pricing and unit economy is the key for the future of ai products
with low margins and usage-scaled, costs both entrepreneurs and investors will start approach software like a completely different kind of business – more like retail, where physics and logistic really matter
no more money out of the void
should founders look at their ai businesses as arbitrage games?
i mean → find smth ai can do so valuable for customers so they're ready to pay much more than you'll burn on crazy LLM costs
do it fast until the customers figure out how to do it on their own :)
every ai startup turns into a cloud infra business, even if they don't realize it
making an ai product = letting your users run their agentic workflows on the AI compute YOU pay for
AI flips software economy
Previously: build time (create software) was expensive, run time was almost free
Now: build time is almost free while run time costs scale crazily (the more your product is used the more you SPEND)
@pmitu Depends on the use case:
For me (doing everything from app dev to data analysis or homework) – it's Claude Code, most powerful of universal ones
Fast iteration isn’t about better prompts. It’s about better feedback loops.
Our loop in Claude Code:
- new branch per experiment
- define goal as a /command
- command must ask for feedback, reflect, and rewrite itself
Human + AI learn together. Iterations compound.
Vibe coding can’t kill product engineering
Vibe coding simplifies simple “CRUD-like” apps
However, as more and more capable AI technologies emerge, the frontier of what’s valuable shifts to more and more smart, autonomous, and capable apps and agents
Developing them requires good old strong engineering mindset, regardless of how goals, constraints, hypotheses, tests, iterations are expressed - in code or in English