“hey grandpa, what was it like to use a mouse and keyboard?”
back in the good ol days, before the launch of ___ in 2027, we had these weird things called iPhones and iMacs. you’d engage them with these antiquated and limited in functionality things called “applications”. we’d call em apps. some were for browsing the web. some for listening to music and ordering taxis. so strange to think we put up with that stuff. now grab your meal pill and hop in the waymo schoolbus - you’re late!

since the beginning of human-computer interaction, developers have sought ways to translate human intention into skeumorphic affordances on a screen that translate into execution scripts on a chip. this can manifest as dragging and dropping files to a recycling bin icon, clicking a magnifying glass to search and discover information or changing the pixel colors in photoshop for your IG post in ibiza.

we labored away, clicking and clacking, typing and tapping, until the computer rendered our vision (or smth close enough). some processes were more straightfoward: writing your home address, credit card and preferred cuisine category then scrolling to choose a photo of the food to be delivered 30min later. some others a little more difficult: searching for the right movie and music to fit the staycation vibe, for example.

soon enough this will all be a thing of the past. dated, like horses and typewriters.

---

back-to-reality

can the world post phone be Good?

honestly, i think people are ready to move on from the phone. the kids are going offline. theres a growing backlash. parents are forming pacts to delay smartphones and phone-free schools are becoming the standard. “dumb phones” like the Light Phone, stone phone, are gaining traction. as you probably know, I used a flip phone during 2025. hope it doesnt come accross as luddism, bc its not. it’s recognition that infinite connectivity has costs. what form factor and interface respects human attention while still providing AI augmentation?

from Jeff Weinstein, notorious for expressing the sentiment of the ‘early adopter’ zeitgeist:

I’m extremely ready for:
1/ Siri 2.0 that is at least passible for dictation and basic tasks
2/ 3p AI assistants to work more natively on iOS
3/ Trying something better/different than an iPhone

the top comment on the johny ive x openai collab was “can’t wait for a world where Apple Watch is on my wrist; AirPods in my ears; iPhone in my pocket; Apple Glasses on my face; and this product wrapped around my neck /s“

---

software

but how exactly is this gonna look? its finally time to formalize what i’ve been yapping about for years now. since a conversation with ben on metasoftware a few years ago i got pretty ‘personalized AGI’ pilled. this has probably been evident from my other writing but essentially the new UI graphics (GUI) are going to look like … drumroll … exactly how you want it to look like.

A classic example of software is the Microsoft Office suite (word, ppt, xcel, etc). Yesterday (pre-AI), the cost to produce such software was more or less the same whether you built for 50 users, or 50,000 users, which encouraged companies to aim for the masses. the marginal cost of another copy of the .exe file doesnt require 10 more developers, or another gaggle of lawyers. same goes for salesforce, twilio, amazon and whatever canonical B2BSaaS example u have in mind too.

Now lets take the example of Nike. The cost of hosting this front end (website) is $0.000..1 for every additional visitor. A decade ago there were some clever internet advertisers who discovered that personalization drvies sales. resonance over reach. so, with utm= backlinks, IP address lookups and cookie data market identity bucketing, D2C brands like nike would “tailor” deals, featured products and landing page headlines to match that which would lead to the highest conversion rate per user visit.

this costs way too much to do for each user, so they usually have some nerd in HQ build an audience segmentation model (its just a clustering of purchase behaviors and basic demographics, not much more innovative than 100y ago), and then map the cookie data(/etc) to a basket of products. famous example is how Nurofen sells the same active ingredient and dose (ibuprofen lysine) to different branding clusters (e.g migraine pain, period pain, and back pain) and ppl will buy the ‘personalized’ colored box more than the default counterpart. its like only choosing the green mnm even tho theyre all the same.

anyway, the claim here is that we’re rapidly approaching the capability for personalized segment groups of 1 (in the digital world). how is that affordable? its just the cost of a few GPT 6.7 api calls applied to the catalogue of products/items/destinations/etc. those items are drawn from a db with all the relevant information (price, shipping estimate, etc) and you can also draw from your premade assets, or generate them to the prefered style of the visitor. for example, the copy can be in all lowercase for me, as you’ve probably noticed. shopify’s MCP is positioned very well for this in the case of D2C CPG.

