How I AI: Alex Lieberman's 6-Step Claude Workflow to Beat AI Slop

To beat AI slop and the blank page, In this episode of How I AI, Morning Brew co-founder Alex Lieberman built a six-step content machine. In this episode of How I AI, he runs it live, showing how the system finds promising ideas, interviews him for source material, drafts in his voice, reviews its own work, learns from his edits, and repackages the finished piece for other channels.

Alex built the system to solve two problems at once. First, he can only devote about a quarter of his own day to creating content. Second, his employees at Tenex need a practical way to publish their expertise while still doing their full-time jobs. His answer is a shared workflow that eliminates the blank page without outsourcing the point of view.

The central idea is that generic output starts with generic input. Alex believes most AI slop is a human problem, not a machine one. Instead of asking a model to invent a post from a vague prompt, his system first locates a promising topic, then interviews him until it has specific stories, examples, and opinions to work with.

He demonstrates the entire six-step content workflow, then explains the employee program he built to help the wider Tenex team use it. See Launch an AI-Powered Employee Advocacy Program to Boost Brand Reach.

The six-step AI content machine

Before building anything, Alex mapped the content process he already knew from years of experience: finding an idea, researching it, developing a point of view, drafting, editing, and distributing. He stresses that this unsexy step is essential. The map made it easier to decide where an agent could lead, where it should assist, and where he needed to make the final call.

He then rebuilt that process as a directory of skills that runs as a plugin inside Claude. The system connects to Tenex's internal tools, including Slack, Notion, meeting notes, Linear, Git, and Gmail, to draw on his team's actual work.

Step 1: replace the blank page with ranked ideas

The first step, the Oracle, is designed to solve the blank-page problem. Instead of an empty document, it scans recent activity and returns a ranked list of "content spikes", promising ideas ready to be developed.

• Its sources include internal systems like Slack, Notion, and Gmail, plus a curated list of external accounts and websites that Alex follows.
• Each daily run reviews the previous seven days of activity, producing 15 ideas from two distinct pools.

1. The internal scan looks for stories, concrete examples, or strong points of view, scoring ideas with these elements more highly.
2. The internet reader monitors recent posts from selected people on X and LinkedIn, along with chosen websites, and suggests opportunities for Alex to respond or extend the conversation.

• Any unused ideas go into a Notion database called the Vault. Alex considers this idea-finding step valuable on its own, even when he writes the final post without further AI help.

Step 2: interview the subject-matter expert

Once Alex selects an idea, an "interview panel" of six AI personas asks him follow-up questions. The goal is to pull out specific examples and stories, which is the raw material that prevents AI slop. If a topic needs more context, a separate research assistant can prepare a brief first.

"My take is that AI slop is hilariously people just pointing the finger at themselves and saying, 'I'm not intelligent enough.' The only time the content machine actually produces slop is more of an indictment of the person not sharing good enough ideas during the interview step."

• The panel is composed of personas modeled on famous interviewers: Tim Ferriss, Joe Rogan, Michael Barbaro, Barbara Walters, Howard Stern, and Larry King. Alex answers their questions aloud using a voice-to-text tool, turning the conversation into a transcript that becomes the raw material for the draft.
• This step anchors the model in Alex’s own arguments and experiences. The AI’s job is to organize and shape his answers, but all the substantive details must come directly from the interview transcript.

Step 3: draft with a codified voice

To ensure the draft sounds like him and not a generic model, the system pulls from a personal folder containing three key files that codify his context, voice, and accumulated feedback.

1. `style-guide.md` records the user’s role, recurring topics, owned assets like newsletters or podcasts, and preferred sources.
2. `voice-guide.md` is based on an analysis of Alex’s strongest past posts. It captures his hook formulas, common structures, topic categories, characteristic language, and his core rule: write like you're texting a friend.
3. `content-lessons.md` is a running list of reusable lessons from previous edits, such as recurring problems with tone or structure, so the AI does not repeat mistakes.

The drafting agent combines the interview transcript with these three files. The transcript provides the substance, while the guides constrain the style and presentation.

Step 4: run the writer’s council

Before Alex sees the draft, a second panel of AI personas called the "writer’s council" reviews the work.

• The council includes personas modeled on writers like David Perell, Shaan Puri, and Morgan Housel, plus a custom persona Alex calls the "AI slop allergist." Each agent scores the draft on a scale of one to 10 and provides written criticism.
• If the draft's aggregate score is below nine out of 10, the system automatically revises it and resubmits it to the council. This revision loop continues until the piece achieves a perfect score.

Step 5: learn from the final edit

The machine does not publish automatically. Alex always takes the final pass and decides what goes live.

• After he makes his final edits, a "Lessons Loop" compares the machine’s last draft with his published version. It then proposes general lessons based on the differences.
• The lessons are not added automatically. The system asks Alex to approve which ones should be saved to `content-lessons.md`, which keeps him in control of how the workflow evolves.

Step 6: adapt the finished piece for each channel

Once an anchor piece is approved, Alex can ask the system to repurpose it for different channels, such as turning a LinkedIn post into a series of shorter X posts. The same voice and style files guide the adaptation, but the source material is always the final, human-approved content.

Turning content creation into a team sport

To get the whole team involved, Tenex is extending the content machine to all employees through a month-long program called the Tenex Creator Cup. The goal is to make publishing easier and more social, not to force everyone to become a full-time creator.

• Participants post on LinkedIn or X and share their work in an internal Slack channel called "Reply Guys," where teammates can see and support one another’s work.
• The program is built around a simple game structure:
• Participants earn 10 points for publishing a post and three points for engaging with a teammate’s post.
• Tenex set aside a $5,000 prize pool and created weekly categories so the program would not only reward people who already have large audiences.
• The business case is broader than just generating leads. Alex believes that if the program helps Tenex recruit a single strong engineer, the prize money is more than justified compared to a typical agency fee.

By treating employees' public expertise as an asset, the program uses participation, peer support, and clear incentives to make the shared tooling part of the team’s routine.

Two smaller personal workflows

Alex also applies the same build-for-a-specific-person approach outside the content system.

• For his wife’s job search, he built a personalized board that sent 10 scored openings each day and included an option to help complete initial applications. See Create a Personalized AI-Powered Job Board for a Smarter Job Hunt.
• For his young daughter, he has been creating custom children’s books based on their family.

The verdict: a system for substance, not just speed

Alex's content machine works because it separates idea quality, source material, voice, editorial review, and learning into distinct, diagnosable steps. When something goes wrong, it is easy to see where. A weak idea points to a problem with the Oracle or the interview. A poor draft means the voice guides or the writer's council need work. A recurring mistake becomes a new entry in the lessons file. You can follow the full implementation in Build an AI Content Machine to Beat Writer's Block and Scale Your Brand. See Build an AI Content Machine to Beat Writer's Block and Scale Your Brand.

The most useful starting point here is not the six-agent stack, but the initial workflow map. By breaking a process into its component parts, you can identify the real friction points. Alex's system is worth copying because it focuses on getting better inputs, through the Oracle and the interview, rather than just trying to polish a generic output. It still requires a human to provide the raw material and make the final publishing decision, but it successfully automates the tedious steps in between.