Intent-based frontends: when commerce becomes situational

Frontends that breathe: a glimpse into intent-aware commerce

1. From clicks to conversation – the customer lens

You land on a storefront looking for a lightweight spring jacket. The grid loads, but as soon as your cursor drifts over linen, the gallery tilts toward earthy tones. Pause at the size guide and it sprouts an inline fit-finder, shaving a full click off your journey. Float toward checkout and the page adapts again, sliding a one-tap PayPal button into thumb-reach. No banners, no nags, just an interface that pays attention, reacts, and then fades from the story.

Quick peek at the build

This is v0.1 of the adaptive storefront running live in my browser. I pointed it at a plain JSON catalog, tracked my own mouse like a lab rat, and set the model free. Result: layout flips, PayPal button appears at the precise “buy-now” moment, colors shift toward what I actually hover. No manual rules, no refresh, just an interface that edits itself faster than I can think.

Trust me, it looks even cooler in motion.

2. Margins, momentum, and merchant magic

Traditional sites treat every visitor the same until late in the funnel. That leaves conversion upside on the table. A shape shifting UI changes the game:

Opportunity areas	Classic	Intent-based
Merchandise	Manual reorder weekly	AI ranks in milliseconds
A/B Test	Two variants per sprint	Infinite variants per visitor
Checkout	One fixed sequence	Dynamic journeys ( PayPal shortcut for impulse, full form for invoices)
Differentiation	Brand voice and price	An experience signature unique to each shopper

When algorithms soon browse and buy for us around the clock, an interface frozen in templates will not keep up.

3. How the engine really works – think game loop, not web stack

1. Input phase (≤ 5 ms)

A micro-script captures scroll vectors, dwell time, click bursts, viewport stats. Nothing invasive, but enough to feel the pulse.

2. Network tick (≈ 800 ms)

Events stream through a WebSocket heartbeat. Light, encrypted, no waiting for “page change”.

3. AI frame render (≤ 600 ms)

The server feeds three assets into a low-latency model:

current DOM scaffold
product graph
the fresh behavior log

The prompt asks for two outputs: an intent sketch and a UI diff (HTML + CSS + optional JS).

4. Commit phase (≤ 150 ms)

The diff patches the DOM in place, no repaint flashes, scroll position stays. You could measure the whole loop like FPS in a shooter: aim for 30 “UX frames” per minute and you feel the smoothness.

In short: the web page stops being a document and starts being a real-time scene graph, re-rendered by AI every few seconds. Games have done this for decades; commerce finally joins the party.

4. The larger vision – commerce becomes situational

Agentic Commerce isn’t an add-on, it is the new playfield:

Serve the agent at lightspeed → Ultra-lean APIs, pre-baked decision hooks, zero-latency payment rails (hello, PayPal Smart Buttons right in the diff).
Hyper-personalise – not 10× but 100x → Session-based UIs that compile on arrival, adapt on every gesture, and retire when the tab closes.
Lock-in through experience, not coupons → Each visit feels hand-crafted, so the brand signature follows the shopper, or their shopping bot, wherever they roam.

Once autonomous agents do the buying, humans will only glance at the UI when something feels off. Interfaces that tune themselves in real time become the trust anchor that keeps both people and bots loyal.

Agentic Commerce isn’t "next" but it’s the logical consequence of real-time, intent-aware frontends.

5. See it in action

I recorded a demo that shows the prototype in action to get a better feeling. Its not perfect. Just a small showcase.

6. Call to builders, merchants, and payment pioneers

This prototype is a proof that real-time, intent-aware UI becomes feasible. If you want to pilot the tech, challenge the approach, or help set the open standards that will guide it, let’s connect.

Commerce should feel conversational, not transactional. The interface is ready to listen – are we ready to let it speak?

— Stefan Hamann