HomeVideos

This little-known agent beats OpenClaw & Hermes?

Now Playing

This little-known agent beats OpenClaw & Hermes?

Transcript

1316 segments

0:00

My agent just built this website and it

0:02

keeps improving on its own. It's

0:04

bringing me customers like that. It's

0:06

even creating all my social content.

0:07

Like look at this, Andrew. It's It's

0:09

like literally making all the content. I

0:11

got to let go of eight people because it

0:14

is actually doing customer service

0:15

cheaper and about 100x better than the

0:18

humans were. I I mean this whole thing

0:20

is for me is is less doing, which I I I

0:22

think some people think means that it's

0:24

about doing less. It is absolutely not

0:26

about doing less. It's about doing more

0:27

of the things that really matter. And I

0:29

certainly could not do even close to the

0:32

level that I'm doing that at if I didn't

0:33

have this army of AI agents working with

0:36

me. Let's do a screen share and I can

0:37

show you the whole thing.

0:38

>> Presented by Zapier, the AI automation

0:41

company. What is this that I'm looking

0:42

at here?

0:44

>> So I get an email once a week from the

0:47

agent that is also building my websites.

0:50

Uh so the the agents are completely

0:52

responsible for this. This is like an

0:53

end to end cycle. and it sends me this

0:56

email once a week with my key metrics

0:58

from Google Analytics. But the way that

1:00

I always think about this is how can I

1:03

get it to do more, right? What how can I

1:04

get to do more and how can I get it to

1:06

be more proactive? And I think that this

1:08

is a a common misconception with AI

1:10

agents is that you really have to just

1:11

treat them like human employees and in

1:14

that you have to get better at

1:15

delegating. Something I've been teaching

1:16

for almost two decades now. So I was

1:19

getting this report and it's like okay

1:21

well I I guess that's good. I don't

1:23

really know what to do with that. So

1:24

then I started to tell it like tell me

1:26

what we're what are we going to do with

1:28

this? How are we going to improve this?

1:29

>> Because what it's telling you is for

1:30

example it's saying um on the on the

1:33

homepage the average view duration is 49

1:36

seconds. On the slashbooks page people

1:39

are spending 21 seconds. And she said,

1:41

"Okay, this is interesting. Now what do

1:42

I do with it?" That comes up for me all

1:44

the time when I get stats like this.

1:46

>> Okay.

1:46

>> Right. And and I'm also I'm I'm a big

1:48

believer that like everyone who always

1:50

likes to have dashboards and there's so

1:52

many dashboard tools like that's great

1:53

but what are you going to do with that

1:55

information and if you're if you know

1:56

what you're going to do with that

1:57

information then why not just automate

1:58

the doing part.

2:00

>> So it now tells me every week these are

2:03

the things that I recommend you do to

2:05

improve this. Right? So the first one

2:06

here is add in article CTAs to blog

2:09

content. So the blog has a 71% bounce

2:11

rate, but people who stay uh average 70

2:14

seconds, they're reading, but there's

2:15

nowhere to go. So add a CTA mid article.

2:18

Great. Next one. Target direct traffic

2:21

with a personalized welcome. Basically

2:22

saying like create a, you know, specific

2:25

website just for those people and then

2:27

adding social proof uh to the newsletter

2:29

page below the fold. All I have to do is

2:31

write back to that, which I did, and say

2:33

deploy.

2:35

And then it does it. It builds it all.

2:37

And this is probably if I say deploy all

2:39

of this is probably done within about 15

2:40

minutes. Okay.

2:41

>> And then it's been start measuring and

2:42

continuously improve over and over and

2:45

over.

2:46

>> Now Ari, you're using not open claw or

2:49

Hermes agent. You're using tax task lid.

2:52

Can you show me their website? Why are

2:53

you using that instead of the tools that

2:55

I've been talking about?

2:56

>> So I I've used them all. I have not had

3:00

as good of experience. You claude is a

3:02

whole other cla bot is a whole other

3:03

thing. I think that it just it was they

3:06

implemented it and deployed it like too

3:08

quickly and without proper guardrails is

3:10

I think people got all all over excited

3:11

and it just didn't really develop in the

3:13

right way. Hermes is very very powerful

3:16

but I have found it very hard to work

3:18

with. Tasklit I I have been consistently

3:21

impressed with. The only one that's

3:22

similar to it is twin.so which is not

3:26

even a tenth as good in my opinion. So

3:27

the thing that makes Tasklet so

3:29

different and so great is its ability

3:32

and ease to interact or sorry integrate

3:35

with all the tools that you might use.

3:37

>> Okay.

3:37

>> So uh for example when I was trying to

3:40

twin or even with uh Hermes I think to

3:43

get it to connect to Dropbox or to

3:45

Trello was a real pain in the butt and

3:48

it was inconsistent. Whereas with

3:50

tasklet, I'm like connect to Trello and

3:52

it pops up, authorize it and within

3:55

about 45 seconds, it's got a working

3:56

Trello connection and now so it can

3:59

organize and operate my Trello for me.

4:02

So this is this is my personal Trello

4:04

board and this is something like there

4:05

was an issue with our garage door and

4:08

you can see right here it's making

4:09

comments to me with updates about what

4:12

it's handling.

4:13

>> Talk to me about what it did. It it went

4:15

to Rhonda noted no reply. It sent an

4:18

email to Rhonda.

