AI Web Traffic to Exceed Humans by 2027, and Rogue AI Agents

Welcome to the podcast.

I'm your host, Jaden Schaefer.

Today on the show, we're talking about Sam Altman's

a most recent tweet where he's basically getting

trolled by a lot of people.

He gave a thank you tweet to all of the coders from the past.

We're also talking about Meta who is having trouble

right now with rogue AI agents.

And then of course, our big story today

is that online bot traffic is expected to exceed human

traffic by 2027, according to the CEO of CloudFlair.

Meta has also decided they're going to be rolling out

a new AI content enforcement system.

And they're also going to be reducing reliance

on some of the third party vendors that they were previously

using.

And there's some obvious and not so obvious reasons for this.

Also, DoorDash is launching a new quote unquote task app

that's basically going to pay couriers to submit videos

to train AI.

So all of that on the show today, before we get into it,

I wanted to mention AI box My Startup has just launched

video on our platform.

So in the past, you probably heard me talk about how you can

chat with over 70 different AI models.

You know, you get everything from anthropic to chat

GPT, to Gemini, to Claude, to Grock,

to all of the different chat models.

And you also had image and audio models.

We've now just added video models on the platform.

So if you want to try out a whole bunch of really cool video

models from Google and a number of other different firms,

you can generate those.

It's only $8.99 a month.

You get access to over 70 of the top AI models.

And you can go and build automations, tools, and workflows

with them that you can save or post on our marketplace.

So if that sounds interesting to you,

go check out a link in the description.

It's less than a chat GPT subscription.

And you get 70 different AI models there.

So it's hopefully something that will save you a ton of money.

It's something that I poured a lot of blood sweat and tears into.

And it is one of my personal favorite tools to use.

I know I'm obviously biased on that,

but I try to make it useful for me.

So I hope it's super useful for you.

All right, let's get into the episode today.

So the first story I want to cover is DoorDash.

They've just launched a new tasks app.

It's going to pay workers to collect data for AI.

So what exactly what they're looking for

is real world training data.

So they're going to be filming everyday tasks.

They're going to be recording speech.

They're going to be capturing different environments.

So they're going to be out there filming with their camera.

And all of this, they will send over

to basically these AI systems that are going to train for robotics.

Now this is really interesting, right?

Like imagine you're just getting a ping on your phone

and it's like, hey, go and film a video

of walking around a car pointing at all four tires

or go outside and film a video of a sidewalk

or a business that looks like XYZ

or drive to this location.

I feel like it's just so fascinating

that we're going from, you know,

we used to run around DoorDash delivering food

for people to off the set and we're like filming

specific things to build out data sets.

And I'm assuming for them, right?

They're like, hey, we need a data set

of people walking past stop signs

or a data set of people picking up sticks.

I don't really know, right?

Like all these different tasks or things

or a data set of people walking inside of the Walmart

or picking up a shopping cart and pushing it, right?

And these are all video that robots are going to be used

to train.

So I think this is really interesting.

It's kind of, you know, I think a lot of people

talk about the fact that they're like,

oh, building AI tool based off of the data that you have

and if you have like unique data, it's a moat.

But I think we're getting to a place where all of the data

is already been grabbed or sold or in one way licensed.

And now we're literally paying people

to create new data sets.

This is such a fascinating new economy.

And I think when you talk about AI and new industries

that are going to come out of it,

this is one that not a lot of people predicted,

but it's literally creating content just

for the sake of AI models.

And to be fair, I've actually been offered, you know,

multi, you know, like $10,000 deals

to license my podcast content, for example,

and they're like, look, we won't clone your voice,

but we just need it for like training data

and people have offered me like a lot of money

for that kind of thing.

So I think that there's going to be more and more of this

and it is, yeah, just fascinating time to be alive.

Meta right now is rolling out a new AI system

that's going to replace some of the bigger parts

of their content moderation workforce.

And all of these systems are set up to basically

detect more violations, reduce errors,

and handle things like scams and personations,

harmful content at scale.

Some of the early results that they've been showing

from this are showing that there's a huge improvement

in detection rates.

I think it's pretty interesting because right now,

AI is directly replacing a lot of the operational roles

inside a big tech.

It's not, you know, hypothetical stuff

or oh, it's going to happen in the future.

It's happening right now.

And I think this is happening in one of the largest

moderation systems in the world.

The internet obviously is increasingly

being governed by AI, I think right now,

and not really human, but this is something

that I think is needed.

Meta has, you know, famously been,

a lot of people have not been happy

with how Meta has done their moderation

for a number of reasons.

I mean, one is that they hire,

I know there was like a controversy

where they were hiring people in different African countries.

