Permission is forever

OpenAI is building a local app to get around the permission system of the internet.

Permission is forever

It's Sunday.

On Sundays you catch up with some newsletters and check out what you missed out on, lately that's mostly been new ai apps doing some pretty cool stuff.

Have you seen the foundation model for biology?

Or what about the AI Science Researcher you can hire for $16 / month that has essentially read all human scientific knowledge?

Pretty crazy what some of these tools might unleash.

If it's got a free tier you pop in and hit the "Sign in with Google" button and go check it out.

The app now can say hi to you with your first name and has a nice avatar you recognize as yourself and makes you feel at home.

You write the url down in your notes.

It's a cool app.


Throughout our days we use some version of "Sign in with Google" or "Sign in with Facebook", email and password etc.

This is just part of the everyday infrastructure of the internet.

It's just how stuff works.

A concept familiar to you, that weird guy at the gas station, and a 6 yr old subsistence farmer in India.

We know that logging into facebook is one thing and logging into google is another.

Separate worlds for separate purposes each with their own set of rules.

In some ways it determines the shape of the internet, and by defining the current shape of the internet it's also defining the shape of the future.

"Having permission" is defining the shape of AI as we speak.

There's a reason Apple is throwing caution to the wind and relaxing some of its long held stances related to privacy to build an AI Agent on your iPhone.

It's the same reason why OpenAI is building a local application for your computer to use ChatGPT.

It's permission.

The optimists are already busy building a future where software that has brains that act like ours can give us super powers and automate nearly anything we desire.

Planning a complicated island hopping excursion to Greece?

Sounds delightful.

One 5 minute voice chat about some aesthetic desires or creature comforts and its booked, payed for, on your calendar, and some new clothes are even on the way to your house so your style fits your scenery on Instagram.

In the background the ai used 2 airline sites, 3 ferry booking services, viewed 92 instagram accounts, read 32 blog posts, purchased rentals on 4 platforms, and ordered from 3 Shopify stores on your behalf.

This is magic.

To make this magic possible the ai needed access to everything.

It needed permission.

The tricky thing about permission is that not everyone gives it.

We all have our own priorities.

The airlines have a competing ai agent so don't let other companies integrate.

The ferry companies have had the same wordpress site since 2009 and just don't care.

Instagram has taken the stance that they won't give 3rd parties API access to fight against AI spam.

The blogs have followed a cultural movement to kindly block AI model training on their websites by requesting it in their robots.txt so the ai doesn't already have the data baked into its model about the best way to travel in Greece this season.

Shopify has leaned into it's libertarian streak and turns stores ai services off by default and most stores forget to turn them on.

Your credit cards just want you to spend as much as possible so they actually love this stuff and let anyone do whatever in the hopes you go a bit past your limit.

You get the point.

So if the ferries don't care, the cards go brrr, and 90% of the modern services you use probably put limits on what type of integrations 3rd parties can build how does this future even get built?

If the builders don't have permission to integrate ai doesn't that mean it just doesn't happen?

Not so fast.

The way Apple is "solving" this is by leveraging its power to convince every app developer to give their AI permission.

Instead of going around permission they are trying to convince every app developer to create a way for apple intelligence to do things in apps on the users behalf.

In essence iPhone apps have buttons for humans to click and will have "app intents" that the Apple Intelligence can push on behalf of the user in the background.

Apple, Microsoft, and Google are probably the only companies on the planet that can use this strategy.

So does that mean they own the future of AI?

Thankfully, no.

There is another way around this problem and its the direction OpenAI and a thousand other apps are going.

Skipping permission altogether.

At first glance this one line seems scary.

But wait!

The permission they are going around is not yours.


The permission they are going around is the permissions of the websites and apps you use.

Since every website or app on the planet has been built for humans to use it also means that technology modeling humans can use them.

If you have built an app that lets users log in, read a screen, and take actions an AI model that can see, read, and click can do the same.

It can take action.

This is why OpenAI is building a local app.

It's already possible for applications to run your computer as if it's a human.

An AI working on your desktop this way is also imbued with the same permissions as you.

The ai can log in as you.

The ai can use the credit card stored in your chrome browser.

The ai can read the charts in that niche legacy windows app you still use and click "print".

Anything you can do it can do (better).

And it doesn't need anyone's permission but yours.

Local ai apps are how you get around the legacy permission system of the internet and set AI automation free.


This isn't going to happen overnight but its already happening.

There are probably a lot of tricks up the future's sleeve on how this plays out.

We likely see new companies take this into consideration and use it as an advantage, building tools that can be easily co-piloted by ai.

Maybe we get lucky and move to a more decentralized permission system like exists in crypto.

No matter what changes in AI.

This part will shape it.

Permission is forever.