The Log Is The Agent - Ishaan Sehgal, Omnara

TL;DR

• 1

  Agents are logs, not runtimes: Ishaan Sehgal argues that the persistent, append-only event history of an agent is its true identity, much like a save file defines a video game character.

• 2

  The log enables disposable workers: A worker can read the log, advance the agent one step, write the result, and disappear; any other worker can pick up the session later without coordination.

• 3

  All else is a projection: The model's context, the UI, debugging traces, auditing, and compaction are all projections of the log. The log alone is the durable source.

• 4

  Compaction is lossy and can be forked: Context windows are finite, so logs must be compacted. But compaction is a best-effort lossy fork; the raw log remains the authoritative record.

• 5

  Log lock-in is the deepest lock-in: Models, APIs, and tools are swappable, but if a provider owns your log, they own your agent. This is the most intimate form of vendor lock-in yet.

• 6

  Log-as-agent yields scalability and reliability by default: Failover, multi-model forking, multiplayer sessions, and provider migration become structural properties instead of bolt-on hacks.