Alex Lieberman co-founded Morning Brew while in college, scaled it into one of the most-read business newsletters in the world, and sold it to Business Insider. He's now founder and CEO of Tenex, and he's spent serious time rethinking content creation from the ground up — not by prompting ChatGPT for a draft and calling it done, but by engineering a structured AI workflow in Claude that produces output that actually sounds like him.

The system, which he calls the Content Machine, was the subject of a deep-dive episode on Lenny's Newsletter podcast hosted by Claire Vo.

The Real Problem: Blank Pages and AI Slop

Lieberman's central thesis is that the blank page is the single biggest friction point in content creation. Most people default to asking an AI to "write a LinkedIn post about X," get generic output that reads like the statistical average of the internet, and either publish slop or give up.

His fix isn't a better prompt — it's a fundamentally different architecture.

"AI slop doesn't come from the drafting step. It comes from the interview step — or the lack of one."

Before any drafting happens, his system extracts his actual ideas through a structured interview format. That's the key unlock.

The Six-Step Content Machine

The system runs entirely inside Claude and involves six components:

  1. The Oracle — Scans internal systems (Notion, Linear, Slack) and the web for content spikes: topics gaining traction that align with Lieberman's areas of expertise.
  2. The Interview Panel — An AI-driven interview that pulls out his real perspective on a topic before any writing happens. This is where genuine thinking gets captured.
  3. Voice and Style Files — Markdown files that codify his tone, sentence structure, vocabulary preferences, and rhetorical patterns. The AI drafts inside these constraints.
  4. The Writer's Draft — Only after the interview and voice files are in place does the system produce an actual draft.
  5. The Editorial Council — A panel of AI "editors" that score the draft and suggest revisions, checking for clarity, originality, and voice consistency.
  6. The Lessons Loop — The system learns from Lieberman's feedback over time, continuously refining its understanding of what he approves versus rejects.

The voice file approach is particularly worth noting for founders building personal brands. Rather than fighting AI's tendency toward generic corporate register, Lieberman encodes specificity — the exact phrases he uses, the rhythm of his sentences, even his preferred analogies — into a structured document that acts as a style guardrail.

Distribution as a Durable Moat

Beyond the technical workflow, Lieberman makes a strategic argument relevant to any founder: distribution is becoming the most defensible competitive advantage in an era when content itself is increasingly commoditized.

His argument is that when AI lowers the cost of content production to near zero, the scarce resource becomes trusted audiences — people who want to hear from you specifically, not from a generically competent AI persona.

This is why he's also invested in employee advocacy as a content channel. At Tenex, he launched the Creator Cup — an internal competition that awards a $5,000 prize pool to employees who build the most effective public content. The logic: your team is your most underleveraged marketing channel, and structured incentives can unlock it.

What Founders Can Apply Now

For startup founders and marketers, the practical takeaways are actionable even at small scale:

  • Map your current workflow before adding AI. Lieberman explicitly recommends documenting how you create content today before automating any step.
  • Build a voice file. A Markdown document capturing your tone, vocabulary, and structure will dramatically improve AI draft quality.
  • Use interview-style prompting, not drafting prompts. Ask the AI to interview you about the topic first.
  • Treat employees as distribution assets. Incentivize them to publish, even in small ways.

The Content Machine as a whole may require Claude and some setup time, but the underlying principles — capture genuine thinking first, encode your voice explicitly, and create feedback loops — apply regardless of which tools you're using.