This article finally discusses the new model, but uses as examples new products with less uptake than incumbents without enough time in the market to definitively say they're successful. This is no different than people making grand claims about AI coding and discussing the sheer amount of code produced divorced of actual outcomes for the business the code is emitted for.

I don't like mistaking work for outcomes!

> "200,000 credits at $0.03 per credit with a 15% volume discount, plus 10 seats, plus a base platform fee, with credit rollover into Q2" — that sentence cannot exist in most billing systems.

I used to work on GCP billing and this is so entirely correct and disgustingly hard to get right. The worst part is that even if you implement a way for sales to create these custom contracts down to a T, you're going to get an escalated customer ticket that hits the engineering team because the bill didn't end up how they think it should have so it's assumed to be a bug. You'll confirm this rube goldberg machine of billing state in their contract was in fact correctly implemented and that maybe they didn't negotiate what they thought they did with sales.

The core premise is true, but this is AI slop. Lots of 'it is X, not Y' in here.

The model seems to have imagined we are already in a place where everything is happening via API/MCP with agents, and that is not the on-the-ground reality in companies today at all. The 'agents just run your business' plan has not been cracked, not even close. It seems to mostly be people extending their existing roles with AI so far.

> Ten people with AI can do what a hundred people did before, so headcount stops scaling with output.

Citation needed