The Architecture of Leverage: Moving Beyond Chat to Systems

Most professionals treat AI as a conversational partner, but this misjudges its utility. By treating AI as a chatbot, you limit yourself to one-off tasks that provide only fleeting efficiency. The real competitive advantage lies in shifting from chatting to systematizing. Use tools like Claude Code or Codex to build reusable skills and automated evaluation loops. This transition changes your role from a manual laborer to an architect of your own productivity. Those who master this shift will not just save time; they will create a personal OS that compounds in value, turning the drudgery of repetitive workflows into a durable, automated engine. For the modern creator or operator, this is the difference between being a participant in the AI era and being its primary beneficiary.

The Hidden Cost of Chatting with AI

Most users approach AI through a simple, repetitive loop: prompt, receive, copy, paste. Peter Yang argues that this is inefficient because it treats every task as a blank slate. When you rely on quick back-and-forth interactions, you are essentially reinventing the wheel every time you start a project.

The reality is that high-level productivity requires skills: text files containing specific instructions that the AI can execute consistently. By documenting your workflows and turning them into modular skills, you stop performing tasks and start managing a system.

"I think anyone can save so much time of their week in three steps and the three steps are like number one, just reflect on your past week. What are the tasks that take up the most time or are the most annoying to do? ... Step two is just like literally list out every single step of that manual workflow ... Then use CodeX or Claude Code to turn it into a system."

-- Peter Yang

Why Your Evals Are Failing

The most common point of failure in AI-driven workflows is the lack of objective evaluation. Users often expect the AI to know what good work looks like, but without explicit criteria, the system drifts toward mediocrity. Yang emphasizes that an eval, or evaluation, is not a complex software term; it is a simple mechanism to teach the AI your specific standards.

The mistake most people make is asking the AI to judge its own taste or creativity. As the hosts note, AI has a bias toward its own output when asked for subjective judgment. Instead, the most effective systems rely on formulaic, pass/fail checks. If your newsletter must have a short hook and zero dashes, that is a binary check. By forcing the AI to run these checks, you create a self-correcting loop where the system iterates until it meets your predefined bar.

"If you look at what an eval is, it's actually somebody being very clear and committing to what they want and what is good. And I think so many people have a problem with that, right? Like, it's like, 'oh, I don't know what I'm actually creating so I'm just kind of all over the place.'"

-- Kieran Flanagan

The Risk of AI Psychosis and the Necessity of Taste

There is a profound consequence to automating your workflow: the atrophy of critical thinking. If you outsource every decision to an AI agent, you eventually lose the ability to start from scratch. Yang describes this as AI psychosis, where the reliance on an automated partner makes the human operator feel lazy or unable to initiate creative work without assistance.

The solution is not to avoid AI, but to maintain the role of the genesis of information. Your human taste and judgment must remain the final filter. An automated system can handle the manual assembly, such as the formatting, the research, and the data retrieval, but the core idea must originate from you. If you allow the AI to handle the synthesis of ideas, you end up with AI slop that satisfies the median but fails to stand out.

"If you don't have the human side of it then it is going to feel very average and, you know, shift to the median and not going to actually stand out and feel like AI's slop. And so it's like yeah, if you just outsourced everything to AI, it's not gonna be good."

-- Kieran Flanagan

Key Action Items

• Audit Your Week (Immediate): Spend this week tracking every manual task that repeats more than once. List every sub-step, no matter how small.
• Build Your First Skill (Next 7 days): Choose one high-frequency, low-joy task, such as podcast prep or newsletter formatting. Feed the workflow steps into Claude Code or Codex and ask it to structure them into a reusable skill.
• Implement Binary Evals (Next 2 weeks): Stop asking the AI if work is good. Create a checklist of 3 to 5 objective criteria, such as "does it follow X format" or "is the word count under Y," and require the AI to confirm these before outputting the final draft.
• Create a Skill Editor (12 to 18 months): As your library of skills grows, create a meta-skill whose sole purpose is to prune and tighten your existing instructions, removing AI slop and redundant language.
• Establish a Weekly Reflection Ritual (Ongoing): At the end of each week, feed your activity logs into your AI tool and ask: "Based on what I did this week, which workflow should I automate next?" This creates a compounding feedback loop for your personal OS.