# 27 Winners: How Mitchell Keller runs a quote-post content system on X with agents

Mitchell Keller posts 10 to 15 tweets a day or more on X with a voice-to-text tool, and he leans on quote posts instead of replies. In the period he shows, 82 of his posts are quote posts. His agents are trained on his own tweets and an opinions.md file. They feed on his Fireflies call transcripts, GitHub activity, session logs and X bookmarks. Each morning an agent sends him a brief with scoring and post ideas, and he decides what to use. A Trigger.dev workflow handles routine retweets and auto-plugs, and he says he never uses AI to post replies.

| | |
|---|---|
| **Who** | Mitchell Keller, who runs the agency LeadGrow and has a Legion website. X: [@MitchellKeller_](https://x.com/MitchellKeller_) (verified 2026-11-04: linked from his YouTube channel; X bio links back to his YouTube channel) |
| **Problem** | Posting founder-led content at volume without much daily time and without posting "slop" to his feed |
| **Stack** | X, LinkedIn, YouTube, Grok, Claude, Codex, RapidAPI (X endpoint), Fireflies, GitHub, Trigger.dev, a voice-to-text tool, an opinions.md file |
| **Confidence** | Medium (first-person video; auto-generated transcript; figures self-reported) |
| **Last verified** | 2026-11-04 |

## The problem
Keller started posting founder-led content "not even maybe six weeks ago." He wanted it to be quick. He says "it's not taking me too much time each day" because he focuses on a few specific things. He also wanted to use AI carefully "to avoid posting any sort of slop to my feed."

## The blueprint (as he describes it)
1. **Three rules for where to post.** X is for posting off the cuff to find ideas. LinkedIn and YouTube are only for curated content. Everything you post should come from the work you do.
2. **Quote post instead of reply.** He says quote retweets give "borrowed authority and borrowed exposure at the same time." He quotes people in an adjacent domain and gives something more valuable, expands on the point or offers a contrarian view. His summary: "The new meta on X is actually like it used to be, you know, be a reply guy. No, it's be a quote poster."
3. **Post at volume, by voice.** On X he posts "10, 15 tweets a day or more" and uses a voice-to-text tool. His reasoning: one post a day may give you only one winner.
4. **Train the agent on his own tweets only.** His voice is built "only from my tweets," not his essays, curated LinkedIn posts or client work. He keeps an opinions.md file and credits the idea to an engineer whose name is transcribed as "Kun Shen" ([UNVERIFIED] spelling). He reviews it weekly and says to "only allow things that are true to go into your opinions." He also tracks contradictions and changes of opinion as tweet ideas.
5. **Feed it the work.** An agent ingests his Fireflies call transcripts and his GitHub actions, PRs, issues and merges. It compares them against his session logs to find interesting moments. He also feeds it his X bookmarks every morning. One tweet from this system described his AI using a "Decisions MD" file to set itself a budget and find 1,880 more qualified contacts by scraping over 40 data sources, then asking him to approve adding them to a campaign.
6. **Pull X posts on a schedule.** He has his Grok-based agent pull from the Twitter API on a schedule through RapidAPI, which he prices at "1 ,000 requests for $1, 100 ,000 for $25." He scores posts himself and against X algorithm updates (source named in the transcript as "Peter Mick," [UNVERIFIED] spelling). The agent also monitors friends' posts on a cron schedule and delivers them in the morning.
7. **Read the morning brief, stay in charge.** Each morning the agent compares his top 20% of posts against the previous day and suggests ideas he can use or skip. He says, "The important thing is to be the human in that loop."
8. **Turn winners into LinkedIn posts weekly.** At the end of each week his X winners become LinkedIn posts, each with a generated graphic, scheduled automatically. YouTube content goes back into the system through a weekly article ingestion.
9. **Automate the routine parts on Trigger.dev.** He has Grok build the workflow, then moves it to Trigger.dev. It auto-plugs a tweet only if it is in the top 50% of tweets that week. It auto-retweets a non-winner twice and a winner three times. Articles he posts on X go into his quote retweet system automatically. The plug is a call to action for a free resource or his YouTube channel.
10. **Keep replies human.** "Never, never use AI to post replies." For other agent actions, he says, "you just give approvals."

## Stated results
- About two million impressions on X since starting, in "not even maybe six weeks."
- A monetized X profile.
- 82 quote posts and 55 originals in the period shown. "Top 20 % and 27 winners all have 61 plus engagements."
- Closed one deal from X and was referred to meet the head of talent of a16z.
- "Hundreds of people are coming from X" to his website or YouTube. He reports around 200 visits a day to his Legion website.
- He says the system has gotten him "twice the number of views that I was getting on YouTube."
- Output: "15 to 30 pieces of content a day, one to two YouTube videos a week, sometimes three."

## Caveats
- All figures are self-reported in the video. The source text shows no dashboard or analytics export.
- "Not even maybe six weeks" is approximate. The third-party transcript page's URL reads "did-2-2m-views-in-7-weeks," which differs from what he says in the video.
- The 27 winners figure is ambiguous in the transcript. It follows "55 of which are originals over the last couple, three weeks," so the pool of posts it was measured against is unclear.
- The transcript is auto-generated. Names may be misspelled: "Kun Shen" (also written "Kanchen"), "Peter Mick" and others are as transcribed.
- The video description links to his free GTM tools page on leadgrow.ai, which is lead capture for his agency LeadGrow. He also says he is "about to launch a community and a course."
- The RapidAPI pricing and the claim that it is "the equivalent of $500 from the real X API" are his statements, not checked.
- Handles: X @MitchellKeller_ and YouTube @MitchellKeller_ (verified 2026-11-04: X linked from his YouTube channel, X bio links back). LinkedIn linkedin.com/in/mitchell-keller (verified 2026-11-04: profile lists LeadGrow.ai). Bluesky, Threads, Instagram, TikTok, Mastodon: none verified.

## FAQ

### Does Mitchell Keller use AI to write his replies on X?
No. He says, "Never, never use AI to post replies." The agents handle inputs, a morning brief with ideas, LinkedIn repurposing and routine retweets and plugs.

### What does his agent learn his voice from?
It learns only from his own tweets, which he calls his most raw and conversational writing. It also uses an opinions.md file that he reviews weekly, and he allows only things that are true into it.

### When does his system auto-plug or retweet a post?
His Trigger.dev workflow auto-plugs a tweet only if it is in the top 50% of tweets that week. It auto-retweets a non-winner twice and a winner three times.

### Did the content system lead to business?
He says he closed one deal from X and was referred to meet the head of talent of a16z. He reports around 200 visits a day to his Legion website, but the source does not tie revenue figures to the system.

## Sources
- https://www.youtube.com/watch?v=-u2prz7WCNI (primary: his own YouTube video and description, published 2026-10-09)
- moderncreator.app/2026-10-09-mitchell-keller-this-founder-content-system-did-2-2m-views-in-7-weeks (auto-generated transcript of the same video)

---
*Want agents like these working your distribution? [Get It Built](/get-it-built). Get new teardowns weekly: [subscribe](/subscribe) (double opt-in).*