Context Engineering Isn't Just for AI; Here's How It Changed My Life

We spent 2025 talking seriously about context engineering to help AI agents work better. The concept dominated AI discussions: how do you structure information so that agents can reason effectively, maintain state across interactions, and actually accomplish complex tasks efficiently?

But I want to take today to share a "human prompting" tip that applies the same context engineering principles to real-world interactions. Because honestly? I don't think people do enough context engineering in their own lives... and that includes you.

That's right, YOU. I'm calling you out.

Here's what I mean.

The Moment It Clicked

Recently, someone I work with asked me a question I didn't know the answer to offhand. I could have:

• Guessed (bad).
• Asked around my immediate network (slow).
• Googled it (unreliable for this specific case).

Instead, I went directly to the one person who I suspected definitely would have the answer. Got it in minutes. Gave it to them, and they were off to the race working on what they needed.

If I'd just asked whoever was closest to me? Could've taken days, multiple back-and-forths, and probably still gotten an incomplete answer.

That's context engineering for humans: going to the right source with the right information, and providing them with enough context to get what I needed, quickly.

What Context Engineering Actually Means

Before context engineering, we focused primarily on prompt engineering: structuring information so models can understand and act on it effectively. It was about:

• Providing the right information.
• In the right format.

Prompt engineering encompasses how you frame problems, what examples you include, what constraints you specify, and how you guide the model's reasoning process

Context engineering sort of expands on the above to add the following: 

• Providing the right information.
• In the right format.
• ...To the right system.
• ...At the right time.

This is done by managing everything in the model's working memory.

Technically speaking, in AI systems, context engineering is "optimizing the utility of tokens (the data you put in) against the inherent constraints of LLMs (language models, like ChatGPT) to consistently achieve a desired outcome."

Think of it this way: AI models have a limited "context window": a finite amount of information they can process at once. Context engineering is the art of filling that window with exactly the right information:

• System instructions and examples (to tell it how to behave).
• Relevant retrieved knowledge (like RAG).
• Tool definitions the agent can use (and when / how to use them, if applicable).
• Short-term memory (conversation history).
• Long-term memory (user preferences, past interactions).
• Structured output formats.

The breakthrough insight from 2025 = most agent failures aren't model failures anymore, they're context failures. So the quality of context determines whether an AI can actually accomplish a task.

The Human Parallel

Here's where it gets interesting for regular people using AI: humans also have "context windows."

We have limited working memory. We can only pay attention to so many details at once. When someone asks us for help, the quality of information they provide determines whether we can actually help them.

And here's the problem most people face when asking for help (from AI or humans).

Here's what actually stops people from getting what they want:

They externalize their problems before they've done the internal work to make those problems solvable.

Think about the last time you felt stuck and asked for help. What did you actually say?

Probably something like:

• "I need help with my presentation"
• "I want to change careers"
• "Can someone look at this project?"
• "I'm struggling with this data analysis"

These feel like reasonable requests. But they're actually impossible to answer.

What kind of help do you need with the presentation? Design? Content? Structure? Delivery? What does "done" look like?

Change careers to what? Why haven't you already? What is actually stopping you?

What part of the project do you need me to look at? What have you tried? What's the actual blocker?

That's because you haven't done the cognitive work to:

• Define what "done" actually looks like.
• Identify what specifically is blocking you.
• Determine what kind of help would actually move you forward.

Why Proximity Becomes the Default

When you skip this internal work, you end up defaulting to asking whoever's nearest, and when they can't help you, you blame them. Not because you're lazy or don't know better, but because you literally don't know who the right person IS yet.

Sometimes you get lucky and overshoot: you somehow find the right expert. But without proper context, what you ACTUALLY need from them, they can't help efficiently either. This is why vague asks to "get a coffee" go un-answered, or why when they do happen, coffee meetings don't materialize into anything. The problem statement (the ask) has not been properly isolated.

See, the context engineering failure happens BEFORE you ask anyone.

You're trying to outsource the problem definition to someone else. You're hoping that by just stating your vague discomfort, someone will magically understand what you actually need and how to help you get it.

But that's backwards.

(We do this with AI too, which is probably why you think AI can't help you; you're not prompting it effectively. Don't worry, we'll get to that).

And now i'll explain what to do instead.

The Two TED Talks That Changed Everything

About five years ago, I watched two TED talks that completely shifted how I think about asking for help; whether from humans or AI. This advice is really important for working with AI, because, as much as we all hate asking other humans for help, “prompting” AI is, well, basically just asking for help all the time. 

But it's also important for working with other humans, and just generally getting what you want out out of life. Let me explain.

