AI-heavy spender

Runs a 6-person applied-AI startup whose second-largest cost line is model inference. Scrutinises anything touching tooling or compute spend and wants cards that can be issued to agents, not just people.

Left a staff engineering role at a large model lab to build agent infrastructure. Burns roughly $28k/month, half of it on inference and GPU rentals across four vendors. Has been burned by a shared virtual card being maxed by a runaway job and now treats spend visibility as a security concern. Reads pricing pages like API docs: looks for limits, receipts, and automation before brand.

Reads docs before marketing; opens pricing and API references first; tests limits by trying to break them.

A shared card gets maxed by a runaway job; a new vendor needs its own budget; month-end receipt chasing.