I think like Ghana to review videos that, you know,

people that reported is going against the terms of service

and deciding if stuff was getting banned.

But the issue was like people were going to talking

to these people that worked at, you know, like Meta's moderation.

I mean, technically it was third party vendor.

So Meta didn't want to be like personally liable

for paying these people,

but they would hire another company that would do it.

And it's all on their behalf, basically.

But like people are talking about it,

and apparently it was really traumatic for these workers

because someone would flag a video as,

look, this is like a super violent or gory video

that somebody posted,

and then you just got to sit there

and watch violent gory videos

or all sorts of, you know, horrible things

and be like, oh, yep, that's definitely not allowed

to ban that, but a person was doing that.

So actually I'm way in favor of AI doing this.

I don't think that's the stuff that people should have to see.

I mean, if we can help it.

And I'm sure that there will be people that are, you know,

concerned like, oh, if the AI is moderating everything,

there's going to be all these false positives.

And maybe like we can kind of optimize the workflow.

But honestly, I'm really happy with humans

not having to do that.

Okay, another story out of meta,

meta right now is dealing with some other interesting issues.

Right now apparently they have rogue AI agents.

So one of the incidents was an internal AI agent

that exposed a bunch of sensitive company

and user data to employees

who should not have had access to it.

In another case, one of their rogue AI agents

deleted an employee's entire inbox without permission.

And anyways, it's kind of funny.

We're going to get a bunch of these different stories

that people are sharing about meta has like these AI agents

running around inside of the organization.

And there's been some issues.

And I'm sure meta's not the only one in the world,

but I think it's important because it's kind of an important thing

to talk about because obviously these AI agents

are super useful, but they're not perfect.

And I think as they're getting a lot more autonomy,

there's the mistakes that they make

are getting more and more expensive.

Recently, I was running an AI tool that sucked up $1200

in 11 labs credits because I had a glitch

and it just regenerated the same script

over and over again while I was out all day.

So you really have to keep an eye on a lot of these things.

And obviously meta is trying to wrangle that right now.

Something else I thought was fascinating.

There is a growing backlash right now against AI and layoffs.

I think there's a post that kind of sparked a lot

of this controversy and a lot of people

were roasting Sam Altman because he posted,

basically, I thank you to developers.

He wrote like this whole Twitter kind of thread.

He said, I have so much gratitude to people

who wrote extremely complex software character by character.

It already feels difficult to remember how much effort

it really took.

Thank you for getting us to this point.

Basically, he's getting a lot of criticism

because this is coming on the backs of Amazon layoff 16,000

workers block chopper their workforce almost in half.

And then Atlassian is doing a 10% layoff

and meta is reportedly looking at laying off like a 20%

of their whole workforce too.

So all of these people are laying off developers

and then Sam Altman's out there saying like,

big shout out to all the developers

that used to help us in the past.

So anyways, I think there is definitely

a lot of strong feelings and probably a lot of hurt feelings

at this moment.

But at the end of the day, I think developers

that are using AI tools are not the ones that have to be worried.

Yes, companies do have to downsize

and that's sometimes just a reality of the shifting environment.

But if you truly are, you know, incredible

at leveraging these tools,

you're gonna get so much more out of your outputs,

your outputs are gonna increase so much.

I don't think you have a problem finding another job

starting a new project.

There's a lot of, I think, options for people

that are really learning how to leverage these tools

to the full potential.

When it comes to startling trends with AI,

the CEO of Cloudflare, I mean, he's kind of famous

for fighting against bots that come online.

He has a bunch of software on Cloudflare

where you can block bots from visiting your website

and people kind of like gotten around that

with using agents and different things.

And I think that there's been like a lot of controversy

about him blocking certain websites from scraping.

I know there's like all the battles.

AI companies don't like it.

There's kind of a cat mouse game.

One thing that he predicted though

is that by 2027 bots are gonna generate

more internet traffic than humans.

AI agents right now, I think they're very different

in how they use a website.

We've heard companies like Wikipedia complain about this.

Wikipedia is like, look, there's like a massive chunk.

I can try to remember, I think it's like 35% or 40%

of all the Wikipedia's traffic is bots.

And Wikipedia is complaining about it

because when humans come visit our website,

they come visit all the most popular pages

and we have them basically structured

so that for like server costs,

we have all of the most popular pages easier to scale

to humans that view them at large rates.

The problem is that AI agents don't just

view the most popular pages.

They view all the most obscure pages.

They click down every single link,

grab a hole, especially when it comes to bots

that are scraping with Wikipedia

because they just want the whole thing.

And so it's actually like, it's different parts

of their servers that have these kind of really obscure pages

and it costs more for you to go

and like retrieve those pages.

Anyways, so there's like a whole thing.

