The AI content machine that turns ideas into posts that don't sound like slop | Alex Lieberman
B · Useful, limited How I AI episode in which Alex Lieberman (Morning Brew co-founder, 10x) demonstrates his firm's skill-based content pipeline live; a detailed, copyable workflow, but results are self-reported and the speaker sells AI transformation services.
Summary based on the full transcript·Original: about 43 minutes to watch
In short
A 43-minute How I AI episode (uploaded 20 July 2026). Alex Lieberman, co-founder of Morning Brew and now of 10x, an applied-AI services firm, shows the "content machine" his company built as a Claude plugin of chained skills. It scans the author's Slack, Notion, meeting notes, email and followed accounts for ideas, interviews the author with interviewer personas, drafts only from the author's own spoken answers in a codified voice, scores the draft with editor personas, and logs every correction as a lesson for next time. His claim on slop: when output is generic, it is because the person gave thin answers in the interview step.
Key numbers
- 15content ideas ('spikes') the pipeline proposes each day, half from internal systems and half from followed accounts
- 9 out of 10aggregate score from six editor personas below which the draft is sent back for revision
- 40%share of inbound leads in one quarter attributed to a staff-posting campaign at Lieberman's earlier business
- $5,000prize budget for 10x's month-long staff posting challenge
As reported by the source.
English·Video·opens youtube.com
What it reports
- Process mapping first: he wrote out his manual content process (idea, research, brain dump, format, draft, edit, publish, repurpose) and decided where AI drives and where it assists.
- "Oracle" step: scans the last seven days of internal and external sources and scores ideas higher for stories, strong views and specific examples; 15 a day. Unused ideas go to a Notion "vault".
- Interview panel: six interviewer personas ask follow-up questions until there is enough material; answers are dictated.
- Each person has a style guide and a voice file built from their best-performing posts. Drafts may use only transcript words apart from the hook and close.
- Editor council of six personas scores 1-10; under 9 triggers a revision loop.
- Lessons loop: after publishing, the system diffs the draft against the final post, proposes general lessons, and adds them to a lessons file if the author approves.
- Staff advocacy: an earlier campaign at another of his businesses drove 40% of inbound leads in a quarter; 10x now runs a points-based posting challenge with $5,000 in prizes, aimed mainly at engineering recruitment.
How they got it
A live demonstration and the speaker's own experience. The lead and hiring figures are his recollection; no measures of post quality or engagement before and after the tool are given.
Limits
- Self-reported results from a firm that sells AI transformation work.
- Persona "councils" named after real interviewers and writers are a prompting device; their scoring is not validated.
- The pipeline reads Slack, email and meeting notes, which raises confidentiality questions for public posts.
- Sponsored podcast with ads in the transcript.
Why it matters for companies
The design choice worth copying is the constraint: the model shapes the person's own words and is not allowed to invent content. That keeps expertise and accountability with the author and is the most reliable way to avoid generic output. The lessons file that grows from each edit is a simple, auditable form of feedback that works for any drafting assistant, from HR communications to policy notes. Firms that want staff to post should pair this with clear rules on what internal material may be mined, because the idea scanner reads internal channels by default.
Summary based on the full transcript. Read the original for the full argument and data.Watch the original on youtube.com
- Tags
- AI enablement, Lecture