Context Is the New Superpower

Everyone's talking about AI models getting smarter. Bigger context windows. Better reasoning. Faster responses.

But I think we're having the wrong conversation.

The real unlock isn't smarter models. It's better context.

---

The chef who knows you

Before I was a software engineer, I was a chef. And one of the things I learned early in kitchens is that knowing your regulars changes everything.

A guest walks in and you already know: they're allergic to shellfish, they loved the lamb last time, they always want extra sauce on the side. You don't ask. You just deliver. The experience feels personal — because it is.

Will Guidara takes this idea to another level in Unreasonable Hospitality. At Eleven Madison Park, his team didn't just remember regulars — they systematically captured context on every guest. They'd note why someone was booking: an anniversary, a first trip to New York, a birthday. They'd overhear a guest mention they'd never tried a New York hot dog, and then have one beautifully plated and brought to the table mid-meal.

It wasn't magic. It was paying attention and remembering — then using that context to deliver something that felt unreasonably personal.

That's the gap between good service and unforgettable service. And it's the exact same gap in AI.

A new chef working their first shift in an unfamiliar kitchen doesn't have any of that. They might be technically brilliant — but they're guessing. Every plate is a cold start.

Most people use AI the same way. You open a new chat, re-explain who you are, what you're working on, what you've already tried. The model doesn't know your codebase. It doesn't know your goals. It doesn't know you.

It's a brilliant chef in someone else's kitchen, every single time.

---

Same model, completely different results

Here's what made this click for me.

I've been experimenting with giving an AI assistant access to my fitness and sleep data. A few weeks ago, I had a long run scheduled as part of my marathon training. But I'd slept terribly — the kind of night where you wake up more tired than when you went to bed.

A generic AI would have told me to push through. Maybe thrown in a motivational quote. Mine didn't. It could see from my sleep data that I'd had a rough night. It could see from my training history that I was in a heavy block. So it recommended a recovery day — sauna and a cold dip — instead of a run that would have done more harm than good.

Same model. Same intelligence. Completely different output — because the context was different.

That's the thing people miss. We keep asking "which model is best?" when the better question is "which model knows the most about my situation?"

---

Context at work

It shows up professionally too.

When I create Jira tickets, I give the model context of the existing codebase, the reasoning behind why a task has been requested, and the design files. The tickets it produces are genuinely good — detailed, specific, bespoke to the project.

Strip that context away and ask the same model to write a ticket from a one-line description? You get generic boilerplate. The kind of ticket that creates more questions than it answers.

The model didn't get dumber. It just lost the context that made it useful.

---

Intelligence without context is just trivia

This is the bit that keeps me thinking.

We're in an arms race over model intelligence. Every few months, a new model drops that's supposedly smarter, faster, more capable. And that's genuinely impressive.

But intelligence without context is just trivia.

A doctor who knows your medical history, your family background, and your lifestyle will give you better advice than the world's best diagnostician reading your symptoms off a card. The diagnostician might technically know more — but the doctor who knows you will get it right more often.

The same is true for AI. A slightly less powerful model with deep context will outperform a frontier model with none — on the things that actually matter to your life.

---

The cold-start tax

Every time you start a new AI conversation, you pay a tax. You spend the first few messages catching the model up. Explaining your project. Describing your constraints. Re-establishing things you've already said a dozen times before.

Most of us have accepted this as normal. It's just how AI works, right?

But it's not. It's a limitation we've normalised. And it's costing us more than we realise.

Every re-explanation is wasted time. Every missing detail is a less accurate response. Every cold start pushes the AI further from the thing that would make it genuinely useful — understanding your situation.

The people who figure out how to eliminate the cold start — whether through persistent memory, structured context, or just being more intentional about what they feed the model — will get dramatically more value from the same tools everyone else is using.

---

The gap is growing

Here's the uncomfortable truth.

As models get smarter, the gap between context-rich and context-poor usage is getting wider, not narrower. A more capable model with your full context is exponentially more useful. A more capable model with nothing? It's just a better autocomplete.

We're heading towards a world where two people using the exact same AI will have completely different experiences — not because of the model, but because of what they've given it to work with.

The differentiator won't be which subscription you're paying for. It'll be the quality of context you're able to provide.

---

What this means for you

You don't need to be a developer to think about this.

Start noticing when you're re-explaining yourself to AI. That's the cold-start tax. Start noticing when a response feels generic — that's usually a context problem, not a model problem.

The models are already good enough. The missing ingredient is almost always the same thing: they don't know enough about you and your situation.

Context is the new superpower. The question is whether you're building yours — or starting from scratch every time.

---

I'd love to hear how you're thinking about context in your own AI workflow. Reach out on LinkedIn or Twitter.