Built to last: what AI will make, and what it won't

Want to guess whether AI will make something on its own? Don't ask how hard it is. Ask how long it has to last.

That one question sorts almost everything.

The shorter a thing's life, the more likely a machine made all of it. The longer its life, the more a person shaped it on purpose.

AI will keep getting better. That doesn't change the rule. Lifespan does the sorting.

Short-lived things: you move up a level

Think of this week's promo banner. A thumbnail. A summary nobody reads twice.

AI will make most of these. Good.

But you don't disappear. You just stop drawing the banner and start designing the machine that draws banners: what goes in, what each step does, where a human checks the result. The output isn't the final piece anymore. It's clay. You push it, cut it, reshape it.

Building those flows is a new skill. Few people are good at it yet. That's your opening.

Long-lived things: intention is the product

Now think of a novel. A film. Software a team will use every day for ten years.

AI still helps here. It drafts, explores, prototypes, argues back. What it can't do is care.

Work that lasts has fingerprints on it. Someone tried five approaches before picking one. Someone sweated details you'll never consciously notice (but you'd feel it if they were missing).

Skip that work and nothing breaks. The result is just forgettable. Nobody notices it. Nobody loves it. So it doesn't last.

Your software is a house

Here's the thing about software: people hate when it changes. Ship a redesign and your inbox will tell you.

So picture a house.

The fridge door changes every few days. New photos. A school notice. A takeout menu.

Every few months, you move the couch.

Every few years, after living there long enough to know what bugs you, you plan a renovation. Slowly. On purpose.

Same house, three speeds. Software works the same way. The content inside an app (a report, a suggestion, a summary) can be fresh every minute, and AI is perfect for it. The structure (the navigation, the core workflows, the way things fit together) should barely move.

When people say "AI will generate software," they mix up the two. The fridge door will be generated. The floor plan will be designed.

That's how we build Enobase. Your invoices, drafts and daily documents change all the time, and that's where AI earns its keep. The model of your business, the screens your team knows by heart, the flows they run a thousand times a week: those stay put, because your team builds its working day on top of them.

Could you just rebuild Postgres?

Sure. Maybe a strong enough AI could spit out a database or a browser in one shot.

Then what?

Writing software was never the expensive part. Owning it is. Every bug, every security patch, every new feature, every support ticket, for years. Postgres and Chrome carry decades of that work. Rebuild them and you inherit all of it, to end up with something worse than what you had for free.

And the best software keeps moving the line. Look at Notion.

Real-time collaboration and offline editing sound like solved problems. They aren't. Notion sponsored Slim Lim, one of its first ten engineers, to research the problem with the Ink & Switch lab. With Geoffrey Litt, Martin Kleppmann and Peter van Hardenberg, she published Peritext in 2022: a way to merge two people's edits to the same formatted text. The paper points at Notion itself: back then, if two people edited the same block at the same time, one edit got thrown away.

In December 2025, Notion's engineers explained how offline mode works: a local database that keeps the pages you make available offline, updates pushed from the server, and text edits that merge on their own. A few months earlier, Geoffrey Litt had joined Notion from Ink & Switch to chase the next frontier: software people can reshape themselves, with AI.

That's years of research, papers, and hires to make one experience feel effortless. No prompt does that for you.

The frontier keeps moving. And there's plenty of room out there for good engineers, now with AI in their hands.

Why everyone won't get their own app

If AI can generate interfaces, why not give every user a custom version?

People will build more software for themselves. That's already happening. But companies won't ship a different product to every customer. The reason is social, not technical.

You learn software from other people. A coworker shows you the shortcut. A friend shows you where the setting hides. A new hire watches the team. A shared interface is a shared language. Make it different for everyone and nobody can teach anybody.

Some apps are even a little hard on purpose. You can learn them, but you can't guess them. So you learn from the people already inside, and that's what makes them feel like a club. The steep first minute isn't a bug. It's the membership fee.

(And more and more, the one clicking won't be a person at all.)

Your next user is an agent

Agents are becoming real users of software. And they get better every few months.

That's great news. A smart agent doesn't need a polished screen. It can read the data directly, or build itself a view for the task at hand and throw it away when it's done. Screens become the ultimate fridge door.

So what does a brilliant agent actually need?

The same thing your team needs: solid ground. What an invoice is. Who approved it. What state it's in. When five agents and three people work on the same business, they all need one version of the truth. That's not a limit of intelligence. It's how teamwork works, for smart people too.

And the more you hand over, the more the foundation matters. Clear permissions. A full history of who did what. A structure solid enough that a fast agent can move quickly without breaking anything.

The smarter the tenant, the more the house matters.

That's why Enobase gives agents the same model, the same actions and the same permissions as your team. One house. Everyone works in it.

What this means for you

The amount of software in the world is exploding. More of it won't make yours stand out. These will:

• Design that someone thought about, not just produced.
• Craft in the details people feel before they can name them.
• A story about why the thing exists.
• A point of view bold enough that someone could disagree.

Use AI everywhere it helps. Let it handle the fridge door. Just don't ask it for the idea nobody has had yet. That one's still yours.

Building software your team will live in for years? Book a demo of Enobase.