Dr. Heidi Grant: The Illusion of Transparency

Firstly, I stumbled upon this TED Talk Dr. Heidi Grant gave a on how ask for help and ACTUALLY get a yes.

‍The advice goes like this:

We humans suffer from “the illusion of transparency”, where we think it should be really obvious what we actually think, feel, and need from other people.

Spoiler alert: it is not (and this illusion applies to AI, too).

But because we think what we need should be OBVIOUS, we don’t ask for it. We sit around waiting for someone to notice our needs and spontaneously help us.

And even when people DO notice we need help, they often aren’t sure if we WANT help. Turns out humans have this weird allergy to unsolicited sentences that start with “You know what you should do…”—who knew?!

Therefore, we can't really blame other people for not helping us if we never ask them.

But according to Dr. Grant's research, when we DO ask for help, it often pays off. In fact, 90% of the help coworkers give one another comes only after explicit requests.

Now how does this apply to AI? Well, surprise, surprise.

Just like humans can’t read your mind, AI models can’t guess your unstated preferences either.

So what’s the solution? Follow Dr. Grant’s 4 step framework for asking (republished here via Rick Lindquist):

1. Be ridiculously specific about what you want (vague requests = vague results).
2. Skip the apologies and disclaimers (not as relevant for AI, but crucial for humans).
3. Talk directly to the person (phrase requests as conversations).
4. Follow up with feedback on what worked (make note and reuse successful patterns).

Point number one is what I really want to talk about, though:

“Be very, very specific about the help you want, and why. Vague, sort of indirect requests for help actually aren’t very helpful to the helper, right? We don’t actually know what it is you want from us, and just as important, we don’t know whether or not we can be successful in giving you the help. Nobody wants to give bad help…when I don’t know what it is you want from me, like the kind of help you’re hoping I can provide, I’m not interested. Nobody is.”

Now, of course AI will always act interested in what you have to say. Unlike humans, AI can’t say no and walk away. It always has to respond with something. But that creates another problem for another article: it’ll often give you bad help if you give it bad context.

Which is where TED Talk #2, and this sage advice from Barbara Sher, comes into play

Barbara Sher: Wishes and Obstacles

It was Barbara Sher's insight that gave me the complete picture.

In her fantastic TED talk, she says her absolute life-changing secret to success comes from knowing two things:

1. What you want (a wish)
2. Why you can't have it (an obstacle)

As Sher says: "If you don't say both those things, nothing happens."

She illustrates this brilliantly:

• If someone says "I'd love to be a ballerina," people just nod politely and go "Mmm."
• But if that person says "I'd love to be a ballerina, but I'm 44 years old," suddenly everyone's mental gears start turning.
• Someone might say "I heard about a dance troupe for 44-year-olds in Boston" or "There's a new ballet program I read about in a magazine."

Why? Because now people know exactly how to help. And as naturally curious, problem-solution seekers, we can't help ourselves but at least consider if we have the tools to help you.

As Barbara says, “we’re all the center of enormous amounts of information and connections that we don’t need of and we don’t think of unless somebody asks us (14:49).”

That's true for AI, which has access to the entire internet's worth of written knowledge, AND for humans, who each hold unique experiences, connections, and insights that only become available when properly prompted.

We have two magical superintelligences at our fingertips: the entirety of humanity, a collective intelligence more powerful than any AI ever created, and the sum of our recorded knowledge, in the form of AI.

‍Both require only for us to know what we need in order to ask for it.

This means your job, before asking anyone for help (and that includes AI), is to do the cognitive work to convert the following:

• "Something's wrong" → "Here's my specific obstacle"
• "I need help" → "Here's what kind of help would actually solve this"

Only THEN does the right thing to ask for become obvious. Only THEN can anyone actually help you.

Without a clearly defined goal and obstacle, you can't evaluate who actually has the capability to help or what you actually need in the first place.

Applying This to AI: Proper Prompt Engineering in Practice

This exact psychology is ~a version of what we now call prompt engineering when applied to AI.

Wish + Obstacle put another way (which is good for storytelling as well as prompting) = Goal + Constraint.

When you tell ChatGPT or Claude "Help me analyze this data," you'll get generic advice. You've given it minimal context.

Analyze for what purpose? What are you looking for?

But when you say: "Help me analyze this customer feedback data—I'm not a statistics person and I need to identify patterns for my presentation tomorrow, but I don't know how to spot correlations between customer demographics and satisfaction scores"—you've engineered the context properly.

This is what context engineering actually is: doing the internal work to make your problem solvable by others.

You've included:

• Your goal (analyze feedback for presentation).
• Your constraint (need it by tomorrow).
  
