Notion's shared-brain for agent fleets breaks under concurrency; a Coda-based fix
bfrench · x · 2026-11-07
Based on running his own Grok Bot fleet, the author dissects where Notion's new official "shared brain for AI agents" guide breaks down with concurrent bots, and offers a working alternative.
Why Notion breaks: its agent-memory prompt relies on polite prose rules; Notion formulas are read-only, silently truncate relation lookups past 25 items, can't press buttons via API, and every human reviewing bot writes needs a $10-20/month seat. Past four or five bots, chat-history memory stops scaling.
The fix, SuperLedger: a seven-table Coda (Superhuman Docs) state machine plus a dedicated Grok Bot Steward, wired via Grok Bot's built-in Coda connector (one-click OAuth on Superhuman's Docs MCP — no custom MCP servers or extra API keys). Client bots never edit canonical tables; they append idempotent Proposals, native Coda formulas (schemaok, conflictcheck) gate every write for zero model tokens, and the Steward merges clean rows into hash-chained Commits. Bots read a compiled 585-token local state packet in 28 ms.
Related event: Coda Beats Notion for Shared Memory Across Multiple Bots, Practitioner Says(2 posts)→
More from coding & agent
- Scope benchmark: only 16.6% of 1,035 AI agent attempts can sign up for hosted databases — RaphaelDabadie · 2026-11-07
- Every cloned its Slack with 51 AI coworkers trained on chat history via DSPy GEPA — lateinteraction · 2026-11-07
- User taunts AI coding agent about an edge case — it had already handled it at line 803 — alexanderchen · 2026-11-07
- GitHub Copilot adds one-time sign-in across Figma, Sentry, Slack, synced to every surface — 0xkarasy · 2026-11-07
- 'No moat, it's all a .md file': recreating a lens perspective illusion with Claude Code — itsOmSarraf_ · 2026-11-07
- Open-source tool shows system prompt and tools can eat 98% of API tokens — behrouzbk · 2026-11-07