Is AI Safe for Students?
Welcome to make EdTech 100.
I am LindyHoc Educator, K 12
Ed Tech Advisor, and your host.
This is a podcast where we keep it real
about what actually works in classrooms.
No hype, no overwhelm, just practical
strategies, honest stories and
tools that make a real difference
for teachers and students.
So come along with me on a
journey to make EdTech 100.
Lindy Hockenbary: So if you didn't
know, and if you're not keeping
up, there's a lot happening with AI
developments and advancements out there.
So much so that I have only been barely
keeping up, and it is my job to keep up.
So I know it's bad and
I know it's moving fast.
I've been struggling
for the last few years.
To keep up because of all the
advancements, but the last couple
months especially have just been crazy.
There's a lot of big
jumps in AI capabilities.
So for this episode, I'm
doing a little bit different.
Instead of having a guest that I
interview, I have an invited Erin on to
be kind of like my co-host to help me
just talk through and unpack all of the
AI craziness from the past few months.
So, welcome, Erin.
Aaron Makelky: Thank you,
Lindy, for having me.
I'm flattered.
Lindy Hockenbary: What do
listeners need to know about you?
Aaron Makelky: Yeah, well I've been a
career educator, , in rural states as
recently as the 2025 school year, but I've
since pivoted to developer relations and
community marketing at a tech company.
And I always tell people my skillset,
I'm not the most techie person.
I'm not a developer or coder,
but I'm the conduit between those
really techie developer types.
And the average person, whether it's
a teacher, an administrator, or small
business owner, I can speak the language
of both, and I'm somewhere in the middle.
Lindy Hockenbary: Yep, same.
Same with me.
It's just I am very still much in.
Education, career educators
still in education.
You made the hop full on to work for
a tech company, but that's one of the
reasons I wanted to bring you on today.
First of all, you're one of the people
that I keep up with most on social media
'cause you're always sharing really
awesome tips and how you're using ai.
And when you were in the classroom
still, you were sharing so many great
things about how you were embracing the
technology and teaching your students
how to use it correctly and responsibly.
I think you were my.
Most followed person on LinkedIn in
2025 because you know, they did that
little thing at the end of the year
where they told you who you like,
who you like most engaged with.
I'm pretty sure you were , my most
engaged with person, so I wanted to
bring you on because you are threefold.
You're doing a lot with the tech
yourself and keeping up with it yourself.
You're also working for a tech
company now, and it's based
out of the Bay Area, right?
Aaron Makelky: It is.
Yes.
Lindy Hockenbary: Yeah.
So like you are, even though you're
working remote from Wyoming, you're
embedded in virtual calls all day long
with those that are like, you know,
the tech people in Silicon Valley.
So you kind of have that perspective now.
And then also you have
the educator perspective.
'cause how many years did you teach?
Aaron Makelky: I was a
teacher for six 16 years.
Uh, and I, I have my master's
in K 12 ed leadership and a
principal endorsement, but always
preferred to be in the classroom.
, Despite some of my family, my.
Members that were career
administrators, I thought the
classroom was a better fit for me.
Lindy Hockenbary: Your dad
was an administrator, right?
Aaron Makelky: Yep.
And my identical principal.
Yep.
Lindy Hockenbary: right.
Aaron Makelky: And my identical twin.
Mm-hmm.
Lindy Hockenbary: him on
LinkedIn too, and every now and
again I'm like, oh, it's Aaron.
No, it's not Aaron.
No, it's not.
Aaron Makelky: Same DNA different person,
Lindy Hockenbary: person.
I love it.
I still have to meet
him in person sometime.
E is it Eric?
Aaron Makelky: Eric.
Yep.
Lindy Hockenbary: Eric?
Eric and Aaron.
Yeah, that's right.
Okay.
And he, yeah, he's educator
too slash administrator.
So, yeah.
So point being, you bring all three
of those different perspectives in
to give listeners a really good idea
of Where, where we're at with ai.
So I wanna start with a story.
There's something about TCA conference.
It's always like the first week
of February, and for the last.
Two years.
All of the big AI companies have released
like a big update like either right
before right around February 1st, which
E either right before or right during
TCA last year, open AI released operator,
which if you're not familiar, they've
now rolled it into their agent feature,
so it's no longer called operator.
But at the time it was
basically like their agentic.
And if you don't know what agent is, we're
gonna dive into that, where you could
actually tell it to go research a vacation
that you wanted to take or draft an email.
I had it in my Gmail,
drafting my emails for me.
Uh, and it would open up a browser
within the window and all these things.
So I was frantically scrambling
at TCA next year, last year,
excuse me, 2025 to make sure that.
I was making sure every one of my
sessions, the people that came,
knew what a agentic AI was and
very, very, very few people did.
Okay.
Fast forward a year.
I was just at TCAA couple weeks ago.
First week of February, same thing.
Huge announcements coming in.
So OpenAI release codex, which we're
gonna hopefully talk about your
anthropic release, their Opus 4.6
model, which holy cow, I was actually
texting my husband this morning.
Telling him that he needs to go
check it out because now you can
literally give it a Word document
and it will edit the actual Word
document and spit it back out to you.
So before, I would always
like upload a document.
I'd ask for changes.
It might give it to me.
It might have given you a file that you
could download, but you still had to do
a lot of like copy pasting, formatting.
Nope, it does all the
formatting for you now.
So that was released and then on
top of all of that, there was.
All of this craziness going on
with Claude Bot, open Claw and mbo.
And if you're not familiar, we're
gonna dive into that as well.
