Field Notes - Sept 25, '25

Executive Signals

AEO is the new SEO: long-tail answers and video TLDRs capture AI-overview demand
Community is the new outbound: live conversion beats sequences when meetings become media
Docs are the new memory: versioned knowledge repos feed agents and Q&A reliably
Projects, not personas: scoped micro-agents with strict triggers outperform general assistants
Autonomy needs brakes: time-boxed runs and kill-criteria prevent agent thrash

SEO traffic can stay flat while MQLs crater. Answer engines and AI overviews are capturing intent upstream. Treat AEO like SEO: ship extreme long-tail help docs, each paired with a 60–90 second TLDR video (YouTube authority carries). Track ChatGPT referrals as a first-class channel; a Series A reported roughly 20% of that traffic converting to MQL.

If SEO MQLs are down 50% for 90 days, shift 50% capacity to long tail + video
Publish 1–2 TLDRs per day; put the primary CTA in the YouTube description
Instrument ChatGPT referrals as a distinct channel; map to MQL and pipeline

Sales

Community-First, Sell In-Call, and Factory Your Meetings

Outbound email is tapped. Community and LinkedIn work when you do not sound like sales and you convert live. Turn every meeting into media to scale founder time and compress cycles.

Book the demo on the call, not via follow-up
Record all meetings; auto-generate one post and three clips per call via n8n or Zapier
Review weekly: meeting-to-content-to-pipeline conversion and time-to-next-step

Product

Micro-Agents via Project-Level Instructions

Spin up role-specific micro-agents with long-form project instructions and a tight reference set. Give each a trigger phrase and strict behavior. Treat the project as scoped memory for that job, from blog helper to ops playbook to tax research. Turn off model training on your data by default.

Write 6–8k characters of instructions: role, tone, inputs, outputs, triggers
Attach 3–5 key docs per project and keep them current
Disable “use my content to improve models” in settings

Engineering

Turn Knowledge Into a Queryable Repo

Stop treating docs as a website project. Keep a /knowledge folder in your code repo and use CI to embed every Markdown change into a vector store. That single source powers Q&A and gives agents durable memory. Seed it with personal history, wins, decisions, and a lightweight relationship graph from your contacts.

Create /knowledge and ship weekly; one merge request equals one new fact
CI trigger: on /knowledge/**.md changes, embed to your vector database
Add a relationships/ subfolder from your contact export; normalize names and contexts

Run Autonomous Coding Agents Without Overload

Autonomous coding agents perform best in 45–60 minute sprints. Humans are not good at supervising ten at once. Cap concurrent runs, time-box them, and require check-ins. If there is no measurable progress, kill the task or downscope.

Hard-cap concurrency at two to three agents; queue the rest
Enforce 60-minute loops with a three to five bullet status and next step
Define kill criteria upfront, such as no PR diff or unchanged failing test over two loops

Field Notes