I keep coming back to a slightly uncomfortable thought: a lot of product UI is going to matter less than we want it to.
Not because design does not matter. Bad UI still makes users hate you. Confusing workflows still create support tickets. Taste still compounds in ways that are hard to measure and very easy to miss.
But a polished interface is becoming less defensible on its own.
The thing that feels more durable is the boring stuff underneath it: the data, the storage, the permissions, the history, the integrations, the audit trail, the API, the ability to answer "where did this come from?" without everyone staring at each other.
That is what people will keep paying for.
The UI gap is shrinking
For years, software companies could win a lot of points by making the same underlying thing easier to use.
Take a messy workflow, wrap it in a cleaner interface, add a better table view, make the charting less painful, make the defaults smarter, and suddenly you had a product that felt meaningfully better than the old one.
That still works, but I think the shelf life is getting shorter.
AI makes the rendering layer cheaper. Not free, and not always good, but cheaper. Vercel's docs on generative user interfaces are a useful example of where this is heading: tool results can become dynamic UI, not just text in a chat window. If the user wants a table, chart, summary, draft, dashboard, form, workflow, or small internal app, the first version is getting easier to produce.
Some of it will be bad. A lot of it will be weird. I am not arguing that generated UI is magically replacing product teams.
The point is smaller than that: the surface area that used to require a lot of custom product work is becoming easier to reproduce.
If your advantage is mostly "we show the same data in a nicer way," that advantage has a leak in it.
The data is harder to fake
The model can make a chart. It cannot invent ten years of invoice history.
It can summarize a customer account. It cannot know which contract terms matter unless the system has them.
It can generate a dashboard. It cannot tell whether "active user" means logged in, paid, enabled, invited, recently used the product, or counted by whatever definition finance blessed last quarter.
That is where products start to separate.
Not at "can you render this nicely?", but at:
- do you have the data?
- is it clean enough to use?
- is it stored with the right history?
- can the right people access it?
- can the wrong people not access it?
- does the system know what the data means?
- can an answer show its work?
This is why I think data and storage get more valuable, not less, as AI improves.
The better the rendering gets, the more important the source material becomes.
Storage is product memory
Storage sounds like infrastructure until the user asks a question that depends on time.
What changed since last week?
Why did this number move?
Who approved this?
What did we know when the decision was made?
Can I reconstruct the state of this customer, transaction, report, account, ticket, or pipeline when something went wrong?
If the product cannot answer those questions, the AI layer does not save it. It just gives you a more confident-sounding shrug.
This is the part that makes me skeptical of thin AI wrappers. They can be useful, but without memory they do not have much gravity. They answer the current prompt and then disappear.
Systems of record are different. They collect decisions, events, states, relationships, and exceptions over time. They become valuable because they remember the work.
That memory is very hard to replace once other workflows depend on it.
Access is the product now
The other thing that feels underpriced is access.
I do not mean "we have an API" in the checkbox sense. I mean the product has a real access model: stable identifiers, permissions, exports that preserve meaning, webhooks that fire when they should, integrations that do not randomly flatten important context, and enough governance that an agent can use the system without becoming a liability.
That is not glamorous. It is also the kind of work that makes a product worth connecting to.
Users are not going to want every SaaS app to have its own little AI universe. Some will use ChatGPT. Some will use Claude. Some companies will have internal assistants. Some workflows will happen through agents that sit across multiple tools.
When that happens, the product's job changes a bit.
It still needs a good UI for humans. But it also needs to be a trustworthy source and destination for other systems.
If the only good way to use your product is through your own UI, AI makes that feel more limiting. If the product exposes meaningful data and actions safely, it can show up wherever the work is happening.
Messy data is not a moat
One caveat: "we have the data" is not automatically defensible.
Lots of companies have data. Much of it is stale, duplicated, under-modeled, permissioned in three different places, and understood mostly by whoever has been around long enough to remember the incident that created a weird column.
That is not a moat. That is a liability with storage costs.
The useful version is data with meaning attached:
- definitions people trust
- lineage that answers real questions
- permissions that are hard to bypass
- history that can be reconstructed
- quality signals
- ownership
- examples of real usage
- domain rules encoded somewhere other than vibes
This is the part I think gets more important as UI gets more fluid.
When the interface can change per user, per task, or per prompt, the system underneath has to be more opinionated about truth. Otherwise every generated view is just a new way to misunderstand the same messy data.
Where this lands
I do not think the future is "UI does not matter."
I think the future is closer to "UI is the cost of entry."
Users will still expect software to look good, feel fast, and make sense. But they will also expect the data to be available through whatever surface they are using: a dashboard, a report, a chat, an agent, an export, an API, or something custom generated for the task.
The rendering layer gets more flexible.
The valuable part is what sits behind it.
What do you store? How well do you store it? Who can access it? Can it be trusted? Does it have enough history to explain itself? Can another system use it without breaking the user's world?
That is the part people will pay for.
The UI gets them in the door.
The data keeps them there.