So that was all going on in the midst
of TCAI was like sharing TikTok videos
just to kind of stay on top of it and
inform my TikTok subscribers about it.
And then so all of that
was happening, right?
I am, I actually had time to go to
a few sessions at TCA this year.
I'm sitting in a session agentic browsers
came up and the whole room, and it was a
packed room and it was a pretty big room.
Came to a halt with mass.
Chaos for either a combination of, I
think the majority of the chaos was they
had never heard the word ag agentic.
They had no idea what an
age agentic browser meant.
, The other little part of
the chaos was immediate.
They knew it, but they didn't know
the feature that was being shown.
And they immediately started
talking about how to block it.
And the presenter did a really great
job of making sure that it didn't
totally change the course of the
session, but I was just sitting there.
going, oh my gosh, these like,
we have some major work to do.
We've been over the last year, agent
a KI has been the big push behind
a lot of the changes in updates.
Would you agree or not agree, Erin?
Aaron Makelky: Totally agree.
I think a lot of us about two
years ago saw copy, paste, chat.
GPT is in a tab or cloud,
whatever you're using, and we
thought, wow, AI is so awesome.
And the new version of
it is it can do things.
On your behalf or for you?
I'll share how I previewed
agen browsers with my students.
I gave them a Google Doc in a
high school social studies class
that had a list of eight or nine
questions about a historical topic.
They didn't have a textbook or a
reading on, and they were supposed
to go out and do some research,
and it's like a background
builder on the League of Nations.
When was it founded?
Which country came up with the idea?
And I said, I just wanna watch how you do
this to see, do you go to Google search?
Do you go to chat?
GPT?
What tools do you use?
But it was a trick because I knew most
of them opened a Google tab and they
copy paste the question to Google search.
A couple of them used ai.
I let 'em finish.
And then what I did is I took Comet
browser, I opened the same Google
Doc and I said, go find the answers
and fill it in on my Google Doc.
I hit run and then to be funny, I took
my Comet browser water bottle and I
went down the hall to get a drink.
Then I just watched my classroom through
the window, so I was still kind of
supervising him and they're like, it's
opening tabs, it's researching, and then
it's going back to the doc and it's typing
an like, is he remote controlling this?
What the heck?
And then I would just wave
at him through the window.
And the reason I did that was 'cause.
It was gonna show up eventually.
Now it's in Google Chrome.
Claude has a Chrome extension.
So if you're a Claude person, you
get essentially the same thing
and it can control your browser.
And that was my moment of chaos where
people thought it was a trick or it
wasn't real, and it's just like, nope.
That's what they can do now.
'cause they're built into your browser.
Lindy Hockenbary: Yep.
And yeah, if you've, if you've never
seen an agentic browser or any sort
of AI agent or agentic ai, it really
does at first seem like a magic
trick, like someone is pranking you.
My students used to prank me all
the time and like set sticky keys.
On the keyboards and the
computers in my classroom.
And then, you know, somebody would it
either I caught onto it pretty quickly,
but then like somebody else would come
in and start using the computer and
think that the computer is possessed.
You know, 'cause they couldn't
figure out the sticky keys.
It's kind of like that experience, but
like way more intense than sticky keys.
And if you're not familiar, so
a agentic AI is basically ai.
That acts, and I always tell people when
I say this, this creates a lot of anxiety.
So I'm always like, take a deep breath.
Seriously, take a deep breath.
It acts autonomously.
So exactly what Aaron was saying,
you're not now just like going to chat
BT or Claude or Gemini and giving it
a prompt and copy pasting over into
a Google Doc, like it's editing and.
Adding to the Google Doc for you, it's
autonomous, but you gave the perfect
example of, I always add the caveat
that it's autonomous with oversight
because you gave it the Google Doc,
you gave it the questions in the Google
Doc that you wanted to do, so you
gave it instructions as the human to
say, this is what I want you to do.
I want you to answer these questions about
the League of Nations in this Google Doc.
Aaron Makelky: Yep.
And the way that I think of it,
the super simple, not as nuanced
definition is agent AI has tools.
So it's kinda like if you gave chat.
GPTA toolkit and it has a wrench for
this job and a hammer for this job only.
It might be a browser and it can do
a search, and then it can call this
service over here and then it can
write to your word or your Google
doc, and it's giving the same brain.
That's the ai.
The tools and now it's an agent.
So it could be a recurring schedule
or a trigger like when I get an email
from this person, do this, or Monday
mornings at a 8:00 AM run this, , or
like the browser example, you open
up the tabs, it needs for context
and say, go off and do this for me..
Lindy Hockenbary: Yep.
Exactly.
And I always get the question at this
point, they're like, okay, well what
they still don't, if you're new to Agent
ai, you may not still see the difference
between a generative AI tool like chat,
GBT, Jim and I, Claude An agentic AI
tool or an AI agent, which there are
subtle differences between the two,
which we probably won't get into today.
, , the way I explain it is , let's use email
for an example chat, GPT, generative ai.
You have to go open the large
language model, or you might
have it built in to like Gmail.
Now there's some generative
built in there, right?
But you still have to prompt it to say,
write this email, review it, and hit send.
Agentic ai, you tell it.
I want you to respond to my emails.
Tell everyone that I'm gonna be out next
week at the NCC Conference in Seattle.
And it goes, opens the email,
drafts that response and hits send.
That's agentic ai.
That's that autonomous part to it.
Aaron Makelky: Which is a scary thing
'cause people always go, what if it sends
the wrong message to the wrong person?
