top of page
Search

Fine. Put Me on the Clock.


Audio cover
The Intent Aperture

I've spent the last few years writing about polymorphic software, disappearing interfaces, synthetic labor, collapsing SaaS economics, persistent intelligence, machine-readable trust and systems that increasingly execute instead of merely advise.


I think they've all been pointing at the same thing. And I finally have a name for the part humans are going to touch.



"The Intent Aperture"



I just wrote an article about receipts...Things I had written before the headline showed up. Berkshire. Alphabet. The Navy. AI infrastructure. Power. Physical interfaces. And I made a big deal out of saying I don't really think of those things as predictions. That's convenient when you're looking backward. So let's make this uncomfortable.


August 9, 2026.

Put me on the clock...I think we are at the beginning of the first real inversion in human-computer interaction since the smartphone. Not another app. Not a better chatbot. Not an AI phone. Not whatever beautiful little object Jony Ive eventually pulls out of his pocket while somebody whispers the word magical.


Something underneath all of those. I'm calling it an intent aperture. And I think once we understand what it is, a ridiculous amount of what is happening in AI suddenly snaps into focus.



We Have a Trillion-Dollar Last-Mile Problem

Start with the absurdity of where we are. We are pouring staggering amounts of capital into GPUs, data centers, power generation, substations, networks, cooling, land and model development.


We are effectively industrializing intelligence. Great.


But somewhere at the end of that enormous machine sits a human being. And eventually that human being has to say: I want this.


Better yet: I will pay for this.


Best of all: I don't want to go back to how I did things before this existed.


That's where the entire economic argument closes.


You can build a trillion dollars of AI infrastructure. You can build another trillion after that. But if the result is still primarily a website people occasionally visit to rewrite an email, we have a problem. This is the piece of the AI bubble debate I think we've been missing.


The infrastructure can be real. The intelligence can be real. The capability can be extraordinary. But capability still has to cross the last few inches between machine intelligence and ordinary human behavior.


The aperture is where the rubber meets the road.



We're Still Trying to Build the Next Phone

Look at the AI hardware conversation. Everybody wants the phone killer.


  • Glasses.

  • Pins.

  • Pendants.

  • Earbuds.

  • Screenless devices.


AI companions. Beautiful little objects. Something you wear. Something you carry.

Something that replaces the rectangular slab we've all been staring at for fifteen years.


I think that framing is already wrong. They're looking for the next device.


I think the device is exactly what's starting to collapse.


The smartphone's great achievement was consolidation.


Camera? Into the phone.

GPS? Phone.

Music player? Phone.

Recorder? Phone.

Calculator? Phone.

Remote control? Phone.

Computer? Increasingly, phone.


A hundred specialized physical objects collapsed into one universal piece of glass.


AI may partially reverse that. Not because we suddenly need a separate calculator again. Because the intelligence no longer needs to live inside the object. That changes everything.



One Intelligence. Many Apertures.

This is the architecture I think we're heading toward: One persistent intelligence surrounded by contextual apertures.


  • Your watch is an aperture.

  • Your earbuds are an aperture.

  • Your television is an aperture.

  • Your car is an aperture.

  • Your laptop is an aperture.

  • Your glasses may become one.


Some $19 ESP32 box stuck beside a machine in your garage can be one.


The aperture is not the intelligence. That distinction matters.


It is simply the opening through which the intelligence gains enough intent and situational context to understand what you want. And that means an aperture is not merely an input device.


A microphone captures sound. An intent aperture understands that the sound came from you, here, now, in this situation, referring to this thing.


That's a completely different object.



Context Is the Compression Layer

Two things make modern AI fundamentally useful. It can interpret intention. And it gets dramatically better at interpreting intention when it has context.


Those two things are going to reshape physical computing.


The better the context, the less language I need. That's the entire game.


Imagine I'm watching a movie. Somebody blows up a hotel on a beach. I say to my television: "Is that where I stayed in 2006?"


That is a terrible prompt to a stateless chatbot.


What beach?

What hotel?

What trip?

Who are you?

What movie?

Which scene?


But a contextual aperture doesn't begin from zero.


The television contributes: I am displaying this movie...This is the current frame...This appears to be this location.


