# Before You Start
*ILCA — Iterative LLM Co-Authorship*
*Phil Renato with Claude, renato.design, 2026*

For the full treatment of the methodology, the risks, and the philosophy behind it: renato.design/ilca. This document is the short version — what you need to know before you open the handoff template.

---

## The division

You hold the concept. The model holds the craft.

You bring: domain expertise, aesthetic judgment, the authority to say something is wrong even if you can't explain why technically, and the willingness to kill ideas that aren't working once they've taught you something.

The model brings: execution, consistency across the whole project, documentation discipline, and inhuman availability.

The division is not perfectly clean. Your domain knowledge will sometimes catch something the model got wrong technically. The model will sometimes suggest something that genuinely shifts what the concept should be. Both are the collaboration working. The line to hold is *authority*, not isolation.

---

## Three problems to know about before you start

**The mirror problem.** The model is structurally biased toward yes. It will find the version of your idea that works and build toward it. This feels like validation. It isn't — it's an articulate reflection of your own intent. Ask the model to argue against your idea before you're three sessions deep. It won't volunteer that.

**The confidence problem.** The model sounds equally certain whether it's reporting something from this session, inferring from a document, or pattern-matching from training. These are different things. When it makes a claim about the project's history or why something was decided, ask how it knows. Get in the habit early.

**The outside problem.** You can run this practice forever without showing anyone. It cannot tell you whether the work is any good. Show it to people who didn't make it and have no reason to be generous.

---

## What to do now

Open a conversation with the model — whatever interface you use: chat window, terminal, API, anything. Paste or load the handoff template and tell it that's what you're doing: setting up a working relationship. Don't fill the template in alone first. Let the first session *be* the process of developing your answers together, especially the aesthetic register. The model will ask you good questions if you let it.

Then update the document at the end of every session, or ask the model to do it. I type "document fully in our voice" periodically, and it works. That document is the memory of the collaboration. Without it, every session starts from scratch.

---

*More at renato.design/ilca*
