World's Fair 2026
The Log Is The Agent - Ishaan Sehgal, Omnara
Overview
This talk argues that the core identity and state of an AI agent reside not in its model or execution environment, but in its append-only log of events. This log, containing every input, output, tool call, and state transition, serves as the agent's durable history, enabling reliable resumption and state reconstruction even after system failures.
Who should watch
- AI Engineers
- Product Managers
- Builders of AI agents
- Those concerned with agent reliability and state management
- Developers looking to understand agent ownership and lock-in
Key takeaways
- An agent's identity is best represented by its event log, analogous to a video game save file, rather than the underlying hardware or software.
- The log acts as the primary source of truth, with models, tools, and UIs all deriving their context and functionality from projections of this history.
- Treating the log as the agent naturally leads to improved reliability, as any worker can resume an agent's state by reading its log.
- Scalability is enhanced because agents are no longer tied to specific machines; any worker can pick up a session by reconstructing its state from the log.
- The log facilitates multi-user collaboration and easier migration between different models or platforms, as it decouples the agent's identity from specific implementations.
- Compaction of the log is a lossy projection, not a replacement for the full history, and should be treated as a way to create new, smaller representations.
- External state changes made by tools are outside the log's scope, but the log still accurately records the agent's actions and its view of the world.
- Log lock-in is the deepest form of vendor lock-in, making it crucial for users to own and control their agent logs.
Notable quotes
*The agent is its data. It's specifically the log.*
*The log should be the primary durable session.*
*The deepest form of lock-in is actually log lock-in.*
Unofficial community note. Prefer the recording for nuance.