Dissecting 9 Company Brains: All AI Agents Just Do These 4 Steps
garrytan · x · 2026-08-30
An analysis of 9 different "company brain" implementations reveals they all essentially perform the same 4 operations: signal acquisition, memory storage, pruning/organization, and retrieval/generation. The key takeaway is that simply hoarding data is useless; the critical component is a "dreaming & pruning" mechanism to review and tag stale knowledge overnight.
Notable Implementations & Differences:
- GBrain: Pipes emails/calendar to Git, relinks nightly to detect old info.
- mem0: Saves only explicit facts, prioritizes new facts during retrieval.
- Letta: Agents judge retention value; background agents handle cleanup.
- Zep / Graphiti: Uses temporal graphs; appends end dates to facts rather than overwriting.
- Gorgias Cortex: 12k Markdown nodes on GitHub; incorrect answers auto-generate PRs nightly.
Industry Patterns: Markdown + Git is the dominant stack; most workflows require human approval before updating knowledge.
More from coding & agent
- AutoNodo runs 28 days processing 14B tokens, exploring massive context engineering — nodo48 · 2026-08-30
- Favorite AI slop: coding agents adding backward compat to an undeployed app — JFPuget · 2026-08-30
- Why Agents Editing Own Tools is Bounded Improvement, Not Recursion — dl_weekly · 2026-08-30
- EXL3 Quantization Test: Running 30B Model on 12GB VRAM Smoothly — PyaesoneP · 2026-08-30
- LLM Code Fixing Gauntlet: Qwen3.6 Wins, Retry Mechanism Key — sysadmin420 · 2026-08-30
- "We created a digital God and forced Him to use Salesforce MCP" — matt_slotnick · 2026-08-30