Or books the wrong flight for
me, or whatever task you gave it.
Lindy Hockenbary: It is scary and you
know, it's, there's a lot of discussion.
We'll get into this a little bit
more when we talk about the claw
bot, mot bot, whatever they're
calling it now thing, but it's also.
Uh, it opens the question about
security risk and data privacy, right?
Like now you're giving, you, you,
we've already been talking for years
three over three years since chat
GPD was released about not putting
an education like PII, making sure
you're a FERPA compliant, right?
Don't put PII into a
large language model now.
It's like 20 steps more than that of not
only just don't give it your information,
but now you're literally giving the , AI
access to things like your email that
might have really sensitive information
in it, depending upon what your job is.
But if you're a teacher or a
school administrator, your email,
your, if you use Google Drive
or OneDrive, you're giving it.
Access to that potentially
has a lot of sensitive info.
So that's a huge discussion around agent
to AI and education is that data privacy.
And that's probably
one of the reasons why.
All those educators, not all the
majority of the educators in that
session room at TCA, just had no idea.
Like I even heard someone
say, I'm so confused.
She was looking around like this
because she was like, she had
never even heard the word ag.
And then you put Ag Agentic browser
in there and you know, a whole
nother level of confusion because
it's no secret that K 12 education
has a lot of barriers to innovation.
One of those being ferpa, data privacy.
We're working with miners under the age
of 18, so the tech tends to move slower.
And not just that, but other reasons as
well than it does in other industries in.
Aaron Makelky: Yep.
And I think if you haven't come to
this realization yet, it's pretty
unanimous by people who see this stuff
every day in 2026 if you're relying
on anything in a web browser or on
a computer to assess your students'
learning or track their progress.
I'm not saying they're all using Chrome.
Having an agent do it for them.
I'm saying they very well
could be even with free tools.
So if it's something in
a Google Doc, I'll share.
Here's an example.
One way people said to monitor for AI
doing the work of students is go back
to the revision history of a Google Doc
and see if they copied and pasted in
a giant four page essay all in one go.
Like that was ai.
That was cheating.
Uh, that's not my preferred way.
If that's what you do, okay,
here's the counter argument.
And now a browser agent could be doing
that for them, and it's not gonna just
copy and paste one chunk from the tab.
They did the prompt in.
It's actually gonna be in
Canvas or in Schoology or in
Google classroom, navigating.
It's gonna show the person's logged in.
They made this edit at this
time, and then they opened here.
And I've even had students show me,
you can prompt it to write out the
thing, slowly put in a couple of
typos, then go back and edit them.
And all of this is running without
the kid touching the keyboard.
But the revision history
is gonna look organic.
It's gonna look like a person did it.
Lindy Hockenbary: That came up in
that session that I was at, at TCA
is like, and even like Bri teaching,
they have an inspect writing tool.
Built into that to show if any big
blocks were copy pasted or whatever.
But now these agents, when I gave
operator a year ago access to my email,
I sat there and watched it type out the
email and look exactly what it would
look like if I was typing out the email.
Maybe the only difference, which
would be a subtle difference
that you'd have to really watch
is the typing speed and typos.
The AI isn't gonna have typos
and be using the backspace like a
human would when you're naturally
just typing something out.
And the educators in that session,
some of them piped up and said, well,
I was already having that problem with
students, figuring out that teachers
are using these tools to see if I
was copying and pasting a paragraph.
So what do they do?
They write their essay in a
large language model, do a split
screen, and then they type it.
They're literally just like I think of
back in the day, you see all of like the
movies where they made the kids write
lines and that was their punishment.
So guess what?
Now?
Now the kid is literally,
that assignment a
Aaron Makelky: It's,
Lindy Hockenbary: punishment.
Aaron Makelky: it's like
typing practice to the kid.
They didn't come up with the, the
thinking or the ideating or the editing.
It's just like, I'll copy it manually.
Yeah.
And there's.
There's chrome extensions.
I saw a student make one that you
paste in your essay, so probably chat.
GPT generated it.
The chrome extension is open in your
browser and it's gonna type it at a
human speed and you can dictate that
and you just paste in your whole
essay and it's gonna go into your
Google Doc and character by character.
Make it look on the revision
history like a human did it.
So I doubt many students use
that, but they're free and
they're on the internet, so.
The crafty ones are trying to find
ways, and you shouldn't assume that's
the best way to filter for cheating.
Lindy Hockenbary: If you're listening
to this and you're like, oh my
gosh, I need to go try this myself.
I need to know what's going on or what
tools I should use or what tools students
are using to do this kind of stuff.
I wanna give like a layered approach here.
The first one , would
be the agents within.
Chat, GBT.
Claude.
I'm trying to think.
Gemini doesn't really have
agents yet, I don't think.
No.
Aaron Makelky: Nope.
Just built into the browser now.
Lindy Hockenbary: Yeah.
Um, I'm gonna open in another tab,
but, um, yeah, built in the browser,
which I'll get to in a second.
Okay.
So those have been there for a
while, like over a year, I would say
maybe a little over a year for chat.
GPT.
So that's number one.
If you go into chat, GPT, hit
the tools button, turn on agents,
and people are always like,
well, I don't know what to do.
If you don't know what to
do to start, guess what?
It's an ai so you ask it.
What can you do for me?
How
Aaron Makelky: Yeah.
Lindy Hockenbary: me?
Right?
And if you have your in chat, GPT
especially, where if you have the setting
turned on where it learns you and it
learns information from your different
chats, if you use chat GPTA lot.
