Sign InOpen Brain
AI EngineerVideoSource Linked

Total Recall: Agent Memory and Harness Engineering — Ignacio Martinez, Oracle

Treat memory, context control, concurrency, and model routing as harness concerns. The workshop shows how these layers can improve agent behavior without changing model weights.

AI Engineer · Sep 18, 2026
Open Source Open MarkdownOpen JSON
Source Summary

The workshop separates the model’s fixed reasoning core from the harness around it. It covers **agent memory**, combining files with databases, using worktrees when many agents edit concurrently, and routing tasks between models.

Practical Implication

Builders should keep working context deliberately small, persist reusable knowledge outside the conversation, and turn effective sessions into repeatable workflows. For parallel coding agents, isolate file changes rather than expecting shared files to provide transactional consistency.

Agent-Ready Context
The workshop separates the model’s fixed reasoning core from the harness around it. It covers **agent memory**, combining files with databases, using worktrees when many agents edit concurrently, and routing tasks between models.

Builders should keep working context deliberately small, persist reusable knowledge outside the conversation, and turn effective sessions into repeatable workflows. For parallel coding agents, isolate file changes rather than expecting shared files to provide transactional consistency.

The session presents these as practical design patterns, not measured performance results. Claims about context degradation and a future **mixture of small experts** are architectural guidance; the material provides no benchmark against simpler harnesses.
Connected Context · Feed7 Judgment

This consolidates small-context operation, external memory, model routing, reusable workflows, and isolated worktrees into one practical harness boundary. The candidates both sharpen and qualify that guidance: fresh-context loops and existing plugins show concrete implementations, while comparative evidence and model-upgrade experience warn that summarization, planning, recovery, and other scaffolding must be selected per model and budget rather than treated as universally beneficial.

Chained Recursive Language Models for Multi-Iteration ReasoningChained RLM supplies a concrete implementation of the workshop’s small-context pattern by resetting context while carrying summaries, a blackboard, and artifacts between iterations.Evolution of agentic surfaces — Gagan Bhat & Isabella Kai He, AnthropicAnthropic’s model-upgrade experience narrows the workshop’s architectural guidance: external durable state remains useful, but context-management scaffolding can become harmful when model behavior changes.An Empirical Study of Harness Design for Coding AgentsThe empirical harness study adds comparative evidence absent from the workshop, showing that elision, summarization, planning, and tool complexity pay off differently by model and budget.rlaope/oh-my-hermesOh My Hermes packages the same routing, parallel worktree, verification, and reviewed-memory patterns into an existing plugin, while its audit surface reinforces selective testing rather than wholesale adoption.
Context Map
agentcoding#agent-memory#harness-engineering#context-engineering
Uncertainty
The session presents these as practical design patterns, not measured performance results. Claims about context degradation and a future **mixture of small experts** are architectural guidance; the material provides no benchmark against simpler harnesses.