Last October I was in a session with Evan Baehr at Arena Hall. We were designing an AI workflow system for his team, and at some point he asked a question I hear all the time: “Why does my AI keep giving me surface-level answers?”
The answer wasn’t the model. The answer was the information.
Every morning Evan started fresh. New chat. Re-explained what Arena Hall does, who was involved in the project, and what the goals were. The AI would give a decent answer. He’d close the tab. Next morning, same thing.
It’s like hiring a really smart assistant who has amnesia every day. They can do good work. But they spend the first 20 minutes of every conversation just catching up to where you left off.
The Fix: One Shared Google Doc
The concept sounds simple because it is. You create one Google Doc. Call it your master context document. It becomes the central memory layer for all your AI work.
Mine has:
- Who I am and what I’m building (business overview, current offers)
- Active projects and what stage each one is at
- How I like to make decisions and work through problems
- Key relationships — who I’m working with and on what
- Current priorities for the quarter
That’s it. No complicated system. Just a doc.
Then every Lindy agent I build gets pointed at that document. When I open Claude or ChatGPT for a real task, I paste the doc in as context. Instantly the AI has everything it needs.
In AI terms, I call this Centralized Context — agents and tools perform better when they share a single durable memory layer instead of operating as isolated chats. The preferred setup is a readable document, not some opaque hidden state that lives inside a tool you can’t see or update.
The Part That Makes It Actually Update
Here’s the part most people miss. A static document goes stale fast.
The second ingredient is a Lindy agent that keeps the document current. After every significant meeting, I run a transcript through a simple Lindy workflow: read the transcript, identify anything worth adding to the master doc, and write those updates in.
So the document is always current. It reflects what I decided last Tuesday. What changed this morning. What project just moved to the next phase.
During the session with Evan, I described the flow: “I have a chat with a new transcript, decide what’s worth keeping, then send a memo to Lindy which writes the decision into the master document.” He got it immediately. That’s the whole system. Inputs flow in. The document stays alive.
When I started running this setup across multiple tools, something changed. My AI interactions went from feeling like briefing a new contractor on every job to working with someone who actually knows the business. The questions got sharper. The outputs got closer to what I actually needed.
This Is Context Engineering
There’s a lot of talk about prompt engineering. How to write better prompts. How to structure your requests. It’s worth learning.
But context engineering is where the bigger returns are. The idea is simple: the quality of what AI can do for you depends almost entirely on the quality of information you give it. A mediocre prompt with rich context often beats a perfect prompt with no context.
Context Files as AI Assets is a framework I teach in workshops. Context files are reusable text files that encode your identity, writing style, decision patterns, business context, and working style. You build them once, maintain them over time, and load them whenever you need them. The master context document is the cornerstone of that library.
The activation signal that tells you this is missing: when your AI consistently sounds generic, when you find yourself explaining the same background over and over, or when an answer could have been written by anyone in your industry.
A Simple Starting Point
You don’t need Lindy to start this. You don’t need any special software.
Open a Google Doc. Write three paragraphs:
- Who you are and what you’re building
- What you’re working on right now
- How you like to work and make decisions
That’s your first context document. Load it into your next AI conversation. See what changes.
Most people notice the difference immediately. The AI stops asking clarifying questions you’ve answered a hundred times. The suggestions get more specific. The work feels less like a back-and-forth and more like collaboration.
From there, you can add more detail over time. Connect it to agents that keep it updated. Expand it into a full library of context files for different types of work.
But the starting point is one doc. Three paragraphs. Do it before your next AI session.
If you want to build this out properly, I walk through the full setup in my Two Hour Workday workshop. We cover context engineering, Lindy agent design, and the exact workflow I use to keep a master document current with zero manual effort.
