george_ciobanu

the proxy uses a cheap, small model (like gpt-5.4-mini by default) behind the scenes to save tokens on the expensive main model.

Because the proxy has a little bit of overhead per turn, the break-even point depends entirely on session length.

Short sessions (e.g., 2 rounds): The proxy's overhead might actually cost you more than you save.

Long sessions (e.g., 69 to 190 rounds): The token savings on the main model are massive and completely dwarf the small model's overhead.

It's not a universal win for quick, one-off queries, but the math becomes highly favorable on long, complex debugging sessions.

What was different than you expected? Would love to get your feedback in detail, positive or not. George (at)human(dot)software

Thank you - we both map out the dependency graph :-) I'll check yours out too asap!

Haha we literally posted at the same time - very cool, also check out mine in the same vein: getpando.ai

I'll be online for the next 3-4 hours, ask me anything!

If you have long chats scrolling becomes tedious. So I made a simple, clean UI to easily navigate. No data sent anywhere, no login required, 100% free (I can share code if interested). Feedback much appreciated.

I built a 100% composable SQL at www.human.software

I think it's about the ability of someone who perhaps has a great idea (imagination) but lacks time, resources or the skills (execution) to make it happen. Of course that a talented producer will create a more compelling song (at least for now) but if the tool is an amplifier it should also boost talented composers - their prompts or inputs to prompts might be much more detailed, more interesting, more creative than those of a neophyte. This of course makes some assumptions but I think the draw for most people is that they can "make" something that sounds cool with almost no effort. My belief is that someone who puts more effort in a GPT and has more expertise can get a lot more out of it as well. I could of course be wrong and GPTs might be the big equalizers, but I doubt it.