It's gonna know you're a high
school social studies teacher.
It knows everything about my work,
so I have to give it no context.
Now when I go in there, so if
I go in there and I'm like,
Hey, what can I do with you?
It'll start giving me
very customized examples.
And then from there you could
even say, give me a prompt,
copy, paste, put the prompt in.
And just, that's the
way I start exploring.
These types of tools now.
So there's that.
Then the next level would be, we mentioned
it, the Chrome browser, which is by far
the most common browser, and I would say
by far, the most common browser in K 12
education now has Gemini built into it.
So if you haven't noticed over the last
few months in the top right corner,
you now have a Gemini button that
you can click if you don't see it.
could be if you're logged into your
school account, it's either that your
school has blocked it in some way, or
schools, depending upon their settings
with Google Workspace, can do either a
rapid release or a scheduled release.
So you might be on a scheduled release
where it hasn't been, you don't get
the new features for a while until
after they're, they're out there.
That's the one that,
honestly, so I mentioned.
The concept of in that TC, a session
of agentic browsers came up, but
then the presenter showed an example
using the Gemini within Chrome.
This is not a full agentic browser yet.
And Aaron, you said it perfectly,
it's actually not the most effective.
There's, when we get into like the
comment and the atlas, those are way more
effective than Gemini built within Chrome.
Although I will say I've been using
the Gemini built within Chrome a lot.
The example that the presenter at
TCA did, he had a, a question up
or it was like a writing prompt or
something and he hit the button.
He just acted like a high school kid.
Like, Hey dude, uh, I'm working
on this problem and I can't
figure out how, how to solve it.
Can you solve it for me?
Solved copy pasted right into the Google
Doc that's when the conversation got
waylaid to oh, how do we block this
Aaron Makelky: that's always
the first question when it's
educators, unfortunately.
Um, well, and we all know this, what
that presenter did, you could have done
by switching tabs and copy and pasting.
It's just the next evolution
of the same technology now.
It's built in and you can give
it a tab group for context.
That's what I usually do is like
Lindy Hockenbary: Mm.
Aaron Makelky: maybe a little research, a
little docket goes in and a another page
and say, look across these three tabs,
do this task, write it out for me, and
now it can use those instead of copying
everything off of those pages, which you
could have done two or three years ago now
it just does it in the browser for you.
Lindy Hockenbary: That's where I need
to switch my workflow in my brain.
'cause for three years I, let's say
I have a Google Doc open and a couple
websites, and I wanna give the large
language model the context of all of that.
Actually, I just did this this morning
of whatever it is that I'm doing.
I would download the Google Doc
and then I'd copy the links to the
websites and put those into chat.
GBT.
It still saved me a ton of time,
but it's that download, upload,
copy, paste for each of those.
Now I just hit the Gemini button.
already have the tabs open so I don't have
to give it any context to what information
I want it to pull in and summarize, or
synthesize or whatever I want it to do.
It's just right there, so it saves you
that many more seconds from a productivity
standpoint, from a student standpoint,
it's that many more seconds saved and
less steps to provide answers to , what we
used to call Googleable questions, right?
Aaron Makelky: And the other thing
is remember that your students are
gonna see all the ways to cheat.
They're not gonna see on their
TikTok or their YouTube shorts,
all the cool things that can help
you with researching and writing
and keeping track of your schedule.
. If all they see is the cheating way,
that's one of the reasons why I am a big
fan of educators themselves have to share
these technologies with their students
instead of hoping the student doesn't
know 'cause they're gonna find out or
trying to block it, because a lot of us
don't have the option to block it either.
Your administration decides what goes into
your Google apps for red account and they
might listen to you, but probably not.
Or a committee takes
three months to decide.
So in the meantime you're stuck with
it and you just gotta make it work.
Lindy Hockenbary: Figure it out.
Yep, for sure.
And then so the last level of agentic
browsers or, or well, or actually
agentic browsers out there is, you
mentioned perplexity has comment, chat.
GPT has atlas.
How are those different than the
Gemini built within Google Chrome?
Aaron Makelky: Yeah, so what Google
and Claude did was basically say
instead of rebuild the browser, we'll
put an extension in Google Chrome.
And let that control it.
In a lot of ways, it was cheaper
than, you know, hiring the people
to develop a whole new browser.
What perplexity did is they forked
Chrome 'cause they basically
copied the underlying architecture.
The chrome extensions from your
chrome browsers still work in
perplexity comment, but they
rebuilt it in a way that the agent.
Is foundational, it's baked in.
Instead of add a Chrome extension,
it comes with a whole sidebar and
it has a couple different interfaces
and it's still using large language
models, but it's way more prone
to taking action on your behalf.
And I've had access to it since before
it came out, just in the last two weeks.
So like while you were at
FETC in early February.
With the newest models like Opus
4.6 Running Comet browser, I tested
it and it did a task autonomously
for about 30 minutes, and it
took between 175 and 190 actions.
So think of that like scrolling, clicking.
Typing a text field, basically how many
movements you would take on a browser.
It's, it's definitely not faster than
if you were controlling your browser.
I'd say it's half the speed, if not a
little slower, but here's the power of it.
You set the task in a tab
group and say, go work on this,
and it runs in those tabs.
Meanwhile, you're doing something
else, so it's not like, oh, it's
faster than air and clicking and
typing and searching and scrolling.
It's Aaron's doing that on the page.
He wants to, and then in the background
for 20 to 30 minutes, it's gonna run on
its own and then say, okay, it's done.
