Why decision traces are the future of AI agents
As businesses move beyond simple automation toward autonomous agents, a critical gap has emerged between structured data “rules” and the “tribal knowledge” found in human conversations. Context graphs represent a new layer of enterprise intelligence that captures the “why” behind decisions, turning informal precedents into searchable, canonical truths. This shift suggests that the future of work isn’t just about building better models, but about engineering the decision lineage that allows agents to navigate complex, real-world exceptions.
Points clés
- Jamine Ball argues that the primary failure point for AI agents is not the model, but the lack of a “canonical answer” across fragmented systems like Salesforce, Netsuite, and Zendesk.
- Jay Agupta and Ashug from Foundation Capital identify “context graphs” as a trillion-dollar opportunity that fills the gap between rules and actual decision traces.
- Context graphs capture “tribal knowledge”—exceptions, precedents, and overrides—that currently live in Slack threads, DMs, and human heads.
- Systems of record (like CRMs) excel at recording “state,” but are traditionally poor at documenting “decision lineage” or the reasoning behind a choice.
- AI agents are uniquely positioned to build these graphs by documenting every input, policy, and approval used during a specific execution path.
- The “cogent enterprise” perspective suggests that these graphs should not be pre-defined but should emerge organically from agent usage patterns.
- Aaron Levy, CEO of Box, predicts that “context engineering” will be a key enterprise AI trend for 2026.
- Levy suggests that humans will adapt to AI by moving from individual contributors to “managers of agents,” focusing on oversight and escalation.
- Standard data warehouses often act as “retrospective mirrors,” whereas agents require a “transactional front door” to make real-time decisions.
- Differentiation between companies in the AI era will rely on how well they leverage their unique organizational context and judgment history.
À retenir
So, it turns out your company isn’t actually run by that expensive ERP system, but by a series of frantic Slack DMs and your boss nodding in a hallway. While we’ve spent billions on “data lakes” that mostly just function as historical graveyards, the real gold is the “why” behind that 20% discount you gave a client because their procurement office is a nightmare. My advice? Start documenting your “vibes” and “gut feelings” now, because soon your AI agent will need a searchable database of your human irrationality just to get through a Tuesday. If you don’t, you’ll be the person explaining to a robot why it shouldn’t follow the rules—which sounds like a great way to spend your golden years.
Sources
Quiz sur la vidéo: 5 questions
