The evolution of AI agents and context graphs
This discussion explores how AI agents require a “context graph” to serve as a decentralized system of record that captures cross-functional decision-making. The participants argue that the next generation of software will shift from static human-centric databases to dynamic agent-centric systems that preserve the “why” behind business actions. Strategic dominance will likely belong to companies that own the interface and the implicit data flywheel created by AI-driven workflows.
Points clés
- Jamin Ball, Ashu Garg, and Animesh Koratana discuss the transition from traditional systems of record to agentic AI systems.
- Current AI agents often fail in production because they lack a unified “source of truth” and create disorganized data state changes.
- Amadeus, a $30 billion company, is cited as a historical GDS system of record that maintained value while newer OTA interfaces like Booking.com captured 10x more market cap.
- Traditional systems of record (Salesforce, ServiceNow) are built for human manual entry, which creates data silos and low-quality decision traces.
- “Decision traces” are defined as the underlying reasoning for actions that cross multiple organizational departments, such as support, engineering, and finance.
- Animesh highlights Player Zero’s approach to using support engineering as a wedge to build a context graph across Jira, Zendesk, and GitHub.
- The shift toward agents is expected to compress sales cycles and drive ROI by moving from “intent” to “information delivered” in minutes.
- Predictions for 2026 suggest a massive enterprise adoption of autonomous agents, potentially leading to the first major AI application IPO.
- Data infrastructure giants like Snowflake and Databricks are positioned to challenge traditional SaaS by becoming the “transactional backdoor” for AI apps.
- The concept of “world models” for enterprise logic is identified as a significant upcoming driver for AI advancement over simple continual learning.
À retenir
If you still think your job is manually updating Salesforce fields, I have some news for you—and your replacement is already reading the documentation. We are moving from “software for people who hate data entry” to “systems for agents who don’t need us to click Send.” My advice? Either build a context graph that captures your company’s collective brain or get comfortable being the “human in the loop” whose only job is to provide the occasional, increasingly redundant, thumbs up before the AI takes over the world. But hey, at least we won’t have to argue with the travel agent anymore, right?
Sources
Quiz sur la vidéo: 5 questions
