News peg + a category-killer line
The top three posts (1.88M, 1.35M, 1.32M) all rode a platform launch or a labour reframe.
I run go-to-market with agents instead of a team.
Competitor intelligence, content, SEO and outbound for an AI startup, as one operating machine of Claude Code skills, subagents and MCP servers. The model does judgment. The code does counting. Nothing invents a number.
Scroll↓
01
The machine
Inputs come in on the left. A shared context bus feeds every part. Four layers of agents do the work. Artifacts come out on the right. Click any part and it runs: the pulse travels the wires and the drawer shows the result you would actually get.
context bus
input
output
idle · generated from the repo on every build
02
A week in the operation
A system map proves scope. Cadence proves it runs. Five days, five artifacts.
03
Receipts
Platform numbers for the product this machine runs growth for. The machine is the work behind them, operated by one person. Per-customer figures are deliberately absent, per the rule the machine enforces on itself.
- A competitor radar that rates every competitor post, every day, and separates engagement from buyer relevance.
- A 1,300-keyword SEO universe collapsed into 18 clusters, a 29-row plan, a per-keyword impact model and a citation-priority list for AI Overviews.
- A measured X playbook from a full-population teardown that changed how every post is written.
- A weekly ad-format series that finds, deconstructs, rebuilds and renders the post video without an editor.
- A lead pipeline with an intent signal on every row that has never emailed a competitor.
04
What 11 posts taught us
A full-population teardown of a founder account, two months of posts, ranked three ways. It became a skill.
Replies are engineered,not earned.
A “Comment [WORD]” gate
Remove the gate and replies collapse to single digits, even at a million views.
A named, portable move
Bookmarks cluster on posts that teach something reusable: 147, 93, 1,581.
Craft does not buy reach
A sincere founder essay did 914 views against a ~465K median. No outcome, no visual, no CTA.
05