Thematic Work.

The work you do with AI is real and should not be lost across sessions or people. It deserves broader visibility across the team, especially as these tools get better. But there is a real risk: AI-generated outputs can dilute your systems and the signal from primary work if they feed back into themselves. This module is about building homes for those outputs — places where the team can reach them, where the AI is explicitly instructed not to train on them, and where the quality compounds instead of looping.

Why this matters

The work you do with AI is real and should not be lost across sessions, across people. There needs to be broader visibility for the team, especially as these tools get better. The work quality they offer is valuable time and time again.

The failure mode is the firm that uses AI and throws away what it learned. A good memo, a sharp market map, a synthetic expert call — these are assets. They should live somewhere the team can find them, not in a chat history that disappears when the window closes.

The risk of dilution

The caveat — and it is a real one — is that AI-generated outputs can dilute the systems and dilute the signal from primary work. If you are writing a memo on a specific company and drawing from an AI-generated output, you risk entering a cyclical loop of AI synthesis over AI synthesis. That is what you want to avoid.

The answer is not to stop saving outputs. It is to give them a distinct home and configure the AI explicitly — in prompts and in system instructions — so that it does not train on or look at those areas when it is doing primary synthesis. The outputs are available for the team, but quarantined from the production signal path.

Homes for AI-generated outputs

Inside each thematic folder in your drive — the structure from Module 01 · Organizing the Content — create a folder labeled AI-generated outputs. This is the holding area for:

• Reports — documents produced through Claude or other platforms, including briefs, competitive scans, and synthetic research.
• Chat history or logs — exports of AI conversations that surfaced useful reasoning, sources, or framings the team can build on later.

AI projects as compounding workspaces

A newer feature that most teams underuse is the project layer inside AI tools. In Claude, and increasingly in other platforms, a project is a home for multiple conversations along the same thread. Rather than starting from scratch every time you want to refine a thesis on energy or robotics, you open the project and the model already has the context — the files you uploaded, the instructions you wrote, the prior turns you took.

Projects have their own memory, their own instructions, their own files, and their own conversation history. They get better over time because they compound learnings instead of resetting them. The desktop version of Claude and the platform version look a little different, so think about the trade-offs — sharing, exportability, integrations — before you commit a body of work to one or the other.

The discipline is: one project per major theme or thesis thread. Not one chat per question. The team should know where the living conversation lives, and anyone picking up the thread should be able to see the history.

What comes next

Once the content is organized and AI outputs have durable homes, the question is how to turn that intelligence into action. The next module — Module 05 · Weaponizing the Pipeline — covers how to connect thematic conviction to sourcing so the right companies surface at the right moments.