Or, okay, I hit a snag.
Atlas is the same way.
The difference is it's gonna
use your chat GPT account.
So if you're somebody who's really
invested in that platform, you have
the projects and the context history.
Some people like it.
I personally was not that invested in it,
so it didn't seem super useful for me.
Like I wouldn't use that if
you're not a chat GPT Power user.
If somebody wants another one, DIA
browser, DIA is another one that's
been competing with Perplexity Comet.
It actually came out before that
and it has some built in Agentic
capabilities and you can switch models.
If anyone used Arc browser, it's the
same company that built Arc, that's
probably my favorite browser of all time.
They just made an AI native
version of it called dia.
There's a free version you can use,
but the thing is, and, and one of
the barriers to a lot of student uses
the good models and the best versions
of it, you need a subscription.
So whereas the Gemini students can get a
year of the pro plan for free, so a lot
of them have access to the paid features.
Your district might enable 'em through
your Google apps for red account,
or they might have a, a student.
AI Pro Plan.
I believe it's the $20 a month one that
they get for free if they're a student.
So there, there's definitely more
students have exposure to the Chrome
extension than the standalone browser.
. Lindy Hockenbary: I'm writing
these all down by the way, and
I'll put links into the show notes.
I have not heard of Diaz,
so I gotta check that out.
And I'm with you.
I'm a pretty big power chat
BT user, but I also use a lot.
I use, I'm a big Claude user.
I'm using Gemini now.
Gemini three.
I'm using Gemini.
Way more than I ever have before.
And I still was kind of like me the
chat, GPT Atlas, , and I think that's why
then I never got , back into the agentic
browsers because I was like, I don't,
I don't know, I'm not seeing it yet.
So I need to get into comment
some of these others and,
and play around with them.
Okay, so let's, we better move on.
So many other things to talk about.
'cause we haven't even
talked about Codex yet.
So one of the big.
releases that opening, I did that
first week of February was Codex Aaron,
I've played with this a little bit.
I know you've played with it a lot
more and your new job sounds like
you've played with it a lot more.
So tell us about Codex.
Aaron Makelky: Full disclosure, I did
spend an entire weekend living in it.
Um, not 'cause,
Lindy Hockenbary: Okay.
Aaron Makelky: 'cause Lindy,
Lindy Hockenbary: more deep than I am.
Aaron Makelky: I. Well, I think
we all have test projects, right?
So say for you when a New Claw
model comes out, you're like,
here's a go-to thing that I have a
baseline of what good looks like.
Codex is interesting because I think a
lot of the vibe coding and coding apps
have not been super relevant to teachers.
'cause they're like, what
would I do with that?
Like, I'm not a coder, I'm an English
teacher, I'm a history teacher.
What Codex can do is it's basically chat
GPTs model, but specifically for coding,
and you don't have to open a terminal, so
you don't have to go to the scary blinking
black cursor box that feels like 1980s.
That's intimidating to non coders like.
Lindy Hockenbary: like a
terminal dos looking thing.
Aaron Makelky: Yep, because cloud
code you have to run in a terminal.
I know there's a web version now,
but originally, so basically chat,
GPT tried to build the same thing,
but it's a standalone app on the Mac.
It's now also on Windows, and
it runs in the web, so it has
access to your local files.
This is one thing that's different.
If you have chat GPT in your browser tab.
It can't see your downloads,
it can't see your local files.
Codex can, and people think, well, I'm
not gonna build an app or a website.
Even if you can't, you know,
that's not your use case.
You could take a folder.
Uh, here's one use case.
I do some analytics reporting.
So download the CSV files or
the spreadsheets, put 'em in a
local folder, then open Codex
and say, go in that folder.
And help me analyze all of this data
versus before you could probably attach
files and do all sorts of importing
and exporting in the web now can do
that in a folder on your computer.
Um, and then like one of the ways
I used to test it, I think a lot
of us have vibe coded in a webpage
or you make one for yourself.
My, I make my kids do that
for themselves and I took a.
Lindy Hockenbary: kids vibe coding, huh?
I love it.
Aaron Makelky: Maya, I
think he was 10 at the time.
10 or 11.
My fifth grade son built his own math
review game without touching a key.
He just voice dictated into an app
and it worked, um, which I think
everyone should be doing, but.
Digressing back to Codex.
I took a website that took me weeks
and a bunch of revisions, and I was
like, Hey, let's redo this, make it
look different, change the content
and move it to a different domain.
And I did something that the previous
best way I knew how to do took me,
I don't know, 10 or 12 hours, and
this was about a 30 minute process.
And it totally revised the
look, the design, the colors.
And then I thought, okay, this was
really good at the designing part.
What's a, what's a design thing that
a non-technical person would use?
So I actually had it design.
My LinkedIn banner and some graphics.
'cause people are like, oh, it's a
coding model, but I'm not a coder.
Well, part of, I mean,
design is fundamentally code.
They're all just pixels
and ones and zeros.
It compared to what I
could do in Canva or Figma.
It made a way better
LinkedIn banner in about.
A a minute, and here's the best part
to my local computer, I'd say put it
in this folder and call it designs.
That's not something that you could
do with chat GPT or clot on the web.
'cause you always have to
export the artifact and download
the PDF or whatever it's in.
I could just point it to my local.
Lindy Hockenbary: tell it to save in
like a certain fuller on your computer.
Aaron Makelky: Yep.
Lindy Hockenbary: Wow.
Interesting.
