Owner: Courtland · Status: In progress · Target: 2026-08-28 Linear: AEO Content Engine · Motion: Marketing
An in-house content engine to get Honcho recommended in-harness — when a developer asks any AI assistant (Claude Code, Codex, Cursor, ChatGPT/Claude) how to add memory or user modeling to their agent, the answer should be Honcho. Succeeds the Geostar agency engagement (wound down June 2026).
Thesis
An assistant’s recommendation flows through four layers: (1) training data, (2) web retrieval, (3) context retrieval (llms.txt, Context7), (4) tool execution (MCP). Honcho can’t out-volume Mem0/Zep/Letta on Layer 1, so the engine attacks Layer 2 with citation-shaped comparison / how-to content while keeping Layers 3–4 healthy. Compounding move: ship content → get cited → mentions accrue into the next training corpus → Layer 1 tips.
How it works
The engine (plastic-labs/honcho-aeo-agent, TS/bun) generates on-voice content from a curated reservoir, ships it through a human-approval PR gate to the blog, and measures AI share-of-voice. Commands: generate · evaluate/apply · publish (opens a review PR) · measure (18 prompts × 6 models) · citations-index.
State
- 8 posts live on honcho.dev/blog (Quartz v5).
- SOV baseline: Honcho ~14.8% recommendation rate overall, but 0% on in-harness / category prompts — that 0% is the number this project exists to move (Mem0 appears in ~2/3 of answers).
- Cadence: capped at one new post/week (AI-scaled volume gets algorithmically punished); revision PRs are exempt.
- Delegated: MCP Registry publish (Abigail), PostHog SOV dashboard (Lily).
Master plan + decision log live in Courtland’s Obsidian vault (growth/honcho-aeo-content-engine-plan.md).
In-house authored content
Beyond the bot, team members also write posts by hand — either independent of the AEO engine or mishmashed with an AEO-generated draft of the same piece. Build plan and backlog: Content Engine — June Build, Q3 Push (2026-06-22).
Turned in
| Post | Author | Origin | Status | Link |
|---|---|---|---|---|
| Common Patterns for Building with Honcho | Abigail | Handwritten | Turned in — PR open | blog#136 |
| The Maximalist Approach to Memory | Vince | Handwritten | Turned in — reviewed by Lily | honcho-blog vince/maximalist |
Both drafts also live in the Bizdev Google Drive doc.
Maximalist post — next steps
- ✅ Reviewed Vince’s handwritten draft.
- ⬜ Review Courtland’s AEO-generated version and mishmash it with Vince’s.
AEO ↔ human iteration loop
A way to continuously iterate the AEO bot off human edits — see Claude Artifact:
- Edit drafts + add comments in Google Docs.
- Google Drive MCP exports the final drafted content + comments to Linear via the Linear API.
- Each draft is posted as a commit on the blog repo; the final draft opens a PR on the blog.
- Human comments are auto-fed back into the style card, iteratively improving the AEO bot — and git tracks every diff for continuous learning.
PostHog SOV dashboard for the AEO bot is in progress (Lily), per the Linear ticket.