AI Is an Amplifier

The simplest way I think about AI in products: it amplifies whatever experience already exists.

Put it on a clear workflow and people move faster. Put it on a confusing workflow and people get confused faster — with more confidence, and more ways to make a mess.

That is why I get uneasy when a roadmap starts with “add AI” before anyone can say what the product helps someone finish.

Intelligence on top of friction is still friction. Just louder.

The Adoption Bottleneck

Most products I work with do not fail because they lack a model. They fail because people cannot find the next step, cannot trust the outcome, or cannot finish the job without a call to someone who “knows the system.”

That is UX. Labels. Defaults. Empty states. Error copy. The shape of the happy path. The cost of recovery when someone goes wrong.

AI cannot rescue a product that asks users to invent the process. It can only decorate the confusion.

So when someone asks whether we should invest in a smarter assistant or a clearer flow, I almost always pick the flow. Assistants are optional. Clarity is not.

Impressive vs Usable

Demo culture rewards impressive. Businesses survive on usable.

An impressive screen generates a nod in a meeting. A usable screen generates a habit. Those are different products that happen to share a codebase.

• Impressive asks: “Did this look smart?”
• Usable asks: “Did someone finish without help?”

I have watched teams polish a generative feature while the primary form still hid required fields, mislabeled statuses, or forced people through steps that did not match how the work actually happens.

Nobody remembers the clever answer if they cannot complete the boring task that brought them there.

Where AI Actually Helps

I am not anti-AI. I use it. I ship it when the job is messy input, pattern matching, or drafting something a human will review.

It earns its place when:

• The user already knows what success looks like.
• The interface makes review and correction cheap.
• Failure modes are visible, not silent.
• The alternative is hours of manual sorting or rewriting.

In those cases AI is a lever on a solid machine. Without the machine, it is theater.

Where It Quietly Hurts

The quiet damage is not a bad demo. It is the product that becomes harder to reason about.

Users stop knowing whether a result came from a rule, a default, or a model guess. Support stops knowing what to tell people. Engineers stop knowing which layer to fix when something feels “off.”

That opacity is a UX tax. Every unexplained outcome teaches people not to trust the system — and distrust is expensive to unwind.

If people cannot explain the product to a colleague in one sentence, intelligence will not save it.

What I Optimize For First

Before any model conversation, I want answers to:

• What is the job the user is hiring this screen to do?
• What is the shortest path to a correct outcome?
• What happens when they make a mistake?
• What can the product decide for them safely?
• What must stay under human control?

Those questions sound soft. They are not. They decide whether the next six months of engineering compounds or evaporates.

Good defaults beat clever prompts. Clear states beat magical answers. Recoverable errors beat confident nonsense.

I also care about the boring craft: loading states that do not lie, empty states that teach the next action, copy that names things the way the team already names them. None of that trends. All of it compounds.

Pitch Energy vs Product Reality

Pitch energy loves novelty. Product reality loves repetition. The features that matter most are the ones people touch every day without thinking — not the ones that photograph well in a deck.

AI often enters roadmaps as pitch energy: something to say in a meeting, something to differentiate on a landing page, something to calm the fear of looking behind. UX work rarely gets the same theater, even when it is the difference between a product people keep and a product people demo once.

I am not asking teams to ignore AI. I am asking them to stop using it as a substitute for finishing the basics. If the primary path is muddy, a smarter side panel will not drain the swamp.

Spend the first budget on making the ordinary path feel inevitable. Spend the later budget on intelligence that shortens judgment where judgment is genuinely expensive. That split keeps both tools in their lane.

The Order That Works

My order has become stubbornly boring:

AI can be extraordinary. UX decides whether anyone sticks around long enough to notice.

That is why I care more about UX than AI. Not as a slogan. As a sequencing decision. Usable first. Impressive later — if it still earns the room.