Not All Tokens Are Created Equal: Cutting the Cost of Agentic Code Search

AI agents spend most of their time reading code, not writing it, and that cost never shows up as its own line on the invoice. We've found the fastest way to bring it down.

Join Cockroach Labs and Takara.ai for a practical session on token efficiency in agentic coding workflows. We'll cover why agent demand is growing far faster than compute supply, where the tokens actually go, and how sharper code retrieval changes the economics. We'll walk through the architecture for holding the vector index in CockroachDB, so search stays fresh, distributed, and inside your own perimeter, then run Miru, Takara's open-source, agent-agnostic code search built on that architecture, live against a large codebase, with benchmark results against a grep baseline.

45 minutes, including a live demo and Q&A. Built for engineering leaders, platform teams, and anyone accountable for an AI spend line.

Agenda:

- Why Tokens Matter

- The Architecture: CockroachDB as the Vector Index

- Introducing Miru

- Demo: Miru in Action

- Getting Started and Q&A