Who Spent All the Tokens? Per-Team AI Cost Control as Code
One agent task can fire off hundreds of model calls. When several teams share one Amazon Bedrock account and nothing is set up to split usage, the bill shows up as one line item, and it's hard to tell which team burned the tokens.
In this live workshop, we'll build per-team AI cost tracking on Amazon Bedrock as code with Pulumi. Each team gets its own inference profile and a role scoped to it, so usage can be measured team by team. We'll turn Bedrock's logs into per-team token counts and use Pulumi Policies to catch mistakes before they deploy.
The code is open source, so you can take it home and run it in your own AWS account.
Join us to learn:
How to give each team its own Amazon Bedrock inference profile with access scoped to it.
How to measure token usage per team from Bedrock's logs.
How Pulumi Policies catch cost and access mistakes before they deploy.
Presenters: Adam Gordon Bell (Community Engineer, Pulumi) and Engin Diri (Principal Solutions Architect, Pulumi)