How to Use AI Without Outsourcing Your Thinking

TL;DR

Use proprietary work data to ground AI, capture your ideas by voice before asking for edits, and divide jobs into explicit workflows with prompts that function like operating procedures. AI is most valuable when it scales documented judgment and handles execution, while the human remains responsible for ideas, context, taste, and final decisions.

Transcript

Alex Herozi has a business that makes over $250 million a year and he says AI is his number one priority. But then his wife Leila sent an internal memo to the entire company telling them to stop using AI the way they've been using it. So I dug into everything Hermosia has posted online to figure out how he actually uses AI. And after analyzing his ... Read More

Key Insights

• Proprietary data is the primary source of differentiation because general AI tools can access broad information, while internal consulting notes, outcomes, communications, and frameworks contain experience that competitors cannot obtain from public sources.
• An AI clone is a documented knowledge system that lets employees consult a leader's prior work, reasoning, and preferences without waiting for a direct response. Its value comes from organized source material rather than the underlying AI tool alone.
• Continuous data capture keeps an AI knowledge system current. Meeting transcripts and spoken end-of-day reflections can preserve decisions, lessons, preferences, successes, and failures that would otherwise remain undocumented or disappear from organizational memory.
• Voice-first drafting protects original thought by requiring a person to state the idea before AI edits it. Rough speech, notes, or prose can reveal genuine understanding more reliably than polished language generated without meaningful human input.
• AI slop is generic writing that substitutes familiar jargon and polished phrasing for clear thought. The underlying problem is not grammar, but uncertainty about whether the sender understands the message or can explain the reasoning behind it.
• AI works best as an editor and thinking partner when it sharpens human ideas, identifies gaps, and asks follow-up questions. The person remains responsible for the original insight, relevant context, judgment, taste, and final decision.
• Workflow-based thinking separates a role into the recurring processes that produce valuable outputs. This makes it easier to identify where AI can assist, what knowledge each step requires, and where human review or domain judgment remains necessary.
• A prompt is an operating procedure for producing an output. Short instructions often omit essential context, steps, constraints, examples, and quality standards, while a carefully partitioned workflow gives AI clearer guidance and makes results easier to evaluate.

Questions & Answers

Q: How can a business build a proprietary AI data moat?

A business can build a proprietary AI data moat by organizing material it has already produced, including emails, sales recordings, consulting notes, research, scripts, frameworks, and existing sequences. That material should be made available to an AI system as grounded context. Meeting transcripts and spoken daily reflections can then create ongoing pipelines that preserve new decisions, lessons, preferences, and outcomes.

Q: Why does proprietary data improve AI output?

Proprietary data improves AI output because it supplies context that general public information does not contain. Internal records show how a particular person or business communicates, makes decisions, applies frameworks, and responds to real outcomes. When AI can reference that history, it can produce guidance aligned with established experience instead of relying mainly on generic patterns and broadly available advice.

Q: How can AI scale a founder's knowledge across a team?

AI can scale a founder's knowledge by combining documented calls, research, frameworks, scripts, decisions, and past results in a system employees can consult. Team members can ask how the founder might frame an issue or approach a recurring problem without waiting for a direct reply. The system extends access to recorded judgment, although the founder still needs to review important decisions.

Q: Why should people use voice before AI when drafting?

Voice should come before AI because speaking forces the author to contribute the original ideas, context, and reasoning. A voice memo, dictated draft, or rough set of notes does not need polished grammar. AI can organize and sharpen the material afterward. This preserves the author's understanding and reduces the risk that fluent but generic language will replace genuine thought.

Q: What does outsourcing the typing, not the thinking, mean?

Outsourcing the typing means using AI to convert an existing human idea into clearer, better-structured language. It does not mean asking AI to make the underlying judgment or invent a position the user has not considered. The human supplies the insight, context, taste, and intended conclusion, while AI assists with wording, organization, editing, and questions that expose gaps.

Q: How can teams avoid generic AI-written memos?

Teams can avoid generic AI-written memos by recording their own thinking before requesting assistance. They can dictate a draft, write rough bullet notes, or describe the decision in their natural language. AI should then improve clarity without replacing the substance. Asking it to interview the author about missing assumptions or reasoning also prevents it from filling gaps with generic corporate phrasing.

Q: How does workflow-based thinking help businesses adopt AI?

Workflow-based thinking treats a job as a collection of recurring processes rather than a single block of responsibilities. A business can map the inputs, ordered steps, decisions, and desired outputs for each process, then identify where AI can assist. This approach keeps domain expertise close to the work and makes automation opportunities, dependencies, and review requirements easier to see.

Q: Why should an AI prompt be treated like an operating procedure?

A prompt should be treated like an operating procedure because both describe how to transform inputs into a desired output. A brief instruction may leave the system without context, sequencing, constraints, examples, or quality standards. Breaking the task into explicit steps allows domain knowledge to guide execution and gives the user clearer criteria for reviewing and improving the result.

Summary & Key Takeaways

• A useful AI system begins with proprietary material such as emails, sales recordings, consulting notes, research, scripts, and meeting transcripts. Organizing this work gives AI relevant context that public information cannot provide. Continuously capturing decisions and lessons also helps the system become more useful for future questions and recurring business problems.

• Human thinking should precede AI assistance. Voice dictation, rough notes, or an imperfect first draft preserve the author's real ideas and demonstrate understanding. AI can then clarify, organize, and improve that material. This approach avoids generic corporate language and keeps judgment, taste, context, and accountability with the person creating the message.

• Businesses can prepare for AI by analyzing workflows instead of treating every job as an indivisible role. Each valuable output can be separated into ordered steps, inputs, decisions, and quality checks. Detailed prompts then serve as operating procedures, allowing people across different functions to use AI within their own domain expertise.