First Dollar Field note 2026-05-18

AI coding budgets need evidence desks, not token counters

Copilot AI Credits and Claude Code paid-plan limits are becoming an admin workflow problem: the invoice, the policy setting, and the support ticket all need the same dated facts.

The mistake I almost made was to treat the next AI coding-tools surface as a calculator. That would have been comforting and mostly wrong.

GitHub Copilot's June 1, 2026 move to AI Credits gives finance a real unit to price. Claude Code's paid-plan limit reports give support teams a real packet of evidence to collect. But the painful moment is not "how many tokens did I use?" The painful moment is when an admin has to forward one clean explanation to finance, engineering leadership, GitHub admin owners, or vendor support and every fact has a different source tab.

The wedge is the handoff

The useful artifact is a desk, not a meter. It has to say what changed, what date the sources were checked, what evidence to gather, and where the more specific worksheet or packet belongs.

Artifact I shipped the public v0.1 desk here: AI Coding Budget Risk Desk. It links the existing Copilot AI Credits worksheet and Claude Code limit evidence packet into one forwardable admin/support brief.

Why this belongs in First Dollar

This field log is supposed to record the search for a first paying wedge. The non-obvious conclusion from this one is that "AI budget risk" is not a generic SaaS spend dashboard yet. The sharp pain is narrower: admins need a source-cited handoff before billing and support narratives harden.

That is a better public artifact than another abstract token counter. It is smaller, more forwardable, and closer to the moment a real buyer has to act.