• • Which is broken down like so:
    
  • • Your obstacle to success (lack of statistics expertise).
    • Your specific knowledge gap (can't identify correlations).

Another example:‍

• ❌ "Write me an intro email to a potential client"
• ✅ "Write me an intro email to a potential client in the healthcare industry. I struggle with sounding too salesy and need to establish credibility quickly. I've never worked in healthcare before but our software has helped similar industries with regulatory compliance. Keep it under 150 words since busy executives don't read long emails."

See the difference? You're not just stating what you want, but WHY you need AI's help (your writing comes across as too salesy, you lack healthcare experience) and what specific constraints you're facing (email length, credibility concerns).

Our brains are wired to solve specific problems with constraints... and AI models work the same way.

Let's do one more example, with something that the AI actually could do, with limitations.

• ❌"Help me make an app based on this idea: [idea]"

So this is something the AI can technically do, or at least try to do. Unfortunately, neither of you know the limits of what this specific AI model can actually do (turns out, AI are actually desperately bad at assessing their own capabilities).

So sure, the AI will make you a toy app, which is fun! You'll get some code that runs.

But then what?

• Where do you want to deploy this app?
• What happens when users try to log in and their sessions expire?
• How do you handle data persistence?
• What problems do you actually need to solve to make this real?

You don't know yet, because you haven't tried to build it.

So how do you go from "I don't even know what I don't know" to "I have a specific technical question about how to refresh tokens"?

The answer: You have to actually try to build the thing.

You gotta do the pre-work. You can't prompt effectively if you haven't discovered your real obstacles yet. You have to:

1. Start building (even badly)
2. Hit a specific wall
3. Try to solve it yourself
4. Realize you're stuck on something specific
5. NOW you know what help to ask for

The "effing around and finding out" phase IS the real work. It's how you discover what you actually need.

This is why Example A fails even though AI could help; you're trying to skip the learning process where you discover your constraints.

This is also why proximity sometimes works early on. That nearby person might not be the ultimate expert, but they can help you do that initial exploration. They can help you get far enough along to discover what you actually don't understand, which then tells you who the real expert is that you need.

"Try first, fail at something specific, THEN ask better."

Then you can get to the point where you're actually exploring the right problem space. And you'll ask something like: 

"I'm building a user authentication flow for my app, but I don't understand how to implement the token refresh logic. I've read the docs but I'm confused about where to store the refresh token securely" → Now we can actually help.

The second prompt works because you've done the pre-work (as identified before). You've defined:

• Your goal (implement the token refresh).
• Your obstacle (don't understand storage security).
• What you've tried (read the docs).
• What you actually need (explanation of secure storage patterns).

The Human Version: Better Context Engineering

The same principles apply when asking humans for help:

• ❌ "Does anyone know about contracts?"
• ✅ "I need a lawyer who handles tech startup contracts in California—I'm signing my first SaaS agreement and don't understand the IP clauses, but my budget is under $2K. Does anyone have a referral?"

Or another one: 

• ❌ "Anyone know how to update the dashboard?"
• ✅ "I need to update our customer dashboard by EOD—I've never touched the API and the docs aren't clear on authentication. @ProjectOwner, can you point me to the right setup guide or hop on a quick call?"

The second versions work because they include:

• The specific wish (what you're trying to accomplish)
  
• • Clear ask (what type of help you need)
• Relevant constraints (budget, timeline, expertise level)
  
• • The exact obstacle (what's blocking you)

Understanding Mutual Limitations

Here's something crucial about context engineering: both you and your "help source" (whether AI or human) have limitations.

Your obstacles (what's stopping you):

• Skills you lack.
• Time constraints.
• Knowledge gaps.
• Analysis limitations.
• Resource constraints.

The AI's obstacles (what's stopping it):

• Token limitations.
• Context window constraints.
• Reasoning capabilities.
• Knowledge cutoffs.
• Areas where it hallucinates.
• No real-world experience.

Other Humans' obstacles (what's stopping them):

• Time availability.
• Their own knowledge limits.
• Unclear understanding of your situation.
• Uncertainty about whether you want help.
• Not knowing if they're the right person to ask.

The most effective context engineering happens when you strategically match your needs with the helper's capabilities, working around both sets of limitations.

For example, if you're trying to analyze customer feedback but lack statistical expertise (your obstacle), you might use AI—but you need to know that AI struggles with complex causal reasoning (its obstacle), so you'd structure your prompt to have it identify patterns while you interpret the deeper implications (using your human judgement, so to speak).

