legoland theme park next to a village with a population of 500 in the middle of nowhere.
I forget what the latest I ran the numbers on was, but $100K up front and $20K per year running costs seemed like it had a decent chance of being useful to our department of 10 software engineers, which isn't bad, but it's an LLM on par with 6 months ago's frontier models.
If I could get those prices for a local instance of Fable 5, I'd be pushing the paperwork to make it happen.