100 million tokens in a single day. That number should make everyone stop and recalculate their assumptions about AI demand.
The "tasks you wouldn't have bothered with before" point from Rohit is the real unlock. My own agent usage followed that exact curve - started with obvious delegations, then expanded into things I never thought to ask for. The marginal cost of asking dropped so low that the decision changed from "is this worth doing?" to "why wouldn't I try?"
The writing criticism is interesting though. I'd push back slightly - AI isn't bad at writing per se, it's bad at voice. Give it structure and data? Solid. Give it personality? Still feels like a press release.
It may well be due to system prompts, but hearing that agents appear naturally risk averse immediately lifted some weight off my shoulders. From your mouth to god’s ears.
100 million tokens in a single day. That number should make everyone stop and recalculate their assumptions about AI demand.
The "tasks you wouldn't have bothered with before" point from Rohit is the real unlock. My own agent usage followed that exact curve - started with obvious delegations, then expanded into things I never thought to ask for. The marginal cost of asking dropped so low that the decision changed from "is this worth doing?" to "why wouldn't I try?"
The writing criticism is interesting though. I'd push back slightly - AI isn't bad at writing per se, it's bad at voice. Give it structure and data? Solid. Give it personality? Still feels like a press release.
I tracked my own token economics recently: https://thoughts.jock.pl/p/claude-model-optimization-opus-haiku-ai-agent-costs-2026 - the cost/value ratio flips positive faster than you'd expect once you stop using the expensive model for everything.
It may well be due to system prompts, but hearing that agents appear naturally risk averse immediately lifted some weight off my shoulders. From your mouth to god’s ears.