Stochastic reposted this
This is an insightful analysis by Jaya Gupta on why capturing the context behind AI decisions is critical and why current infrastructure like Databricks and Snowflake still falls short. For AI agents in regulated industries, these capabilities are table stakes. Beyond a simple context graph, companies require robust audit trails that capture the specific versions of inputs, policies, and exceptions used, alongside a clear record of any human intervention. At Stochastic, we ship our products fully equipped with context graphs, versioning, and recorded history of every manual intervention. If you'd like to learn more, shoot me a message or leave comments below.
There's a subtle infrastructure problem no one's talking about. Data warehouses like Snowflake and Databricks are in the read path. They receive data via ETL after decisions are made. By the time it lands, the decision context is gone. They can tell you what happened, but they can't tell you why. For example let's say a support lead checks customer ARR in Salesforce, sees two open escalations in Zendesk, reads a Slack thread flagging churn risk, and decides to escalate. The synthesis of that process happens in their head. The ticket just says "escalated to Tier 3." The warehouse gets "escalated to Tier 3" three hours later, but the reasoning doesn't really get recorded. This matters for agents. If you want an agent to make that same escalation decision, it doesn't just need access to the data. It needs access to how that data was synthesized into a decision the last ten times. The precedents. The exceptions. The judgment calls. That requires being in the write path, at commit time and in the execution loop. That's the only place you can capture the full context: what inputs were gathered, what policy applied, what exception was granted, who approved. If you persist those traces, you get something we call a context graph (not embeddings), but more a structured record of how decisions were actually made, stitched across entities and time, so precedent becomes queryable. The startups sitting in the agent orchestration layer are the ones positioned to build this. Not because they're smarter. Because they're in the right place architecturally. Read more about our thesis below and if you are building a company in this space, excited to chat!