Writing

Curate Your Context

AIContext EngineeringCuration
Curate Your Context
Image by Patrick Morgan, created with Midjourney

Your AI is only as good as the context you give it.

Too little context and it guesses wildly. Too much and it drowns in noise. But get it just right and suddenly the work hits the mark.

This is exactly the problem I’ve faced for years when working with new design clients. If all they tell me is “we need a new website,” I can make some baseline assumptions, but the odds of delivering what they actually want and need are low. I need more context.

But if they give me a massive data dump of everything about their company, their competitors, their quarterly reports—now I’m swimming in potentially irrelevant information. What do I pay attention to? What do I ignore?

The same dynamic plays out with AI. That curation skill—knowing what context matters and what’s just noise—is also essential for getting usable outputs from AI.

Over the last two weeks, I’ve written about getting your thinking into text and using voice to make that capture effortless. Now comes the next step: learning to curate that material so it actually works for you, not against you.

The digital hoarding trap

Creative professionals have always curated inspiration. Marketers keep swipe files of ads that work. Designers build mood boards before starting projects.

Digital tools made collecting all of this frictionless. Unlimited bookmarks. Endless screenshot folders. Voice memos that pile up. There’s no limit to what you can save so you just keep gathering.

But there’s a difference between collecting and curating.

With collecting, building the collection itself is often the end goal. Not so with curating. Curation brings intention to what’s being collected and points it towards a purpose. Collection gathers raw material. Curation creates the signal from it.

What’s changed with AI is that your collections can now function as machine-readable context and instruction. Your swipe file, your documented processes, your captured thoughts—none of this is just for your own reference anymore. It can inform an AI agent working alongside you. If you’ve curated this material wisely, your AI becomes a well-informed collaborator. If you haven’t, your information dump just pollutes the context window and you get results no better than if you’d provided nothing at all.

Curating the right signal

Language models have finite attention. Good context engineering means finding the smallest possible set of high-signal tokens that maximize the likelihood of your desired outcome.

You don’t give your agent all your context all the time. You build a library of potential context, then strategically select what matters for the work at hand.

The curation happens in two places.

First, what you choose to keep. You need to be active in maintaining this resource. Add things, remove things, update things. ‘More’ is not better in this environment. Do what you can to tend to the signal in the noise. This is the work that turns collecting into curating.

Second, what you choose to share with the AI agent for a given task. You don’t have to (or want to) attach your entire library to every conversation. You pull the specific examples, processes, or thoughts that are actually relevant to the work at hand. Give the agent just what it needs to understand what you’re trying to achieve. Guide its attention purposefully and the results will show it.

This process is similar to how a seasoned creative director operates. You have years of references stored away, but you only pull what’s relevant for the specific brief at hand. You intuitively know what signal matters for a given project and what would just be distraction. This is the information you use to guide a team from creating anything to creating the specific thing a project demands.

Strategic context compounds

The difference between working with an agent that mostly guesses and gets a lot wrong versus one that mostly understands and gets a lot right comes down to this: whether you’ve done the work of curating its context.

Each piece you curate becomes reusable creative capital. A well crafted artifact doesn’t just help with one project, it provides a foundational reference for many. It helps the agent work from your perspective, getting better results, faster, more often.

That’s a whole lot of new value from adding just a bit more intentionality to the context you save and share.

Until next time, Patrick

In Practice

Try this

Add curation to what you’ve already captured

If you’ve been following along the last two weeks, you’ve been getting your thinking into text and using voice to capture it. Now it’s time to curate some of that material.

The goal: Transform raw captures into curated context you can strategically deploy. You’re not just collecting anymore, you’re building a library you can actually use.

Pick 2-3 notes you’ve created recently and add curation:

  1. Open a note from the last two weeks — Decide whether to keep it, kill it, or archive it
  2. Add the “why” — Add 2-3 sentences explaining:
    • What makes this valuable
    • What it relates to
    • When you might reference it again
  3. Tag it for retrieval — Make it easy to find when you need it (a clear filename, a tag, a folder—whatever works in your system)
  4. Test strategic deployment — Next time you work with AI on a relevant task, explicitly reference one of these curated notes in your prompt

Go deeper