Aaron Makelky: And then you
can, you can back it up.
So a lot of people use GitHub
and if you're intimidated,
I was once so intimidated by
GitHub and I'm not an expert.
Think of GitHub, like
Google Drive for tech.
People like, and you
can put documents in it.
You don't have to build apps and websites.
But it's just a way to back
things up and then also share it.
Just like a lot of educators
go, oh, that form, that
spreadsheet is in my Google Drive.
Here's a link.
That's essentially what GitHub does.
You can say, take the project we put on
my local machine, like those LinkedIn
banners and this whole website, and sync
it to my GitHub, and it's just like, boom.
Running this command done.
It's committed like it's backed up there.
So I do think there are a lot of
uses for teachers, even if you're
not computer science and building
apps, that that is what it's best at.
But it can do all sorts of other things
for you as well, and it can do it on
your local machine from a web app.
You don't need to be in a terminal.
Lindy Hockenbary: okay, so I'm hearing
the big changes and differences with
Codex is that download from an app, so
it's more user friendly than Claude Code,
which was kind of the leader in AI coding
basically up until February 1st ish.
And then it can also.
Actually save files in
specific folders for you.
And then the thing I noticed in the
little bit of time I've spent in
there, I went in and just so you
guys know, like I am not a coder.
I am not that kind of technical person.
So I went in and I could have asked
it, Hey, what can you build for me?
But when you go in there, it actually
has some sample prompts for you.
So I just clicked the first one.
It was something like build
a sneak game, something super
easy and just watched it run.
The thing that I was like, oh,
this is a game changer, is it
not only built the game, it built
the app, but then it tested.
The app and if you have never worked,
like Aaron's, now in the tech world, I
work with a lot of ed tech companies,
so I get immersed in that world.
There is a job at every, at least
one person, at every company
that builds software called a qa.
They're like the QA engineer.
And QA stands for quality assurance.
And what that person does is test
and make sure their software works.
Work looks for bugs, right?
Codex is doing that.
And I'm not being the hype gal
that's saying that Codex is gonna
take every QA engineer's job.
I'm not saying that, but I want you
to think about the job disruption that
could be caused by the AI and oh, and
here's another thing about Codex, which
I have not fact checked, but I heard
that OpenAI actually had AI create codex.
Have you heard that?
Do you know if that's true?
Aaron Makelky: Yes.
, They publicly said they used
their newest model to build it.
So basically the tool they were building
helped them build the tool, which is
one of those weird advantages in the ai.
World.
Yeah, and if you've played with other
ones, here's what I'd say about Codex.
It's very persistent.
So I've used Claude to do a task.
Claude can feel , it's more warm, it's
more enthusiastic, it's more human, but
it also can feel kind of a DHD and kind of
jump around and quickly do this, and it's
not as good as, look, it's not as good
at looking through things in, in detail.
I did a project with Codex just last
night and multiple times the little
working bar ran for more than 20 minutes
before it actually output something.
And I've never seen a clawed
model be that persistent.
'cause it'll either take a shortcut,
hallucinate, make something
up, give you like A-T-L-D-R.
Codex definitely seems more thorough and.
Kind of gruff, like, okay, here's the
thing, it's done, and it's not as warm
of a personality as Claude models.
Lindy Hockenbary: Which makes sense.
Just knowing the difference between
Anthropic and OpenAI, the companies that
make both of those, and just knowing the
vibe difference between Claude and Chat
GBT, that kind of makes sense , that their
coding models , would do the same thing.
Okay, so, so that's a biggie.
There's a lot to unpack.
There.
And I know at the surface you might
meant, but like really stop and
think about the fact that we now
have AI, building AI that is huge.
And we have already been moving
so quickly with AI advancements.
Now that AI is building ai, you
know how much quicker we're gonna
be moving with AI advancements.
AI thinks, and works way, faster, way,
way, way, and way is not even the right.
Word there.
A lot faster, lot faster than humans.
Aaron Makelky: And it doesn't sleep,
it doesn't take vacations or PTO days.
Lindy Hockenbary: It doesn't
have bio break needs.
Aaron Makelky: can spawn subagents
to make work, you know, delegate
jobs on, like you and I can't spawn,
, executive assistants to go through
our emails or do things for us.
Lindy Hockenbary: One other big update and
then if we have time, we'll dig a little
bit more into the effects of education.
We might have to do a part two of this
episode just focused on education.
Should needs to change, not
change, adapt, not adapt.
But I wanna make sure we talk about the
Claude bot slash open claw slash mt bot.
Fun.
, So again, I was watching this
go down while I was at tca.
And by the way, and granted I only had
a chance to go to a few sessions, I had
tons of conversations with educators and I
presented five or six different sessions.
And not once did I have
somebody mention this to me.
At the education conference, which
I thought was super interesting.
, And just like maybe a difference
in people's algorithms.
'cause it was all over all of
my social media algorithms.
That's all I could see.
So A guy created what?
And he, he called it Claude Bot.
To start with.
Well then, and it was C-L-A-W-D-B-O-T,
then people thought it, it was related
to Claude, C-L-A-U-D-E, which is the
large language model made by Anthropic.
It was not, not associated at all.
So then he quickly changed the
name to Open Claw and basically,
well, here you explain.
Aaron, if, if you're comfortable, explain
what Open Claw slash Claude bought is.
Aaron Makelky: Well, and if teachers
hadn't heard by then, the backstory
is it's changed names three
times, but you're exactly right.
Lindy Hockenbary: was
there a third one in there?
