Skip to content

Details

Agents in your SDLC are more expensive than they need to be. When agents connect to MCPs, they waste tokens reasoning to find answers, increasing cost.

But what if the context is structured around the questions agents ask the most? Port ran an experiment and measured the cost.

In this session, you’ll learn about the research findings and how to build a context lake that reduces agent cost at scale.

Join us to find out:

  • Why AI agent token costs are structurally high, and why switching models doesn't fix it
  • What the research found after running 1200 queries across four context conditions and three models, totaling 1,200 runs
  • How to build a context lake that works in practice: which properties to pre-join, how to model data around your agents' actual query patterns, and what catalog hygiene looks like once you're running at scale.

Speaker: Matar Peles
Field CTO @ Port
Matar is Field CTO at Port, connecting product and market in the field. He works closely with customers and internal teams to translate real-world needs into product direction, focusing on productivity and developer experience at scale.

Speaker: Aaron Taylor
Program Manager @ Port
Aaron is a program manager at Port where he researches how engineering teams actually use AI, and translates the findings into clear, practical guidance for the people building with it.

You may also like