Most AI pilots run once, in a controlled setting, with a small group watching the bill closely. Production runs constantly, for everyone, and that's usually where the invoice starts to look haunted. A single request can trigger a dozen lookups behind the scenes before it ever reaches an answer, and most of that cost lives in the parts of the system nobody budgeted for. 

In this session, Chris Carney walks through where AI spend actually goes once something ships, and what to check before finance comes knocking. 
 

What you'll learn:

 Why production costs more than your pilot predicted 
The gap between a pilot's estimate and what a live system actually costs, and where that difference usually hides. 

How to pick the right model for the job 
Not every task needs your most expensive model. A practical way to match the work to the right tier instead of defaulting to the top of the line. 

What actually drives up token costs 
Why a system's output costs more than its input, and the habits that quietly compound that cost with every revision. 

How to use Bedrock's own tools to keep spend in check 
The cost controls built into Bedrock itself, provisioned throughput versus on-demand, batch inference, prompt caching, and how to track spend by team, so cost management isn't something you bolt on after the fact. 

Who should attend 
- IT and engineering leaders supporting an AI tool or agent that's already live, or about to be 
-AWS and cloud cost owners who've gotten a spend question from finance they didn't have a clean answer for 
-Anyone deciding whether to switch models to save money, and not sure how to make that case 
-Technical and non-technical leaders alike, this is a budget and planning conversation as much as a technical one 

Come find out what's really lurking in your AI bill before it shows up at your next budget review.