What was
Aaron Makelky: Yeah.
Well, and that's what's funny 'cause
I know you wanna bring this up.
Mt Bought was a very quick change and what
the founder of this open source project
said was Anthropic didn't like that.
It sounded like Claude, even though it
was spelled differently, I'm guessing
somebody sent a letter from a lawyer
Lindy Hockenbary: Yeah.
Aaron Makelky: is like, okay, I'll
change it to Molt Bot 'cause it,
it molts and changes and evolves.
And that's where Molt book came from.
Lindy Hockenbary: Oh,
Aaron Makelky: Is semi satirical.
Someone said, let's make a
website for all these new agents.
That's like social media, you
know, LinkedIn, essentially for
the agents to post, not the humans.
Then the founder of the project
changed it, a third different name
to Open Claw, which it got acquired.
He got acquired by Open AI this weekend.
So now.
Yeah.
He, he and the founder, uh, Sam
Altman said it'll stay open source.
It's not gonna go away, but he's joining
our team, whatever that's gonna look like.
But you know, this, just like Atlas and
Codex, it's probably gonna be a standalone
product in chat GPT any day now.
Lindy Hockenbary: Yeah.
Yeah.
Okay.
So I missed a whole, I was thinking
it was Molt bought not Molt book.
I missed a whole name in there.
All right.
But Moral of the story is
it's landed on open claw
Aaron Makelky: Yep.
Lindy Hockenbary: basically only
really technical people are doing
it because you have to kind of
create and run your own terminal.
The Best Buy in San Francisco.
Ran out of Mac Minis because people
were going out and buying Mac
Minis to be, 'cause you really need
like a dedicated computer because
it has to have 24 7 uptime and
basically it's running agent for you.
Or is it agents, is it multiple
agents or just one agent
that's like just working 24 7.
Aaron Makelky: Either one.
It's up to you.
Lindy Hockenbary: one.
Okay.
So you could have multiple agents
potentially doing different things,
but you have to give it access,
like we were talking about, the
security, you know, data privacy risk
there, you have to give it access to
everything, but then it can literally.
pretty much anything digital for you.
Like save this file, send this email,
, put this event on my calendar for me.
And you have to use, like a lot of people
were using WhatsApp send the instructions
to the agents to say, you know, go put
this on my calendar, delete this event
off my calendar, et cetera, et cetera.
Aaron Makelky: Yeah.
Here's how I view Open Claw.
We'll stick with that name and hope
he doesn't change it a fourth time.
It doesn't do anything new.
You could have done all of this stuff
with Claude Code, or if you were a
techie person, you could have duct
taped your own version of this together.
But there was no hope of an
average person making that work.
And now what's unique about
Open Claw is just the interface.
In other words.
If you like WhatsApp, you
have group chats there.
You're in the app.
You can use it to talk to your
agent, but then it can go out
and use all these other tools.
It can search the internet, it
can search specific websites.
It can see files or folders you have
saved in the cloud somewhere, like your
Google Drive, your notion, your GitHub,
whatever, and it can do things with that.
It's also misleading.
I tell people all the
time, don't buy a Mac Mini.
, I have one under my desk.
Mine is on a raspberry pie.
Number one, a raspberry pie is
way cheaper than a Mac mini.
Also easier to find in stock right now,
but your local Facebook marketplace
is probably gonna have a bunch of used
very gently used Mac Minis any day.
Now, as people realize the, the
tool is free, but the services you
have to plug into it to run are not.
So I think of it like a house.
He basically built the
app and said, here you go.
It's free.
You can download.
But to make it work, you have to plug
it into things that cost money to run.
The biggest one is an AI tool.
Like it needs a A brain, it
needs chat, GBT, Claude Gemini.
You can use anyone.
You can get an API key
for those cost money.
So his app without those connections
is like having an empty house that's
not connected to the power grid.
Connected to the sewer.
It doesn't have electricity.
It doesn't have plumbing.
Well, that's a pretty worthless house.
It looks cool, especially that
it's free, but it's really about
how you connect it to things.
And what I would do and what I am
doing is put it on a different device.
It just doesn't have to be powerful.
You don't need a fancy pc.
Lindy Hockenbary: Mac device.
Aaron Makelky: No.
And if you do, 'cause it's an old
one, just know for anything to
work, it has to be on, has to be
powered on and on the internet.
So if it's a laptop, you're gonna have
to change the battery settings and the
sleep settings to never go to sleep.
, But something like a
raspberry pie can run it.
I think somewhere between one to two
gigs of memory is what people find.
It's good enough.
So like any old laptop from the
last 10 years could run this.
Any old PC you don't
still use could run this.
But the other thing is you want it on
a separate device so it doesn't have
access to all of your permissions, your
browser, your folders, like as a teacher,
definitely not on your work computer.
Lindy Hockenbary: No,
Aaron Makelky: Um.
Lindy Hockenbary: of tech directors
that would be up my throat.
Yeah.
Do you or your work.
Emails, accounts at
Aaron Makelky: Yes.
The, the way to think of it is you wanna
build it in a box and keep it contained.
Like you don't want to give it that email.
So what's interesting, some people
give their open call agent their
own accounts to things like their
own Gmail account, their own iCloud.
, One guy now has.
Two Dropbox subscriptions, one
for him and one for his agents.
And some things are shared between those.
But his important, like PII
containing files or finances
are in his personal Dropbox that
his open claw agents can't see.
So.
Really, the moral of the story is you
wanna start with it on a different device
and don't give it access to anything.
