Topic
Controlling and accounting for enterprise AI spend, including token consumption, model routing for cost, and centralized budget ownership.
GitLab 19.4 adds a /goal command to Duo CLI that loops until a second model judges the objective met or an iteration limit hits. The release adds finer AI spend controls and keeps agent actions inside existing project permissions.
Reality
Cohere's North 2 agent platform lets administrators track token use down to each user and agent and cap spending across the whole company. Cohere is selling to whoever answers for agent cost and risk, and the launch report backs that with three customer accounts and no prices.
Perspective Coverage
5 publishersReality
A licensing change at ServiceNow moves the AI line into a metered pool held at the tenant, where it used to be an add-on a customer could decline. The per-action values a buyer would need to model the bill are unpublished.
Reality
Enterprise AI pricing has separated the seat from the usage, which moves the cost driver from headcount to demand. The overruns now on the record suggest buyers tend to find out months after the fact.
Reality
The AI functions compose like any other SQL, which is the appeal and also the hazard. The statement that made five model calls behind a LIMIT makes one per row without it. Two meters run the whole time.
Reality
The loop reads Bedrock invocation logs through Athena, prices them at published rates, and denies Claude Opus at 80 percent of an engineer's daily budget. It transfers only if every model call carries a federated user identity.
Reality
Metered spend broke the per-seat budget at Uber, and a dollar ceiling only rations the invoice. Commonwealth Bank shows the version of the same argument that survives a CFO's questions.
Reality
Branden Jenkins found the bill on his phone at dinner. His staff hit approval limits before they can spend; his own token wallet just quietly refilled itself.
Reality
An undisclosed pay-to-continue prompt appears at the moment of lockout, and the weekly cap it lifts is never shown to users as a number they could plan around.
Reality
The company's CTO puts agent cost in terminal noise rather than model pricing. The 30 to 80 percent savings are vendor-reported; the 847-to-9 measurement is the part worth keeping.
Reality
A prompt meant for 10 rows ran on 28,000. That is the buying question now: not price per seat, but what one mis-scoped action costs before anyone notices.
Reality
Work-health measures peak at task-level use, at 75% healthy utilization, then fall about five points once AI is embedded in the workflow. Only 2% of tracked employees got that far.
Reality
The firm is standing up an AI Value Realization Office that takes AI money out of departmental budgets, on the argument that functional funding is what strands most of the value.
Reality