# AI-Native Operating Leverage > [!abstract] The one-line thesis > When everyone has the same models, the durable advantage isn't the model — it's the **scaffolding you own on top of it**: memory, skills, and bespoke software that turn payroll into an asset. That's a new source of [[operating leverage]], and the businesses that build it get [[Process Power]] at a granularity that wasn't possible before. This is my working synthesis of Taylor Pearson's *A Lever Made of Agents*, pulled into my own frame. I'm writing it down because I want to **make this teachable** — condense it into a few core parts that stack into the whole and break back down into clean definitions. Most explanations of this stuff are still too complicated. See [[#How I'd teach this]] at the bottom. See: [[7 Powers]] | [[AI era Defensibility]] | [[085 Systems MOC]] | [[010 Venture Building]] --- ## Start with the abundance trap Every newspaper put the paper online and called it free money — readers everywhere, no printing cost. Then supply went to infinity (anyone can publish), demand didn't keep pace, and US print ad revenue fell ~80% in fifteen years. The lesson repeats: **the price of anything abundant trends to zero.** A lot of the agentic-AI excitement is vibe-coded apps and one-shot prompts. Fun, but abundant. Solve for the equilibrium: the price of anything a one-shot prompt can build is the **token cost to build it**, because anyone else can build it too. So the question I keep asking is the only one that matters — *where does durable advantage come from once the model is a commodity?* > [!warning] The newspaper question, asked of every business > Newspapers adopted the internet just fine. What they never asked was what it would do to the **economics** of their business. AI puts the same question to all of us. --- ## The shift that actually matters: chatbot → agent A chatbot in a browser tab makes you ~20% more efficient on the mildly annoying stuff and never touches the work that moves the business. Agent harnesses (Claude Code, Codex, etc.) add two things that change the category: - **Long-term memory** — it reads and writes notes to its future self that persist across sessions. - **Agentic-ness** — it executes multi-step work autonomously while I do something else. The harnesses started as tools for programmers; they're becoming the **operating-system layer for knowledge work**. See: [[AI Agents Stack]] | [[AI agents]] | [[AI usage is now a baseline expectation]] --- ## Where the leverage lives: the AI business brain The stack has two commodity layers at the bottom — the **model** (Opus, GPT, Gemini, etc.) and the **agent harness**. Don't roll your own unless you're huge. The durable leverage sits in the **local scaffolding** on top, what I'd call the *AI business brain* — the thing you own that pulls context from your services and tells the agent how your work actually works. Three core form factors: 1. **Memory** — files where the business accumulates: state of every project, history of what's been done, and the *reasoning* behind decisions. This is what makes a session feel like talking to someone who already knows the company. It's also where defensibility starts. 2. **Skills** — SOPs written for an AI instead of an employee (draft the newsletter, close the books, turn call notes into follow-ups). The agent pulls the right one when the task calls for it. 3. **Bespoke software** — from a custom CRM down to little scripts for the work that needs the same answer every time (pricing, reports, the bookkeeping math). See: [[Agent Skills as Codified Domain Expertise]] | [[AI Frameworks]] | [[PKM System]] --- ## Why this maps onto Process Power In [[7 Powers]], only some of Helmer's seven are even available to a sub-50-person business: - **Scale economies** and **network effects** are structurally closed early — they need scale and a network you don't have out of the gate. - **Switching costs** and **cornered resource** exist in weak forms (the accountant who knows your books; a below-market lease) but rarely decisive alone. - **Counter-positioning** needs an incumbent with a structural conflict — most of us compete in fragmented markets with no one to counter-position against. That leaves **Branding** and **Process Power** for essentially everyone. Most good businesses *are* exactly this: they do something well, reliably (process power), and have a reputation for it (branding). And early on, **the founder is the process power — and the constraint.** The business can't grow past their calendar; they can't take six weeks off. See: [[AI era Defensibility]] | [[Defensibility Principles MOC]] | [[Bottleneck Business]] --- ## The Four Levels of Organization Every systematization playbook (E-Myth, Work the System, Traction/EOS, Scaling Up) is really moving know-how between four levels: | Level | Where the know-how lives | Cost to run | |---|---|---| | **Software** | code — so explicit a machine runs it with zero judgment | pay once, marginal run ≈ free | | **SOPs** | paper — explicit enough a competent hire can run it | pay an expert once, then lower wages | | **Expertise** | a person's head — runs reliably, can't fully explain itself, walks out at 5pm | pay every time it runs (salary) | | **Culture** | myth — "how we do things here," usually written nowhere | implicit, high marginal cost | The axis running up the pyramid is **explicitness**. The more explicit the know-how, the cheaper each run — which is exactly **operating leverage**: costs that stay fixed while volume grows. Every procedure you push *up* the pyramid converts a slice of payroll into an asset → higher margins, or fuel to reinvest. > [!example] The payback math (memorize this) > A weekly 1-hour task, your time worth $100/hr. Writing the SOP takes ~3 hrs = **$300**. Hand it to someone at $30/hr → save $70/week ≈ **$3,600/yr**. Pays for itself in ~a month, ~1200% first-year return. An index fund at 10% takes ~25 years to return that. Most businesses have *dozens* of these hiding on someone's memory and have never added them up. See: [[10x different over 10x better]] | [[Five Step Process]] --- ## The Fifth Level: Agentic AI cuts across all of it Here's the new move. Agentic AI generates leverage at **multiple levels at once**: - **Augments expertise** — a real thinking partner when given the right context. - **Executes what used to need a human** — "read this 30-page PDF and summarize how it applies to our project" is now a computer task. - **Cheapens software** — non-developers can build bespoke tools in an afternoon. So it doesn't sit *above* software on the pyramid — it's a cloud that overlaps SOPs, expertise, and a chunk of software. The frontier between AI and the old org chart is **jagged** (uneven, domain-dependent), the same jaggedness I noted in [[AI Eats Services Not Software]]. > [!note] Process Power is compound interest, and that's the moat > Process power is a thousand tiny improvements that start to depend on each other until the system stops being legible from outside. When GM toured Toyota and copied every visible procedure, they still couldn't reproduce the results. Helmer calls the barrier **hysteresis** — the advantage can only be built by walking the path, because each improvement is the soil the next one grows in. A competitor copies your footsteps, not the power. --- ## What's genuinely new: granularity and scale Historically one person could only hold so many processes before having to hire someone to manage them. Agents change the calculus — **one person (or a small team) can now oversee an army of agents and custom tools**, operating at a scale that used to require headcount. Two concrete shapes from the article, both worth stealing: - **The newsletter (one person).** A skill ranks the month's reading by how much I marked it up, drafts a report on each top candidate, hands me 10–12 to choose from, routes leftover highlights into the relevant project drafts, and formats per platform in ~2 minutes instead of a couple of hours. Margin *and* new capability — integrating every highlight into a project never happened before. - **The financial advisor (a service business).** An agent keeps a folder per household (structure, plan, tax picture, decision log). A pre-call skill preps every meeting; a post-call skill files the *why*. Solve one client's question well → it becomes a skill that runs against all fifty households. None of the pieces is special; the **combination** is the leverage. > [!tip] Two places the freed-up time/cash can go > **Margin** (payroll-turned-asset drops to the bottom line) or **recombination** into new possibilities. For the founder who was the constraint, the job shifts from *operator* to *designer* — you stop being the router for every decision and start building the system that handles them. See: [[Consultancy-to-Platform Transition]] | [[Bespoke Engineering in Industrial AI]] --- ## So What? — for me / Navon / [[Ventures]] - The defensible asset to build deliberately is the **AI business brain**, not another wrapper app. Memory first (it compounds and it's the hardest to copy), then skills, then bespoke software. - Audit every recurring task and run the payback math. Push each one **as far up the pyramid as it can go.** Most of the ROI is sitting unmeasured. - Treat process power as a **walk-the-path** asset — sequence it like [[AI era Defensibility]]: velocity in the bailey, then fall back into the motte (embedded memory + skills + switching costs). - The opportunity is the same one the [[010 Venture Building|venture]] thesis already points at — find an [[unmet needs|unmet need]], then serve it at a scale one person couldn't before. This is the *4-Hour Workweek* moment again: a new source of operating leverage **and** a new class of business that becomes viable. The playbook isn't arbitraged away yet. --- ## How I'd teach this The reason I'm capturing this is to turn it into a **go-to simplification** — something genuinely digestible, one idea per "page," building up and breaking back down. The spine I'd use: 1. **Abundance → price to zero.** Why one-shot prompts aren't a moat. *(definition: equilibrium price = token cost)* 2. **Chatbot → agent.** The two unlocks: memory + agentic-ness. 3. **The stack.** Commodity layers (model, harness) vs. the layer you own (local scaffolding). 4. **The AI business brain.** Memory / Skills / Bespoke software — three clean parts. 5. **Why it's a moat.** 7 Powers → Branding + Process Power → operating leverage → the payback math. 6. **The fifth level.** Agentic AI cutting across the pyramid; process power as compounding/hysteresis. 7. **Scale.** One person, an army of agents — margin *or* recombination. > [!todo] Next > Build this out as the digestible guide — section tabs, one idea per page, each component added on its own page then defined, so it stacks visibly and breaks down cleanly. Pull in the extra structure/content I'm gathering. *(working surface: `ai-native-site`)* --- **Links:** - [[7 Powers]] - [[AI era Defensibility]] - [[Defensibility Principles MOC]] - [[AI Agents Stack]] - [[Agent Skills as Codified Domain Expertise]] - [[AI Eats Services Not Software]] - [[AI usage is now a baseline expectation]] - [[Bottleneck Business]] - [[10x different over 10x better]] - [[085 Systems MOC]] - [[010 Venture Building]] - [[Ventures]] - [[unmet needs]] - [[Consultancy-to-Platform Transition]] - [[Bespoke Engineering in Industrial AI]] Source: [A Lever Made of Agents — Taylor Pearson (@TaylorPearsonMe), 25 Jun 2026](https://x.com/TaylorPearsonMe)