6 Comments
User's avatar
Neural Foundry's avatar

Your dynamic model switching logic is brillant for handling cost unpredictability. The mistake wasnt the Pareto assumptoin or the thin margins, it was not pricing in your own labor cost as real COGS from day one. When you're subsdiizing with time instead of capital, you're essentially running negative gross margins while only counting token spend. Every founder underprices their own hours until they hire someone to do the same work.

Expand full comment
Imasuen's avatar

Next time you want to raise cost. Lmk. I'll help you calculate it instantly and be in charge of sales copy.

I used your telegram bot. It was crazy cheap.

Expand full comment
mogwai.'s avatar

LOL thanks. you sound deeply offended by my prior strategy lmaoooo.

Expand full comment
Kehinde Adeleke's avatar

Good lessons here. I’m building a chatbot for my website i’ve decided that rate-limiting is the way to go to reduce API spend. 10 messages per day so people don’t use assistant as a general purpose answer machine.

Expand full comment
mogwai.'s avatar

yeah, rate-limiting is definitely the optimal way to go for chatbots. just tough to do in a world where frontier models aren’t rate-limited.

and the more specialized a chatbot gets (eg customer support chatbot), the more counterintuitive it becomes to rate-limit: proof of satisfaction is literal satisfaction.

Expand full comment
Chegbe's avatar

Knowing how you kept the bot running makes me even more appreciative of all your efforts. Thank you for putting in the work.

Expand full comment