Start here • first principles • reference design

Beyond model myth

AI-assisted engineering becomes durable only when workflow, verification, and governance take the place of wishful thinking about stronger models. Start with first principles if you want the argument, or open the control plane specification if you want one concrete reference design for long-running local agent workflows.

Choose a track

Where do you want to start?

01

First Principles

Why the real limits of modern LLMs should shape workflow design, architectural clarity, cost discipline, and the human role in serious AI-assisted development.

Best for: understanding the core argument

Read the argument →
02

Control Plane

A concrete reference design for durable state, slice-based execution, role separation, and executable verification in local agent workflows. Implemented concretely as the letscook Cursor plugin.

Best for: studying one concrete operating model

Study the reference design →

Executive summary

The core idea

AI-assisted engineering is not mainly about getting access to a stronger model. It is about building workflows, verification loops, and governance that can make model strengths useful without letting model weaknesses quietly run the system.

Core takeaways

  1. The limiting factor is not access to an LLM, but whether the organization can use it inside clear workflow and verification boundaries.
  2. Small, local tasks can stay lightweight; long-running or high-risk work needs explicit control, resumable state, and governed stop criteria.
  3. Architecture clarity matters because ambiguity increases both implementation error and the amount of context AI must read to work safely.
  4. Verification, review, and evidence are not optional overhead—they are part of making AI-generated work trustworthy at team scale.