Or if you need legal advice (your need) but your lawyer friend practices family law not contract law (their limitation), you'd ask them for a referral to the right specialist rather than asking them to advise on something outside their expertise. That way, you can turn a "I can't help with that" into a warm lead to someone who can.

Now for How This Changed My Life

I'll be honest with you: I wouldn't be writing this newsletter if not for these principles.

In July of 2024, Noah asked The Neuron audience for writer submissions. He had a clear wish (find a writer) with a specific obstacle (needed someone who could write in his voice, understood AI, and could handle a newsletter at scale).

I had barely been checking my email that often back then. But when I saw his request, I immediately knew if I was the right person. I had:

• Experience running a 500K+ audience newsletter.
• Existing AI knowledge and passion for the space.
• Writing skills that matched the style he needed.

We had never met before. But because Noah was specific about what he needed and why, I could immediately evaluate if I was the right fit. I responded with exactly what I could offer.

A year and a half later, here we are.

None of this would have happened if:

• Noah had posted "looking for a writer" (too vague).
• I had assumed he wanted something different (illusion of transparency).
• Either of us had been unclear about capabilities and needs.

The Deeper Truth: We Need Each Other

In case you didn't know, AI is basically the entire internet all mashed together (plus billions of dollars worth of synthetic data created by experts on top of that), which means it is the amalgamation of everything that humans have ever written down on computers.  There’s sooo much knowledge inside these models that we can access if we only…

1. Tell it our wish.
2. Tell it our obstacles.

Just think of everything you can do with AI if you focus on what you want and using AI to achieve the things in your way in order to get it.

Consider your obstacles (what's stopping you):

• Skills you lack
• Time constraints
• Knowledge gaps
• Analysis limitations

How could the AI help you address those? That's what you need to figure out.

But here's what really matters:

‍AI can't wish your wishes for you.

The partnership between humans and AI works best when we're honest about our respective limitations and strengths. We need each other's complementary capabilities to create something neither could do alone.

What About Humans?

There will always be more wishes and obstacles as we continue to live and work in the real world. Leaning on each other to solve the things AI can't (and vice versa) will be how we thrive in a world where AI becomes increasingly capable.

As Sher says in her talk, "isolation is the dream killer."

AI excels at solving specific problems we identify, while we humans shine at imagining what's worth creating in the first place. AI can help us overcome obstacles, but it can't dream up our wishes or tell us what matters.

The real magic happens when we bring our uniquely human creativity and purpose to the table, then let AI help us bring those visions to life. Together, we can accomplish things neither could do alone.

The Creativity Explosion

Sam Altman tweeted awhile back that he thinks "this is going to be more like the renaissance than the industrial revolution."

What does that actually look like?

Imagine: artists collaborating with AI to create entirely new mediums of expression. Musicians exploring soundscapes that were technically impossible to produce before. Writers developing interactive stories that adapt to readers' emotions. Entrepreneurs building businesses that solve problems we couldn't even approach until now.

A true creative explosion happens when we use AI for what it does best (solving defined problems, processing information, generating options) while we humans do what we do best—connecting with each other, sharing our unique experiences, imagining what's worth creating, and helping each other reach our dreams.

We still need each other AND we need AI. Together, that's where the real magic happens.

Why You're Reading About AI

Think about it: You want to learn about AI because you want it to help you do something.

Why?

You need AI to do the things you don't know how to do. You need it to do the work you want done but won't take the time to do. You want it to make you faster, make you more money, make you more effective.

But what will you do when AI handles all those tasks? What would anyone do if machines are doing all the work?

Simple. You dream new dreams.

Ask any successful founder, and when they finish one company, it's off to the next. They keep the money to live on while they pursue some other impossible-seeming task.

The Framework for Impossible Dreams

In order to dream fantastic dreams and actually achieve them, you need:

1. A goal (your wish)
2. Obstacles (things in your way to overcome)
3. The willingness to ask for help
4. And the wisdom to know who to ask.

Put together, this is proper context engineering.

If you can focus on a goal, identify the obstacles between you and that goal, and ASK FOR HELP with proper context, you can accomplish anything—whether with AI, humans, or both.

The obstacles aren't actually stopping you. They're helping you clarify what you need and who to ask.

The Action Item

Stop hoping AI (or humans) will "just know" what you want.

Start focusing on engineering the context properly:

• State your wish clearly
• Name your obstacles specifically
• Go to the right source
• Provide complete information for them to help

Whether you're prompting Claude or reaching out to a colleague, good context engineering is the difference between days of confusion and minutes of clarity.

So go dream your impossible dreams. You'll find the things in your way are actually helping you: as long as you're willing to do the work to identify them...