tbh so many people focusing on scaling bigger and bigger models and so few realizing how distilled small models could be ‘good enough’ for MOST use cases. and these small models can run on a phone, thus negating the cost of inference, a nice change of pace from the current hosting models on datacenter servers for 8B ppl business model.

not only will the images and copy be custom per user (on websites like nike and L’Oreal) -in fact meta already has begun to offer this- but the entire layout of the website can be recreated too. you can regenerate the entire website with the right scaffolding.MD files and an understanding of the user unique preferences. for example, i might not care to see the kids catalogue, but would be interested in the Summer 2025 collaboration partners from japan since the AI cookies will know i have an upcoming trip - and in the style of tinder swipes.

this isn’t limited to websites, of course. hope that was pretty evident. you can apply the same capabilities to custom apps. reformating a music app to suggest the right music for the right time of day (eg gym, commute podcast, etc), or a transit direction map that knows your prefered ‘scenic’ route and speed limit suggestions.

this is where things can get interesting. for example, when i say “I want to listen to gaga”, it doesnt matter if its spotify, apple music or youtube that supplies the song from the POV the user, since we’re no longer using their interfaces. heck, it could even come from the labels (or artists!?) directly if they’re set up well. i can envision a DSP aggregator offering placements for each of these IP right holders to bid for. kind of like a modern doubleclick or applovin. huge startup opportunity, good luck to whoever reads this and decides to attempt it.

but dont think only of yourself! you can now have apps for you and your 5 friends (if u even have 5 lol). something to help organize the upcoming ski trip’s repayments, or summer camp’s activity votes, your local HOA etc. The internet is already blooming with announcements and launches of niche apps, crafted by and for each of the niche communities they serve.

When the cost of specialization drops, innovation accelerates. the standard will shift quickly to depth over breadth.. the same prompt, filtered through different communities, produces entirely different products. the interface becomes the brand.

---

Just-in-time UI (JIT-UI)

im going to doubleclick into this custom app stuff because it’s the undercurrent of every major AI company right now. its the promise of personal intelligence, mark would say (pastebin for the unsubscribed poor). the appless speaker from sama. and of course, the pioneering work from Google and … Microsoft? (ikr).

Google’s path is pretty obvious. It the same as apple’s siri – but actually functional lol. Gemini, by having metasoftware access, can click and swipe through any of the apps on your behalf. That (computer-use agent) stuff is a pretty skeumorphic approach, but during the transition period to AX native, itll do just fine. During Google IO they announced Android Intelligence will be able to MCP wrap all phone apps (and can make custom apps) to better navigate (beyond screenshot analysis and pixel clicks) and agentically act on your behalf.

There are two things that Google/Microsoft are doing to unlock this personal intelligence. The first is to have that OS access on android/windows/web to enact agency on the phone and operate a task. They’ve done this internally. Meta can’t do this, because they dont own the hardware. There are some hacks; a friend of mine went viral on X (10M views) for building one, and there are some other companies spinning up that can control apps on your (android) phone with agent.

With that limitation, Meta/OpenAI and friends are instead betting on building their own hardware. Zucc’s glasses are catching on and sama’s planning on selling 300-400 million devices. Aggregating across recent investor reports Meta’s spent $80B+ on this effort, and openai spent $6.4B on jony ive. It kinda goes back to the above post-phone craving, people be building their own hardware now! Cyberdecks are suuuuper trendy.

