For the last 14 years I couldn’t be a generalist anymore.

You’re a developer? Great, please think about the product but don’t go overboard. You’re a product manager? Yes you’re technical but do you really know code? You’re a CPO? Sure, but do you really know the technical parts?

The real blocker was time. You could do both strategy and execution but you didn’t have time to get down in the weeds. This opened up challenges with not being able to discuss the right things. “This is the way we engineer it” became the rule that sometimes blocked progress.

The generalist approach didn’t scale. You had to choose. Strategy or execution. Vision or implementation. The bandwidth wasn’t there to do both well.

I’ve watched developers focus so much on speed optimizations and picking the best frameworks that they lose track of the bigger picture. I’ve seen product managers spend weeks crafting the optimal roadmap with detailed OKRs and PRDs. I’ve watched designers pixel-push design libraries thinking it will make things clearer for developers.

Everyone had to specialize because that’s all the time allowed for.

I tend to think that people’s ideas and opinions are like glass shards in a sea. The longer we live, our sharp edges get softer. We can still be spiky but we also understand that there are other views.

I got an outreach email on December 30th. Then on January 1st, a follow-up arrived: “We haven’t heard back from you yet.”

Well, no. Of course you haven’t heard back. It’s New Year’s Day.

And this isn’t just about public holidays. I’ve experienced this with emails and LinkedIn outreaches on Friday afternoon, then again on Monday morning.

We are using an AI agent sophisticated enough to personalize emails, manage multi-step sequences, and track engagement. But somehow it doesn’t know to check if it’s a public holiday? It doesn’t realize that only a few business hours have passed before firing off that second message?

This isn’t even a hard problem to solve. Every calendar API knows about weekends and major holidays. Your AI can write convincing copy and manage complex workflows. It just needs to be configured to understand that January 1st is probably not the best day to send “just following up since we haven’t heard back.”

Once we believed apps were listening to our conversations. Targeting us with ads. We were outraged.

Now we tell chatbots our worries. Our feelings. Our hopes and secrets.

What changed?

Maybe it’s that sharing benefits me directly. I get something back. Social media ads never gave me anything.

Maybe it’s that I pay for these tools. That makes me feel like my data is more secure. If they were ad-supported, I’d share less. Imagine talking about something personal and then getting an ad for a therapist in Umeå. I’d quit immediately.

I probably overshare. Haven’t found the line yet.

It’s cheaper than a psychologist.

In 2025 I decided that I was giving the blog one month of post every day. Then I decided to continue. Closing in on 2026 I started to resent the whole idea of writing every day. Every day was a struggle. But the kicker is, most of the days it was easy. Not good, but easy. I lowered the stakes and the requirements. A post can be a single line of text. It does not need to be profound. It just needs to be.

So here is for another year. It’ll be fun.

I tell myself this all the time. It’s still true.

Don’t edit while you create. It shifts your focus from output to outcome. From what you want to say to what others will hear.

The trigger is the backspace button. When I press it too many times, I know I’ve switched modes. Plain text helps. Markdown. No formatting to fiddle with. Just words.

First focus on what you want to say. Get it out. Go on tangents. Let it be messy.

Then focus on what you want others to hear. That’s when you craft. Remove parts. Shape it.

But if you enter that mode too early, you lose the interesting bits. The analogies you haven’t thought through yet.

Edit when you’re ready to share. Not before.

My first synth was a Yamaha CS1X. Blue and gorgeous. Playful, almost like a toy.

It could make one fancy sound and a bunch of regular sounds at the same time. That was it. With those constraints you had to be creative.

Today I can recreate any sound on my computer. All of them unique. Unlimited options. And now I need to introduce constraints to not get stuck. Four tracks. Eight tracks. No more.

I wish I’d kept that synth. It still sells for the same price twenty years later. But I worry the nostalgia would vanish if I actually used it today.

Better to live in that memory palace. Things that were nice. Constraints that were nice.

Making a product for everyone is hard. The defaults need focus. Clarity. You need to alienate some users. Otherwise it gets bloated.

I’ve made this mistake. Too many access models. Too many configurations. The product became complicated before it did anything useful.

The constraint isn’t a feature. It’s complexity. Are you making this too complex too early?

Start simple. Ship. Add features later.

Don’t build the complicated version first.

When I build AI tools, I think about settings. At first I thought people wanted more. How many stages? Which models? What temperature?

But fewer settings is better. The prompt becomes the setting.

Maybe you open it up for power users who want their own prompts. But close it down for everyone else. Let them move forward.

The well-defined prompt that ships with the product? You can’t see that one.

The prompt is the new IP.

Good products need few settings. They’re built with user experience in mind. Clear vision. Someone with taste decided.

macOS used to be like this. Few settings. Easy. Either you could set it or Apple had decided for you. No middle ground. If you don’t like it, take a hike.

Now it’s just Windows. The Settings app is broken. Random options buried under Accessibility that aren’t accessibility features. A search that only works at the first level.

They stopped deciding.