Your persistent intelligence contributes: This is Rich...He took this trip in 2006...Here are the photos...Here are the locations, if available...Here is what he means when he says "where I stayed."


Then the sentence supplies the missing piece: intent.


Now the system can answer: Yeah. Same beach. Your hotel was about half a mile east.


Then: "Show me."


Up come the pictures from 2006 beside the movie.


That sounds futuristic. The disturbing part is that almost none of the underlying capabilities are futuristic anymore. They just haven't been assembled correctly.



The TV Doesn't Need to Know My Den

Take another example.


I'm walking through Best Buy. I stop in front of an 85-inch television.


I say: "How would this look in the den?"


Best Buy has absolutely no idea what my den is. The television doesn't know whether I even have a den. And it shouldn't!


The television only needs to know what it is.


  • Its model.

  • Dimensions.

  • Panel.

  • Capabilities.

  • Maybe its physical location.


My intelligence knows the rest.


  • What "the den" means.

  • What the wall looks like.

  • What's already there.


Maybe the camera on my computer has seen the room. Maybe I have photographs. Maybe it knows I hate televisions mounted six inches below the ceiling like a sports bar run by lunatics.


The aperture contributes device context. My persistent intelligence contributes personal context. I contribute intent.


And suddenly: "It would fit, but the 85 is going to crowd the shelving. The 77 would probably look better. Want me to mock up both?"


The television doesn't need to know my den. It only needs to know what it is. My intelligence knows the rest.


That is the architecture.



Aperture Greater Than Physical App

This is the distinction I think the hardware industry is going to stumble over for a while.


We've seen ambient-computing concepts for decades...Tiny recorders...Dedicated cameras...Smart buttons...Specialized objects.


The old model was essentially a physical app.


This object records notes...This object takes pictures...This object controls music.


The capability belonged to the object. An intent aperture is different. It does not bring a capability into the world. It brings the world into the intelligence. That's a big distinction.


The microphone isn't there because this is "the recording device." It's there because hearing is useful context here...The camera isn't necessarily there to take photographs. It may exist simply because sight gives intelligence situated understanding. The screen isn't there because this is "the AI computer." It's there because sometimes the appropriate way to return information is visually.


The physical object's job becomes: Help the intelligence understand what is happening here.


That is a radically different industrial-design brief.



I Already Use a Terrible Version of This

My AirPod is probably my most important productivity interface. There is nothing special about it. It's an earbud.


But I've connected enough things behind it — Siri, shortcuts, APIs, AI systems, agents, automations — that I can say something out loud and cause some degree of intelligence to instantiate somewhere and do something.


  • Sometimes it retrieves.

  • Sometimes it creates.

  • Sometimes it starts a workflow.

  • Sometimes it invokes another system.

  • Sometimes it comes back later.


The implementation is still held together with the technological equivalent of hose clamps and prayer. But that's not the point.


The behavior works.


And this is not some special thing I invented. Anybody seriously screwing around with agents has done some version of it.


  • Hotkeys.

  • Voice pipelines.

  • MCP.

  • Bots.

  • Automations.

  • Agent harnesses.

  • Home Assistant.

  • Shortcuts.

  • Local models.

  • Cloud models.


We are all duct-taping together primitive versions of the missing product. When enough technically curious people independently build the same crappy thing, somebody eventually productizes the behavior. That's what I think happens next.



ARIA and Porthole Make More Sense to Me Now

I've spent a ridiculous amount of time playing with tiny ESP32 hardware because it's cheap enough to ask stupid questions. One became the ARIA Node...Another became Porthole.


At the time, I thought of them mostly as physical endpoints for a personal intelligence.


I think I had the right architecture but the wrong noun. They're apertures. ARIA can capture intention. Porthole can capture attention. One can let the human reach into the intelligence. The other lets the intelligence reach back into the physical world.


That relationship is far more interesting than the hardware.


In the source architecture I've been working through, the legacy model is explicitly framed as human-operated applications, dashboards and workflows, while the inversion model moves toward intent-driven capabilities, ambient sensing and "points of intention."


That phrase — points of intention — is almost there. I think the aperture is the physical and contextual manifestation of it.



Human Behavior Gets Less Technological

This may be my favorite consequence.


Computers have spent most of their history teaching humans to behave like computers.