Era	Interface	User Action Required
CLI	Type exact commands	High
GUI	Click through menus	Medium
Mobile	Touch and swipe	Low-medium
Voice	Speak naturally	Low
Ambient	Just exist	Near-zero

The second is the custom JIT-UI. We no longer need to use apps that require button clicks and swipes. Agents can do so on our behalf. We just need to be able to communicate our expectations to the AI and verify the outcome is rewarded if correct (and insured if not - startup idea!). Then the POST Response received from the (e.g music) service that the Agent ultimately regenerates for human eyes can be presented in any format. Microsoft rehashes here:

We are also investing in just-in-time UI: the ability for an agent experience to adapt across devices and modalities without requiring developers to redesign everything for every new form factor. Today, that means semi-structured approaches like adaptive cards and known content types. Over time, it moves toward more dynamic and generative interfaces. This is what makes specialized form factors viable.

I too look forward to this future theyre describing:

The next platform shift is from apps to agents—from software you open to intelligence you invoke; from graphical interfaces of buttons to expressing intent through agents; and from AI operating inside your applications to agents working outside and across your apps, workflows, and devices.

Claude Code, Codex, Hermes, OpenClaw, Lovable, Bolt, Replit, Vercel, and sooooOooO0Oo many more companies are trying to be the agentic harness to creates the JIT-UI. each of them will have a default flavor and tendency. if not tuned with a reward signal from the user, the bland in-distribution creations will flood the internet/app store(/new form factor?) with banal mediocrity. Slop will spread perniciously. But thats another topic for another day.

The point im trying to get to in this essay is agents will be a new interface. No longer neccessitates phones, keyboards, screens, etc. We can write onto and read from its outputs and inputs in a more effective way if we reinvent native form factors and primitives of UI/UX. Every major consumer company is angling towards it. If you own the touchpoint, you own the experience. Delight, value and $trillions in ads is yours if you are best in market.

---

how

does this mean that apps and websites (brands?) go away in favor of the most functional TTFB service on the new form factor’s custom generated GUI? Eh im not sure, i dont want to be a __ (JH armageddon bemoaning doom ahh), but they do call it the SaaSpocolypse for a reason. Brand value may collapse if ‘efficient agents’ purchase services based on the best deal and not the name value. (hundreds of) thousands of UX dev/designers and their downstream colleagues may be repurposed to build for AX. or none of that will happen. Elon agrees with me tho (is thats a good thing?)

what is that new form factor anyway?

idk. i dont think anyone does. whoever says they do is lying and has a hidden agenda to sell u their thing. Shamelessly, i think itll be rings (and BCI inshallah). email me if u want to see what im working on.

Product	Form Factor	Status	Core Bet
Meta Ray-Bans	glasses	working	stylish + useful
Limitless	pendant	acq. Meta	meeting capture
Humane	pin	scrapped by HP	projector + voice
Rabbit R1	dead?	struggling	LAM (Large Action Model)
Friend	pendant	early	AI companion
Bee	pendant	acq. Amazon	memory/notes

in the movie her they thought itd be an airpod. sama and jony ive think it should be a family of screenless speakers. google and meta are working on glasses (apple too!). microsoft is building custom enterpise hardware aka solana. 2041 will look very different.

imo the form factor that wins will most likely be the one that disappears. maybe its ar contact lenses with display and small wrist band or ring for gyroscopic motion caption. or airpods with a camera tethered to a ‘phone’ that can offload a lot of the compute. or maybe itll be totally ambient, instead of “hey siri, what’s on my calendar,” the AI notices u looking at ur watch repeatedly and says “you have 15 minutes before your next meeting”.

there is room for startups in this world too. some of the above had exits and some died. such is the startup life. but its hinting at the early feelers that people are ready to move on from phones and AI will enable such. itll maybe be a completely different interface altogether (eg linq’s imsg + app clips, omar’s folk computer or tauvus’s PALs), TBD!

memory could be the first major success story. Granola, Limitless, Bee and Boz’s (Meta CTO) vision: a “queryable day” — everything you see and hear, indexed and searchable - is a pretty obvious and feasible win for AI. whoever builds the most useful memory system (that respects privacy/convenience tradeoff) will win big.

“For agent builders: Think big. The agents you are creating today will not be limited to the screens and devices we know today. They will be able to show up across a variety of new form factors—devices designed around them, tuned for them, and deployed into the moments where they can create the most value.” - Microsoft