When I sit down to solve a problem with a colleague, I don’t come with all the answers. It’s why I’m not just doing it myself: my coworker can bring to bear her unique experience and skills.

I might start by stating the problem to solve, and a few constraints. Then I’d share how I’ve thought about the problem.

But before we get too far, I’d ask her opinion. She may see something I’ve missed, or have insights or experience I lack.

Together, we’d work out the right approach before anyone gets started in earnest.

But that wasn’t how I used AI at first.

I gave it an assignment — I need a slide deck formatted just so — waited for an answer, and spent time fruitlessly arguing with it when the output inevitably wasn’t quite right.

Then my friend John gave me a better mental model:

Think of AI as a brilliant junior colleague. It’s a fast worker who knows an enormous amount, but it’s new to your field, your organization, and the kind of work you’re trying to do.

That changed how I approach AI.

I stopped treating it as a drone to execute tasks and started treating it as a collaborator.

So I’m less likely to ask for a finished product and more likely to begin with the problem: Here’s what I’m trying to accomplish. Here’s how I see it. What questions do you have? How would you approach this?

The conversation that follows is often where the work gets better. AI can surface ambiguities I hadn’t noticed, suggest approaches I hadn’t considered, or help me make sense of a half-formed idea.

When my AI collaborators make suggestions, I use my judgment about whether and how to apply them. Sometimes they merely show me what not to do — which points towards a better answer.

This is possible because of a continuous process of learning — essentially onboarding a new colleague.

At work, I started by feeding it all our marketing frameworks, legal guidelines, style guides, examples of content that had actually landed well, and details about who else is involved.

On a regular basis I share examples of what’s working and what’s not from what we produce together. We’re constantly refining processes together to make things smoother or address gaps.

I still argue with it plenty. But it’s more constructive now.

I love that open source software exists.

Not that I use it regularly. I use macOS rather than Linux. I use ChatGPT and Claude rather than one of the open-weight alternatives. In fact, most of the software I rely on every day is commercial.

What I value most about open source is how it allows thousands of people with different knowledge and interests to build on one another’s work—often enabling better approaches to come to the fore. (The scientific process and Wikipedia work much the same way.)

Consider Firefox. The open-source browser restored competition to a stagnant market dominated by Internet Explorer. It popularized user interfaces like tabbed browsing that are now table stakes for every browser on every platform. And it spurred a revolution in how websites are built by demonstrating what open standards like CSS could actually make possible.

I spent hours exploring the CSS Zen Garden website and digging into its submissions. Dozens of designers took the exact same HTML and, by changing only the CSS, produced wildly different sites. It showed how people could learn from one another by studying, borrowing, and adapting each other’s work.

Nearly every piece of software you and I use was shaped somewhere upstream by code people gave away for others to build on.

I lived for several years in an apartment building with a “community room.” Nobody used it.

It looked great on tours: a flatscreen TV, comfy-looking couches, a little kitchenette, dramatic lighting. And a pool table, obviously. But there was no reason to go there as a resident.

It was jammed into an out-of-the-way space in the basement — an afterthought filled with generic contemporary decor. It felt designed around a checklist, without any thought to why people might want to come.

A lot hinges on that why.

Lively town squares are surrounded by shops and restaurants and cafes. People come for a sandwich and then sit talking on a bench. They come so their kids can play on the swing set and then chat with their neighbors.

In offices, people gather where there’s food or coffee. There’s often more collaboration around the coffee machine than on the weird set of chairs by the elevators that nobody sits in.

Public spaces that work usually aren’t destinations in themselves. They piggyback on things people already want to do.

Go out to the beach around dawn and you’ll see people staking claims to their spots for the day.

Anyone can claim a bit of sand by placing a chair or towel. It’s first come, first served, so the sooner you get there the better.

As the beach gets busier, people will fill in around the early arrivals. Everyone has a sense of how close is too close, and there are unofficial norms about how much space you can claim, and how much margin to leave so people can walk past.

If you arrive at midday, you can claim an available space. You know almost instinctively if a space is too small — and if you judge wrong, your new neighbors will let you know.

Nobody designed these rules or posted them on a sign at the beach entrance. They emerge from thousands of individual interactions between people and adjust as conditions change. On a crowded holiday weekend, people squeeze in. On a quiet Tuesday, they spread out.

The rules change without anyone rewriting them.

Email newsletters used to be teasers. A few sentences, then “read more” — a link back to the website where the actual content lived.

That made some sense for publishers that ran ads: a click meant an ad impression on the website, and an ad impression meant revenue.

But this became the norm even for senders that didn’t depend on ads, like nonprofits. They built the same teaser-and-click model.

One big reason why? Measurement.

When the content lives in an email, there’s far less to count. No page views, no time on site, no funnel. Even open rates have gotten unreliable as email providers block marketers from seeing who opens. A click on a link is one of the only numbers left that can still be measured with any confidence.

So click-through became the metric — and then, like every proxy that stands in for a goal long enough, the metric became the target. Chasing clicks skewed the actual job of a newsletter, which is to give subscribers something worth reading.

Some writers went a different direction. Ben Thompson was sending full-text emails as early as 2013. Substack made that approach feel normal. A better reader experience, but less data.

That’s a trade worth making. A newsletter’s job is to create value for the people who signed up for it, not to generate a number for the people who sent it.