Every AI agent I've ever built hits the same wall in the first week.
The outputs are technically correct. The tasks get done. But something is off. The tone is generic. The recommendations don't quite match how you'd handle it. You spend time correcting and re-prompting when you could be using the output directly.
Most people assume this is a prompt quality problem. Write better prompts, get better outputs. And that's true up to a point.
But there's a deeper issue: the AI doesn't know who you are.
The Briefing Problem
Think about how you'd onboard a new employee. You wouldn't just hand them tasks on day one. You'd tell them how your business operates. What your communication style is. How you make decisions. What “good work” looks like in your context. What to escalate and what to just handle.
Without that briefing, they'd complete tasks competently while missing the nuance that makes the work actually useful to you.
AI agents have the same problem. Every time you open a new chat or deploy a new Lindy agent, the AI is starting from zero. It doesn't know your voice. It doesn't know how you make tradeoffs. It doesn't know that you want outputs in Google Docs, not email, or that you prefer short answers unless you specifically ask for depth.
So you re-explain. Every time. Or you correct. Or you accept outputs that are good-enough-but-not-quite-right.
The fix is a context profile.
What a Context Profile Is
A context profile is a plain text file that encodes the things your AI would need to know about you to work well. Not instructions for a specific task — context that applies across all tasks.
I was designing AI agents for Evan Baehr and his team at Arena Hall, a real estate and hospitality project in Austin with multiple buildings, events, and a team of people. We built agents for meeting prep, email management, and weekly synthesis. And every single one needed to understand the same baseline things about Evan: who he is, how the organization makes decisions, what format he wants outputs in.
We kept re-explaining this in every new prompt. So instead, we built one context file and loaded it into every agent.
The immediate effect was subtle but consistent. Agents started responding with less generic language. Recommendations got closer to what Evan would actually do. The time spent correcting went down.
I've done the same thing for myself. I have six context files:
- How I write (voice, sentence structure, things I never say)
- How I make business decisions (what I optimize for, what I avoid)
- Current priorities (so agents know what's important right now)
- Specific preferences (how to handle uncertainty, what format to use, when to escalate)
- Team and business context (who the key people are, what we're building)
- AI behavior guidelines (don't give me a summary at the end, don't use bullet points for narrative content)
None of these files is long. Most are a page or less. But I load them into any workflow that needs to produce work that sounds like me or knows how I operate.
Context Files as Reusable Assets
The key shift is treating context files the way you'd treat a good template or a reusable piece of code: build it once, use it everywhere.
Most people write context into individual prompts. “You are an expert assistant who writes in a conversational tone and…” That's fine for one-off tasks. But if you're running multiple agents — a meeting prep agent, an email agent, a research agent, a weekly review agent — you're repeating yourself constantly and getting inconsistent results because each agent has a slightly different understanding of who you are.
When Evan's team now deploys a new Lindy agent, the context file is one of the first things they load. The agent immediately has the baseline: how the organization operates, what Evan cares about, the communication format, the key contacts. They don't start from zero.
I call this treating context files as AI assets. They're not throwaway inputs — they're something you maintain, update, and reuse. When a business priority changes, you update the context file. That update propagates to every agent that uses it.
What Goes In a Context Profile
Here's a practical starting set of headers:
Voice and communication style. How do you write? What words do you use? What do you consciously avoid? If you have writing samples, summarize the patterns.
Decision-making guidelines. How do you make tradeoffs? What do you optimize for? What are automatic no's?
Values and priorities. What matters most in your work right now? What's a distraction?
Specific preferences. These are the things you'd tell a new assistant on day one. Output format, escalation criteria, communication channel preferences.
Business context. Who are the key people? What are you building? What are the current major projects?
You don't need all of these to start. Pick the one that would most improve your AI outputs right now.
Building It Without Writing It
Here's the fastest way to create a context profile: use AI to build it.
Give Claude or ChatGPT the category headers. Ask it to interview you by asking questions one at a time. Answer as you would in conversation. Then ask it to synthesize your answers into a structured context document.
The resulting file is usually better than what you'd write from scratch, because the questions surface things you'd forget to mention — preferences so obvious to you that you wouldn't think to write them down.
The Compound Effect
The improvement from a context profile in any single response is maybe 10%. That sounds small.
But 10% better, across every AI interaction, across every agent, every day — over weeks and months, that accumulates into a qualitatively different relationship with your AI tools. You stop correcting and start using. You stop re-explaining and start building.
The best time to build your first context profile is now. The second best time is after the next frustrating AI interaction where you think “this would be good if it just knew [thing about me].”
Write down that thing. Save it. That's the beginning.
I help founders and operators design AI workflows that actually fit how they think and work — not just technically correct outputs, but work calibrated to your voice, your decisions, and your priorities. If you're building an agent system and want it to feel more like you, reach out or check out my AI consulting and workshop programs.
