Skip to content

Give agents context they can trust

Engineering field guide · Company context · Both days

How do you stop useful context from becoming stale context?

Day 1 tackles changing schemas and expert knowledge. Day 2 adds dated company facts and a shared gateway for tools, memory, and monitoring.

All topic guides · Practical takeaways

The talks treat context as something to build and maintain. A large document collection helps only when the agent receives the right meaning, permissions, and evidence.

Session Approach
Southbridge · Day 1 Adapts schemas to the question and uses different model roles within a data pipeline.
Citadel Securities · Day 2 Tracks dated beliefs about the organization, reconciles sources, and preserves history as facts change.
Intuit Credit Karma · Day 2 Centralizes tool access, monitoring, memory, and context selection through shared infrastructure.

Southbridge · Day 1 · Citadel Securities · Day 2 · Intuit Credit Karma · Day 2

Compression can remove the evidence needed to notice a contradiction. A quiet data source is missing signal, not confirmation. The context optimizer’s reported savings describe its demonstrations and workloads.

  1. Choose one status question and collect dated evidence from two sources.
  2. Keep source identifiers and timestamps while creating a shorter context for the agent.
  3. Test an outdated claim, conflicting updates, and a source that stops reporting.

Compare correct answers, unsupported certainty, stale facts used, tokens consumed, and whether the agent can point back to the deciding evidence.

The comparison and experiment are our editorial synthesis. They are not a reported benchmark or the speakers’ prescribed procedure.