The Log is the Agent

arxiv.org
July 5, 2026
2 min read
51/100
Summary
ActiveGraph is a runtime that inverts the traditional agent framework by prioritizing event-sourced reactive graphs, where the logging layer is central to the system. This design enables auditable and forkable agentic systems, improving observability and state management.
Key Takeaways
- ActiveGraph inverts traditional agent framework design by making the append-only event log the source of truth, with the working graph as a deterministic projection of that log.
- The architecture allows for deterministic replay of any run from its log, cheap forking of runs at any event, and end-to-end lineage tracking from high-level goals to individual model calls.
- The system is proposed to be well-suited for self-improving agents and extends previous research on graph-memory systems.
Coverage & discussions
Hacker News · arxiv.org
Community discussions
Historical count: 5 · 5 saved in sample · Historical sample
Analysis ready
Community Sentiment
Positives
- Very cool work!! This is the same pattern we used at $MY_STARTUP to develop $MY_HARNESS, which persists the entire graph to disk — a game changer for agent harnesses.
- All relevant events that affect the context window are stored in an event log, making it easier to manage sessions and fork agents.
- Excited to check this out! The implementation looks promising and could enhance agentic systems.
Concerns
- Can someone explain why such a trivial knowhow is paper-worthy? Event sourcing is well known and feels like a stretch for a formal publication.
- There are a lot of 'logs for agents' papers that seem redundant and lack innovation, making it hard to take this seriously.
Related Articles

Agentic Context Management: Memory and Cost as Architecture Problems
Aug 26, 2026

Coding agents think ahead of time
Jul 14, 2026

Evaluating AGENTS.md: are they helpful for coding agents?
Feb 16, 2026

A Study of Microsoft's Early 2026 Rollout of Claude Code and GitHub Copilot CLI
Jul 13, 2026

Zero-Mem: Zero-Token Memory Operations for LLM Agents
Aug 5, 2026