  • Folders.

  • Files.

  • Menus.

  • Forms.

  • URLs.

  • Syntax.

  • Applications.

  • Passwords.

  • Keyboard shortcuts.

  • Workflows.

  • Even prompt engineering is another version of the same old bullshit.


Learn how to phrase things correctly so the machine can understand you. But that is not how humans naturally communicate.


Two people standing beside a broken engine can have this conversation:


"That's the noise."


"Yeah. Bearings."


Nobody says: Please analyze the acoustic characteristics of Machine Asset 17 and compare them against historical maintenance records.


Because both people are standing there. They share context.


An intent aperture lets the machine start participating in that kind of communication.


The mechanic stands beside the pump and says: "That's the noise I was telling you about."


The aperture already knows:


  • Where it is.

  • Which machine.

  • Who is speaking.

  • What "yesterday" refers to.

  • Maybe it can hear the noise.

  • Maybe it can see vibration.

  • Maybe it has the maintenance history.


Now normal human language works again. That may be the ultimate success metric for AI interfaces. Not teaching humans a better way to use computers. Allowing humans to stop performing computer behavior.



This Is Where UX/UI Goes Next

I don't think UX dies. I think UI contracts while UX becomes almost philosophical. Because if routine execution disappears behind the intelligence, the interface concentrates around the moments that actually matter.


  • What do you want?

  • Did I understand you?

  • Am I allowed to do this?

  • Should I interrupt you?

  • Is this reversible?

  • How certain am I?

  • Should you make this decision?

  • Something changed. What now?

  • Here is what I did.



That becomes the interface. So instead of asking: Where should the button go?


The design question becomes: When does the human deserve a button?


That's a very different discipline. Sometimes ambient is perfect. Sometimes ambient is horrifying.


Ordering toothpaste? Handle it...Moving a meeting fifteen minutes? Probably handle it...Sending fifty bucks? Maybe quick confirmation...Wiring half a million dollars? Yeah, friend, we're going to have some UI.


The right aperture is: as ambient as context allows and as explicit as consequence requires. That may become one of the core design rules of this whole category.



And This Is Where "Install Software" Starts Dying

Here's one prediction I specifically want timestamped.


Over the next 12 to 18 months, I think the phrase "install software" starts losing its place in normal productivity language. Not literally. We will still install plenty of things. I mean the mental model. When I need something done today, I think: What application does that?


The new question becomes: Can my intelligence do that? ...And if not: Can it acquire the capability?


Maybe the capability is an application. Maybe an API. Maybe another agent. Maybe an MCP service. Maybe dynamically generated code. Maybe an external company. I increasingly don't care.


The atomic unit moves from the application to the capability. That is enormous.

Software becomes less a thing I possess and increasingly something the system does.


I wrote about this in 2023 as polymorphic software.


A tool gets created because a particular need exists. It may disappear when the need disappears. Software starts changing grammatical form. Noun becomes verb.



Which Means SaaS Gets Weird

What happens to Salesforce when my agent is the one using Salesforce? What is a seat worth when nobody sits in it? What happens when I don't actually want accounting software?


  • I want: The books correct.

  • I don't want project-management software.

  • I want: The project on schedule.

  • I don't want a CRM.

  • I want: Customers taken care of.


That's where SaaS starts moving toward something closer to Agency as a Service.


Yes... AaaS.


Apparently the future of enterprise computing is that we will all go online and pay for a little AaaS. The internet remains undefeated. 🤣


But underneath the joke is a real transition: Tool to capability to outcome.


And once the aperture captures the intention, the intelligence can assemble whatever is necessary downstream. The user no longer needs to know which application did what.



The Human Surface Consolidates. The Machinery Explodes.

This is another apparent contradiction that I think defines the next era.


The human will interact with fewer things. Underneath, vastly more things may participate.


I say: "Get me to New York Thursday morning and move whatever you can off Thursday afternoon."


Simple.


Behind that sentence:


  • Calendar.

  • Airline inventory.

  • Payment.

  • Travel preferences.

  • Weather.

  • Transportation.

  • Company policy.

  • Perhaps another person's agent.

  • Maybe five companies.

  • Three models.

  • Ten APIs.

  • Authentication.