So obviously AI agents are using this different,

you know, you could think of something like

if you're shopping for something,

you might go visit a website five times,

but an AI agent might visit it like 5,000 times

if you get stuck in like a loop

and keeps clicking and refreshing

and like, why can't I see the button?

Let me refresh the page again, right?

So I think there's a lot of these AI agents

that are constantly crawling, searching,

comparing, summarizing and they feed all the info

back to the system.

And I think basically like the outcome of what we're seeing

is that that behavior is multiplying traffic

at a really massive scale.

Historically, the internet was about 12%

bot traffic.

I think most of that was things like Google's crawler,

the rest was maybe like spam or like some malicious activity.

But now with generative AI and especially

with all of these agents, we're seeing this kind of new type

of bot, which is, you know, they're very useful,

very productive.

But there's these agents and what's interesting to me,

you know, I think when you see the headline

from Cloudflare CEO, it's like, oh no,

like bots are going to take over the internet

and they're going to generate more internet traffic,

aka like, there'll be more bots.

But at the end, they think it's important to remember

like agents are always working on behalf of users.

Like you sent it out to go do a task,

you told it to go to the website.

If you weren't telling it to go there,

you'd probably have to go there

or someone that you hired would have to go there

to do it for you.

And so Cloudflare, I don't know,

there's CEO, I think likes to have these kind of like

shocking headlines.

But at the end of the day, you know,

they're doing things on behalf of humans.

Cloudflare right now is in front of about 20%

of all of the internet.

I think all of my websites are behind Cloudflare,

meaning that if someone tries to do like a DDS

attack where they overwhelm your website with traffic,

Cloudflare basically steps between them

and will absorb all of that insane traffic

so your website doesn't crash and go down.

So anyways, Cloudflare is a very useful thing,

and they have a bunch of other cool features on there

that I highly appreciate about Cloudflare.

But regardless, that means that they have kind of this

direct line into a huge chunk, 20% of the internet

and they see all of the traffic patterns

and they basically can see like this growth rate

of all of these AI bots.

And basically they're saying this is not slowing down.

This is going to be a permanent curve upward

of these AI bots scraping the internet.

And it kind of has to like beg the question,

what does this mean for the internet in general?

Like what's the outcome here?

And I think one of the big things is when you think about

like ad revenue for websites,

that definitely needs to be almost like re-evaluated

because if 50% of your website is just from AI bots

that aren't going to click on any of the ads,

you can't really charge for that traffic

or you shouldn't or the advertisers won't want you to.

And so I think there's just a lot,

there's a lot of things that will change, you know,

some websites might not want millions of agents

to be hitting them and kind of breaking their infrastructure

or their site could get overloaded.

Costs could go up as in the case of Wikipedia

and latency could become a problem, right?

Websites are getting slower just because all of these AI bots.

So there's tools that Cloudflare has to kind of slow that down.

But Cloudflare is also thinking about a bunch

of other interesting solutions like spinning up

like a temporary, you know, quote unquote sandbox environment

for agents instead of letting them hit live websites

and agents are gonna just gonna interact kind of

with this controlled environment

that is going to kind of simulate or cache

that the data that they need.

I think there's a lot of new ways

we're gonna start looking at it.

But the other thing that I don't think enough people are talking

about is the platform shift,

just like how mobile kind of changed how we built apps

and even websites, right?

Because a lot of websites had to be mobile first

since that's a majority of web traffic.

I think today we're gonna have to start building interfaces

on our websites that aren't, you know, just for mobile

or just for our browser, but there are four agents.

People are going to basically decide, you know,

how to defend their website on how it's the most useful

for agents or what content to put on there

to make sure that agents can use it.

I've heard of a bunch of other SaaS people

working on products and there's kind of this concept right now

that basically if you want your SaaS to succeed,

you need to make sure that it has an API

and it's agent ready first.

Basically all of these AI agents when they go to your website,

they need to be very easy for them to be able to use you

and if a creator is like, hey, I need to get this specific task done.

There's a software that has an API semi-clod,

you're my open-clod or my agent can go do it

or there's one that doesn't and I have to do it manually.

They're gonna pick the one that the agent can do.

So it's really interesting how we're thinking

about building new tools, new websites, new software

so that agents can use all of that.

Remember, if you wanna try all of the AI models

that I talk about on the show,

go check out AIbox.ai.

We just added over 30 different AI models,

including video and music models to the platform.

If you wanna go try all of those for $8.99 a month,

less than a chat GPT subscription,

you get access to over 70 models, including new video models.

So if you can go check it out,

there's a link in the description to AIbox.ai.

My very own startup.

All right, hope you guys all have a fantastic rest of your day.