The problem
Bulldozer’s weekly output covers personal and company LinkedIn posts (one a day), the Bullish newsletter, SEO pages, visuals and tweets. When there is a podcast episode, a webinar or a physical event, it also covers the full sequence that comes with it: emails, reminders and announcements. Before the system, he says the team produced about half of today’s volume and a single week of it took two and a half days. He frames the choice as hiring more people to produce more content or building a system. For a CMO, he says, the question became “what system do they run.”
The blueprint (as he describes it)
- Split the work into small skills. He did not write one giant prompt. He split the content operation “into dozens of small skills and slash commands, each with one precise job”. One writes the newsletter, one turns a transcript into posts, one drafts the signature post for a new client, and one handles SEO briefs. The count is “32 skills, 29 commands” at last count.
- Put one meta command on top. When he runs /week, Claude runs an interview. Is there a newsletter this week? A podcast episode, and if so, the transcript? A new client to announce? A webinar to push? “It asks, we answer, it routes each job to the right skill.” He says this part took the longest to get right. It opens on a Monday, and by lunch a full week of content exists.
- Feed one input into many formats. A podcast becomes a newsletter, the newsletter becomes LinkedIn posts, and a post becomes tweets.
- Fix the order of the cascade. He calls sequencing the real bottleneck. The newsletter sets the week’s thesis, the podcast deepens it, LinkedIn carries it and X sharpens it. He says getting the order wrong turns the whole week into rework. They iterated for months before the cascade settled. The order is now built into the meta command: the newsletter first, then the podcast, then the rest.
- Check against brand and anti-slop rules. Most text gets drafted, checked against brand and anti-slop rules, and queued. Every output passes an anti-slop filter before it ships. He points to Wikipedia’s “Signs of AI writing” page as the public list of what counts as slop.
- Add visuals only at set points. Visuals plug in for the newsletter, LinkedIn infographics and webinar thumbnails. Everywhere else the post is text. Some visuals come from Nano Banana (Gemini) and some are built directly in responsive HTML. His advice is to templatize what works and hand-build the rest.
- Keep the human at both ends. He says the ideas, angles and direction stay his. He wrote more than 1,000 LinkedIn posts and hundreds of newsletters by hand before the system existed, and he says the model learned his voice. “The input is human. The final read is human.” He reads everything before it ships.
- The stack. Claude Code, 32 skills, 29 slash commands, a memory file and about a dozen MCP connections to tools such as HubSpot, Notion, Buffer and Webflow.
Stated results
- Before: about half of today’s output, with a week of content taking two and a half days.
- Now: the same week takes half a day, with twice the volume shipped.
- He combines the two into “a tenfold jump in throughput”: “twice the output, in a fifth of the time”.
- He limits that claim himself. The 10x “is on production, not on judgment.”
- Their content manager moved into field marketing.
- No engagement, traffic, lead or revenue figures are stated.
Caveats
- Bulldozer is an agency that sells marketing services. Per the newsletter about page, it runs marketing and growth for 500+ B2B companies (Outbound, Paid, SEO, ABM, RevOps). The Bullish newsletter promotes the agency.
- The post opens a series on how Bulldozer uses AI for revenue. The method for splitting work into sub-tasks is covered in an earlier edition, not in this post.
- The time and volume figures are his own estimates. The post does not say how they were measured. The 10x figure is his arithmetic on those two estimates.
- The post names the skills’ jobs but does not include the skill or command files themselves.
- He calls visuals “the weak link.” Outside templated assets such as webinar thumbnails, he says the visual layer is still shaky.
- Handles: LinkedIn linkedin.com/in/jordanchenevier (verified: linked from his own newsletter about page). Substack @bullishbulldozer (verified: author of the post). X: a profile appears only on a third-party aggregator, so it is not verified and not tagged.
FAQ
Does the system publish content without a human reading it?
No. He says a human reads everything before it ships, and that human is him. He also says the ideas, angles and direction stay his.
What does the /week command do?
It interviews him about the week: whether there is a newsletter, a podcast episode and its transcript, a new client to announce or a webinar to push. It then routes each job to the right skill.
How much time does it save?
He says a week of content went from two and a half days to half a day, with twice the volume. He adds that the gain is on production, not on judgment.
Does it make images too?
Only at specific points: the newsletter, LinkedIn infographics and webinar thumbnails. He uses Nano Banana (Gemini) and responsive HTML, and he says visuals are still the weak link.
Sources
- https://news.bulldozer-collective.com/p/inside-the-ai-system-that-runs-our (primary: first-person post on his newsletter Bullish, published Jun 02, 2026, signed Jordan)
- Bullish newsletter about page on news.bulldozer-collective.com (used for his role and Bulldozer’s services)
Want agents like these working your distribution? Get It Built. Get new teardowns weekly: subscribe (double opt-in).
Get one teardown a week.
Same format, a new operator every issue. Double opt-in.