The Three Ladders of AI Leverage

Three ladders, three rungs each. The right rung fits the task, not the top one.

Short answer

There are three ladders of AI leverage for PMs: personal, product and systems. On each ladder, rung 1 is AI assisting, rung 2 is handing over a task and reviewing, and rung 3 is AI completing a multi-step task and checking its results. Each rung is roughly an order of magnitude more leverage, and you always apply a final review.

Key takeaways

  • Connect your frequently used products via MCP once, and add cite your sources to outputs.
  • Ask the model to save a working flow as a skill so you stop rebuilding it.
  • Prototype on your actual codebase so the prototype inherits real components and patterns.
  • Use agent-opened pull requests for copy changes, small UI tweaks and views over existing backend code.
  • Build repeatable systems: skills, evals and connected pipelines that produce output at a trusted quality bar.

Common questions

What are the three rungs of AI leverage?

Rung 1 is AI assisting your work. Rung 2 is you handing the task over and reviewing it. Rung 3 is AI completing a multi-step task and checking its own results. Each rung is roughly an order of magnitude more leverage, with a final review always applied.

How do I move from web prototypes to real product leverage?

Web prototypes from tools like Lovable or Replit communicate ideas but the code is disposable. Rung 2 uses your actual codebase as context. Rung 3 has the agent open a pull request that an engineer reviews and merges.

What is systems leverage for PMs?

Systems leverage means building repeatable skills, evals and connected pipelines that outsource work at a quality bar you trust. A one-off good output is not leverage. A system that produces that output reliably is, and it answers the complaint that first drafts are not good enough.

CheatsheetLayout, Section, NumberedItem, Callout, } from "@/components/CheatsheetLayout";

Rung logic is the same on every ladder

Rung 1: AI assists your work. Rung 2: you hand the task over and review. Rung 3: AI completes a multi step task and checks its own results. You always apply a final review. Each rung is roughly an order of magnitude more leverage than the one below it.

Personal leverage: your own to do list

Rung 1: AI writes the text

PRDs, tickets, emails, and you copy and paste into other tools. This is where most people sit.

Rung 2: AI produces the artefact

Slides, a model, a small prototype. Expect heavy revision.

Rung 3: AI runs a task you would have delegated

The model is wired to your real tools over MCP. For example: "check whether users of social share features have higher 30 day retention, and cite sources so I can validate". Connect your frequently used products via MCP once, add "cite your sources" so outputs are verifiable, then ask the model to save the working flow as a skill so you stop rebuilding it.

Product leverage: closing the gap between what you want and what you can ship

Rung 1: web prototypes

Tools such as Lovable or Replit. Good for communicating an idea, but the code is disposable and disconnected from the real product.

Rung 2: code based prototypes

Use your actual codebase as context, so the prototype inherits real components and patterns. Ask an engineer to generate a runnable UI only repo with a local mock data store, no backend and no env vars, so you can prototype on real styles safely.

Rung 3: the agent opens a PR

An engineer reviews and merges. Best for copy changes, small UI tweaks, and views over existing backend code.

Systems leverage

Build repeatable systems: skills, evals and connected pipelines that outsource work at a quality bar you trust. This is the answer to "the first drafts are not good enough yet". A one off good output is not leverage. A system that produces that output reliably is.

The judgement line

It does not make sense to spend your time being a worse engineer than the rest of your team. Knowing when to write a doc, when to prototype and when to open a PR matters as much as the technical ability. Use it when you want to name which rung you are actually on before claiming AI leverage. If you have not levelled yourself yet, start with The Four Levels of AI Fluency. Source: Colin Matthews, "How top PMs increase their leverage with AI", guest post in Lenny's Newsletter (Jun 2026). );