OMX Helsinki — S&P 500 — DAX — NASDAQ 100 — STOXX 600 — EUR/USD — EUR/SEK — BTC/USD — ETH/USD — Euribor 3M — Euribor 12M —
AI Engineer

AI Engineer

Technology and AI 1,040,000 subscribers

The Data Context Layer for Agents — Yoni Michael & Brandon Callender, Typedef

11.10.2026 3,579 views
Published: 11.10.2026 Category: Technology and AI Views: 3,579

Episode details

Episode description

Two Rust types are both called PostingList, but changing them breaks different code. A text search finds both; a context graph preserves their identities and exposes their separate dependencies. Yoni Michael and Brandon Callender use that distinction to connect data engineering and coding agents. For a data system, valid SQL can still produce the wrong answer when a join multiplies rows or a refactor changes a metric's meaning. Michael describes a context layer that computes grain, lineage and relationships from transformation code, checks join assumptions against real data, and makes the resulting evidence queryable before an agent acts. Callender builds a smaller version over selected crates in Qdrant. Tree sitter extracts syntax, Rustdoc resolves symbols, and Neo4j stores the relationships. An agent can then ask which symbols depend on a particular type instead of repeatedly searching files and inferring connections. The workshop's central complication comes next: given better graph tools, the agent may still choose grep. Tool naming, familiar output formats and prompting become part of the interface, and evaluations must measure whether the tools are actually used. Cached comparisons separate getting an answer right once from getting it right across repeated runs, while trace inspection catches flaws hidden by a favorable score. The closing discussion explores incremental indexing, Python's harder resolution problem, small auditable semantic operations with Fenic, and the work required to extend a classroom example into a maintained context layer. Speaker info: - https://www.linkedin.com/in/yonimichael - https://x.com/yoni_michael - https://www.linkedin.com/in/bcallender/ - https://github.com/bcallender/agent-context-workshop Timestamps: 0:00 - The missing context for data agents 3:44 - Impact analysis, joins, and safe changes 5:05 - Distributed context and table grain 6:29 - Valid SQL can still explain the wrong cause 9:02 - Four facts an agent needs 10:13 - Join semantics and fan out 12:42 - Deterministic extraction and data profiling 14:16 - Persist a queryable context graph 17:33 - Analyze downstream changes before acting 19:07 - Safe joins and metric definitions 20:55 - Evaluate data engineering tasks 22:32 - Five context engineering…

Open the AI assistant chat. The chat loads only when you open it.