How a data scientist at Kilo built the business context layer his AI agents were missing

Pedro Heyerdahl, formerly Kilo Code’s sole data scientist and now at Anaconda, argued that although AI agents have become more capable of solving technical problems, they still lack the organizational knowledge needed to make sound business decisions. He found that dbt tools, semantic models, lineage, and documented projects can help agents understand data structures and metrics, but cannot capture the decisions, debates, failed experiments, and changing assumptions stored across sources such as Slack, meetings, and pull requests. After finding that large memory files, raw retrieval systems, and direct access to workplace tools often produced stale or conflicting results, he developed a context pipeline that evolved from summarized articles to atomic facts, decisions, entities, and prescriptions. Its latest version records temporal validity and superseded information, allowing historical context to remain available without contaminating current results. The system uses different AI models according to task complexity, checks incoming information against existing knowledge, routes conflicts for human review, and lets agents submit new findings for approval, creating a governed and continuously updated business-context layer.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.