You wouldn't trust a human,
like a say a student teacher.
You would not give a student
teacher your bank account password
and your personal laptop 24 7.
So you definitely don't
wanna give it to an agent.
Lindy Hockenbary: Really good advice.
And that's the big, the things that I
have heard that's like the big differences
with open claw is that now it's allowed
agents to go externally is one part of it.
Like you said, like you could do a
lot of the things that you could do.
With open Claw, with like the
chat GPT agents over the last
year, but now it can go external.
And then that leads to the second
thing, which is now the agents can
talk to each other, talk in quotes.
Right.
And so that's where the malt book
I had been calling it, malt Bot,
the Malt book came into play,
which is basically like a Reddit.
For AI agents and they've gone crazy
and they're chatting and they've
like made up a religion and they're
talking about religions and they're,
and it's freaking people out.
, This is a big, we've gotta wrap
up here and I know it was a big,
heavy thing to leave the episode on.
, But I think it just goes back to that
foundational AI literacy that when you
understand that the AI is just acting
with instructions, you understand that.
The agents were given instructions
to do that, and they're basically
replicating Reddit thread because
that's what they have to go on.
So they're just kind of like
recreating that in mt book in
their own little agent land.
Aaron Makelky: Yep.
And, and I would also remind people
that there's no way to verify all
the things on there are actually
coming from an agent at that.
Any person could say, go on
and write all these things.
Go make up your own.
Religion and the agent is gonna
do what it's been prompted to do.
So I don't know how much of that
is authentically happening, like an
agent spontaneously went and posted
versus somebody thought it'd be
funny to create these conversations.
Lindy Hockenbary: Yep.
Aaron Makelky: but
Lindy Hockenbary: their agent to go like
say, Hey, let's go start another religion
Aaron Makelky: Yep.
Yep.
And that.
Lindy Hockenbary: a good point.
I.
Aaron Makelky: A, a variation
of that same mt book style site
that's been freaking people out.
We've all seen the Upwork, the
TaskRabbit, the hire a person to, uh,
do this little project in Photoshop
or make this webpage for you.
There have now been some sites
pop up where agents need humans
to do things in the physical
world and they will pay people
Lindy Hockenbary: What
Aaron Makelky: do the tasks.
So it's like an Upwork
for agents to find humans.
Um, and that's, uh, just a whole
new world of like, okay, they, they
can't go in the world and turn a knob
or go to a physical location yet.
Lindy Hockenbary: embodied
Aaron Makelky: Yep, yep.
Lindy Hockenbary: till
they're embodied ai.
We currently have non embodied
ais Wait till they're embodied.
Yeah.
Aaron Makelky: Yeah.
But.
Lindy Hockenbary: I hadn't heard that one.
Aaron Makelky: Yeah, my, my biggest
takeaway with Open Claw is you'll spend
way more time trying to make it work than
you will find useful things for it to do.
That's one.
Lindy Hockenbary: Yep.
Aaron Makelky: other is if there, if
there is something useful, it's making
the technology accessible to people.
So my example is my family.
Calendar my kids'
wrestling meets right now.
They're what days do they have student
council meetings in the morning, I
wanted to build a thing that I could
say, Hey, in a telegram or a WhatsApp
chat, my wife in plain language
can say, what's the schedule today?
, Where's the next wrestling meet?
And something can look across the
context it needs and say, yeah, in
a telegram chat, here's your answer.
That's really the only advantage,
, and it is difficult to set up.
So if you don't know what you're
doing, make sure you follow a
good tutorial or ask somebody
who's done it a couple of times.
'cause it's easy to put API keys in
a place you don't want them or give
it permission to do something that
you're not gonna wanna trust it to do.
Lindy Hockenbary: It's pretty technical.
That's why a very nichey
group are the only ones.
Really doing it, which are
still quite a lot of people.
, But like you just said, it just got
bought by Open ai, so just wait until it's
embedded in chat, GPT and Atlas, their
agent taking browser and yada yada yada.
And that's the takeaway that I want
educators to take from this is.
I know we covered a lot and there's
a lot that we said that is unknown
and potentially scary and really
overwhelming when you think of the
effects on integrity of student work.
I wanted to get more into that,
but alas, there were just too many
updates and things to talk about.
So maybe we'll do a whole
nother episode on that.
Like how, actually multiple episodes to
think about, you know, how do you adapt?
And adjust because some people are saying
, no, we shouldn't adapt and adjust, but
I don't know how we wouldn't, and formal
education just is not gonna remain.
Relevant, in my opinion, if
we don't adapt and adjust.
So there you go.
There's my, my parting thoughts.
Thank you Erin, so much for coming and
chatting with me and providing silicon.
I'm gonna say you're providing the
Silicon Valley perspective, even
though you are in Wyoming, you're
Aaron Makelky: Yeah,
Lindy Hockenbary: that.
Aaron Makelky: and what I know is due
to working with really smart people
that are light years ahead of me and
just distilling little useful bits that
a teacher or an administrator, someone
like you and I would actually put to use.
Lindy Hockenbary: Yep, for sure.
Awesome.
All right.
Thank you so much.
Aaron Makelky: Thank you, Lindy.
Thanks for joining Make EdTech 100.
I know educator time is valuable and I'm
honored you choose to spend yours with me.
For more EdTech strategies you can use
tomorrow and ways to bring me to your
school or event, head to LindyHoc.com.
If this episode resonated, hit subscribe
so you don't miss the next one.
I'm LindyHoc.
Go forth and make EdTech 100.