  • A dynamically generated tool.


Who knows? The execution graph may look like spaghetti thrown into a ceiling fan. I don't need to see it.


The visible computer gets smaller as the invisible computer gets larger.


There's your collapse.



And Then the Enemies Start Collaborating

That invisible complexity creates another inevitability.


No company owns the entire stack. Nobody owns all the models, chips, data centers, power, devices, identity, enterprise data, payments, communications, software and customer relationships required to execute arbitrary human intent.


So companies that hate one another will increasingly have to work together.


Not because they've found inner peace. Because architecture doesn't care about corporate ego.


That means we're going to see alliances among supposed enemies that create genuine WTF headlines.


The early versions are already obvious enough.


Companies can compete viciously at the user layer while simultaneously supplying infrastructure, models, cloud capacity or distribution to each other underneath.


That becomes normal.


Collaboration below. Competition above.


Because the most valuable position isn't necessarily owning every capability. It may be owning the aperture.




First Possession of Intent

This might be the biggest economic implication of the whole thing.


Google built a fortune around the query. Amazon built enormous power around purchase intent. Meta owns attention. Apple controls an extraordinary amount of device interaction.


But persistent AI potentially sits upstream of all of them.


It receives: What the human wants before that desire becomes a search, an app session or a purchase.


"I need a new air conditioner."

"Fix this."

"Get my daughter home."

"Find me a lawyer."

"Replace that software."

"Why am I paying so much for this?"


Whoever receives that intention first can influence what happens next.


  • Which model gets invoked.

  • Which vendor gets considered.

  • Which services participate.

  • Which capabilities are authorized.

  • Which transaction happens.


That's not merely search. That's first possession of intent. And I suspect that may become some of the most valuable real estate in technology.


Which brings us right back to the aperture.



Unreasonable Resolution

I've written before about unreasonable resolution.


The ability to look at something at a level where the seemingly separate pieces stop being separate.


I think this is one of those moments.


At normal resolution: AI hardware is one story.


SaaS disruption is another.

Agentic systems are another.

MCP is another.

AI infrastructure is another.

Synthetic labor is another.

Apple and Google collaborating is another.

The web being read by machines is another.

Polymorphic software is another.


At unreasonable resolution, they start collapsing into the same architecture.


Persistent intelligence receives human intent through contextual apertures, then assembles capabilities across a distributed execution layer to alter the world.


That's it.


Everything else is implementation.


The aperture is where human behavior touches the system. The capability mesh is what happens underneath. And context is the compression mechanism that makes the interaction feel natural. Once I see it that way, the AI-phone conversation feels almost quaint.



We're Probably About to Get the Flip-Phone Era

Unfortunately, I don't think the industry jumps straight there.


We're probably about to get an absolute shit soup of AI devices.


The AI pin.

AI necklace.

AI pebble.

AI glasses.

AI earbuds.

AI phone.

AI not-phone.

AI rectangle-that-is-definitely-not-a-phone-even-though-you-carry-it-in-your-pocket.


Some will be excellent. Some will be hilarious. Most will teach us something. That is exactly what happened before the smartphone interaction model settled.


There were flip phones.

BlackBerrys.

Palms.

Nokias.

Weird Windows Mobile things that absolutely were stupid and I will accept no revisionist history on this point.


It was experimentation around a category whose final interaction model hadn't crystallized.


I think we're about to do that again.


The error will be asking: What does our AI device do?


That is physical-app thinking.


The better question is: What context does this aperture contribute to the intelligence?


That's the design problem.


What can it sense?

Where does it live?

What does its physical placement imply?

How private is it?

How deliberate should its activation be?

What does the human naturally do in this context already?

And how much of the prompt can this object eliminate simply by existing there?


That is the aperture.



The Winning Hardware May Be Almost Forgettable

Which brings me to the prediction inside the prediction.


I don't think the truly successful post-smartphone AI hardware necessarily looks impressive.


It may look boring...Tiny...Cheap...Embedded...Disposable...Already familiar.


Because the expensive thing isn't the aperture. The expensive thing is the intelligence behind it.


We spent decades packing more intelligence into the object. AI lets some of that intelligence move back out. The object can become simpler again.


Your watch doesn't need to be a medical supercomputer. It needs to be incredibly good at being on your body.


Your television doesn't need to possess your entire personal history. It needs to be incredibly good at knowing what it is showing you.


The garage aperture doesn't need to understand your entire company. It needs to know it is attached to Pump 17.


Situated context becomes part of the product. And that can make a $20 object more useful in one context than a $1,500 universal computer.



This Is the Big Call

So let's strip all of this down.


I am not predicting that some particular device kills the iPhone. I actually think that may be the wrong endpoint entirely. My prediction is that over the next 12 to 18 months, we begin seeing the emergence of a new interface phylum. The industry may not call it this.


I will.


Intent apertures.


Small contextual openings between a person and a persistent intelligence.


  • Some visual.

  • Some auditory.

  • Some tactile.

  • Some deliberate.

  • Some ambient.

  • Some wearable.

  • Some bolted to a wall.

  • Some already sitting in your house waiting for the software to catch up.


The physical form won't define the category.


The architecture will:


  • Personal context.

  • Aperture context.

  • Immediate intent.

  • Authority.

  • Execution.


And downstream from that one human-interface change, the rest of the inversion accelerates.


  • Applications become capabilities.

  • Software becomes behavior.

  • SaaS becomes plumbing.

  • Agents collaborate across corporate boundaries.

  • The web acquires machine users.

  • Tiny organizations direct enormous synthetic capacity.

  • Rival companies become reluctant execution partners.

  • Trust and provenance become infrastructure.

  • And billions of dollars of AI compute finally find a natural route into the ordinary moments of human life.


That is why I think the aperture matters so much. It is the fulcrum. There can be unlimited intelligence sitting on the other side. There can be terawatts someday powering it. None of that matters if interacting with it still feels like operating a computer. The breakthrough comes when the intelligence meets us where humans already are. Talking...Looking...Pointing...Moving...Remembering.


Asking half a question because everybody in the room already knows the other half.


That is default human behavior. The technology should adapt to that. Not the other way around.


For fifty years, humans learned how to operate computers. I think we are entering the period where computers begin learning how to operate around humans. And once that clicks, I don't think the smartphone gets killed. I think something stranger happens.


It gets demoted...One aperture among many...One persistent intelligence behind all of them. And a person increasingly unaware that they are "using AI" at all.


August 9, 2026.


That's the call. The future interface isn't the next device. It's the smallest opening necessary for intention to become execution.


Intent aperture.


You're welcome.



Ad: Use code: RICH99 for a discount
Ad: Use code: RICH99 for a discount

Rich Washburn is a technologist and strategist working at the intersection of AI, infrastructure, and capital. He is Managing Partner and Chief AI Officer at Eliakim Capital.




My prior work / the "trail"

From Interface to Infrastructure: The AI Shift Most People Still Misshttps://www.richwashburn.com/post/from-interface-to-infrastructure-the-ai-shift-most-people-still-miss

The Invisible Reader: What 682,000 Hits Taught Me About Who’s Actually Reading My Websitehttps://www.richwashburn.com/post/the-invisible-reader-what-682-000-hits-taught-me-about-who-s-actually-reading-my-website



Current external signals

OpenAI — Sam & Jony / io hardware efforthttps://openai.com/sam-and-jony/

OpenAI — 900M+ weekly ChatGPT users, 50M+ subscribers, consumer/compute flywheelhttps://openai.com/index/accelerating-the-next-phase-ai/

OpenAI — testing ads in ChatGPT / consumer monetizationhttps://openai.com/index/testing-ads-in-chatgpt/

Google + Apple — Gemini models/cloud powering future Apple Foundation Models and Sirihttps://blog.google/company-news/inside-google/company-announcements/joint-statement-google-apple/

Anthropic — Model Context Protocolhttps://www.anthropic.com/news/model-context-protocol

Anthropic — MCP + code execution / connected agent systemshttps://www.anthropic.com/engineering/code-execution-with-mcp

Microsoft — Power Apps MCP server: expose application capabilities as agent toolshttps://learn.microsoft.com/en-us/power-apps/maker/model-driven-apps/power-apps-mcp-server

Comments

Animated coffee.gif
cup2 trans.fw.png

© 2018 Rich Washburn

bottom of page