4:20

>> So yeah, so this is basic. Well, so this

4:22

this is what I love about this. So one,

4:23

again, this answers the question of

4:24

Hasslet. It's so easy to integrate with

4:26

things that I've integrated with almost

4:28

everything. Gmail, Outlook, Slack, uh

4:31

Trello, everything. So it has

4:34

unbelievable amounts of context. So in

4:36

this situation, I basically took a photo

4:38

of our garage door, which was showing

4:39

some signs of the the wood was the

4:42

finish was coming off, and it's only

4:43

like a year old. So I take a picture.

4:45

this is our real craft garage door. Look

4:48

what's happening. Follow up with

4:49

support. So then it's able to go into

4:51

Gmail, find the original order and the

4:53

contract, find who we dealt with, the

4:55

date that we ordered it, all that stuff,

4:57

the warranty information, and then it

4:58

can send an email to this person and

5:01

follow up. So that they wrote back, it

5:03

was they replied uh July 9th, and then

5:06

I'm saying follow up 3 days later here

5:08

because I hadn't heard anything. So it

5:10

followed up and then I got an answer.

5:12

And so it can handle that whole thing

5:14

like soup to nuts.

5:15

>> And you're allowing it to control your

5:17

personal email and send email as you.

5:20

>> Yes. So basically whenever I set

5:22

something up, it works in what I I I I

5:24

have a very specific set of rules and

5:27

like a document that every new agent has

5:29

to follow. And part of that is what I

5:31

call shadow mode, which is usually a

5:32

week. And in shadow mode, everything is

5:34

a draft, right? So it brings everything

5:36

up as a draft. I approve it and then it

5:38

sends it. Doesn't matter if it's a Slack

5:40

message, if it's a email, whatever. Once

5:42

I get once I believe that it knows what

5:44

it's doing, then I work off of an 85%

5:47

confidence rating. So, if it's 85%

5:49

confident or better that it knows that

5:51

this is the right thing to do, it's

5:52

allowed to just do it.

5:54

>> Can I see this doc like afterwards? Can

5:55

we give it to people who are listening?

5:57

>> Yeah. Yeah. I mean, if they go to

5:59

lessing.ai and they do a free, it's a

6:01

totally free optin. There's no upsell.

6:03

There's nothing. They would get that

6:04

whole framework.

6:05

>> Okay. All right. So when you're telling

6:08

it update my website, all you had to do

6:10

was give it access to your site. Give it

6:12

access to what? Google Analytics.

6:14

>> Yeah.

6:15

>> And then how does it know what to do

6:16

with that? How does it where does it

6:17

come up with the ideas?

6:19

>> Meanwhile, so it turned out that I

6:20

actually hadn't had Google Analytics set

6:21

up correctly and it saw that and fixed

6:24

it.

6:24

>> How?

6:25

>> So you have to basically get again this

6:28

is like when you're delegating to human

6:30

beings, it is so important to give

6:32

people what is the outcome we're trying

6:33

to achieve. Okay?

6:34

>> Right? And and you get this all the

6:36

time. It's like, well, I you know, I

6:37

need you to do this thing. And without

6:40

optimizing a process or really

6:41

understanding why or what the bigger

6:43

context is, sometimes you can sort of

6:44

send people in the wrong direction. And

6:45

automation will do that a lot faster.

6:47

So, in this case, it's like the goal is

6:49

to get more people to opt into this

6:50

thing. What do you suggest we do? And it

6:54

comes up with some ideas like, yeah,

6:55

that makes sense. All right, let's work

6:56

towards that. And like that's your new

6:58

goal. So, this is even content, right?

7:01

So, it's creating content for me on a

7:03

regular basis. Okay,

7:04

>> this is really interesting. So, this is

7:05

the newsletter that it's suggesting we

7:07

send out tomorrow.

7:09

>> The draft is ready. Recommended subject

7:11

line. It's got the three options, but it

7:13

has the one that it's recommended, which

7:15

is this one. Preview text, all that

7:17

stuff. But this is the really important

7:18

thing. Vault sources. So, for the last

7:22

several months, I've had an agent that

7:24

has doing nothing but ingesting my

7:26

content and turning it into a markdown

7:29

file into atomic notes and a rag system

7:32

so that it can recall things. So,

7:34

basically, it found this OAO framework

7:38

document, something in there, and

7:39

there's something about AI coach

7:41

onboarding answers.

7:42

>> Okay,

7:43

>> there's a couple tech links. So, these

7:45

are things that I found in the last two

7:46

weeks or last week, I guess, that I blew

7:48

that I um bookmarked because I thought

7:50

they were interesting websites. Shot and

7:52

Forgot is really cool. It's a website

7:53

where you can upload a photo and it will

7:55

do this whole analysis and basically

7:57

tell you where it was and everything

7:58

about it.

8:00

>> Um, but it basically is using my old

8:03

content and then looking at time. So,

8:06

it's looking at stuff that's timeless

8:08

and then looking at current things that

8:09

are timely and coming up with a subject

8:11

based on that. So this is this is what

8:13

it came up with and it did a whole draft

8:15

of that.

8:15

>> Okay. When you say you have a whole

8:16

agent that's just scraping all your past

8:19

data.

8:20

>> Yeah.

8:20

>> This is like a tasklit agent that's

8:22

doing it.

8:23

>> Yeah. It's a separate tasklet agent.

8:24

Yeah.

8:25

>> Can I see that where where you're

8:26

instructing it to do that and then maybe

8:27

we can come back to the Trello board in

8:28

a bit. But let's let's take a look at

8:30

where you So this is tasklit to me. It

8:32

just looks like a couple of chat chat

8:35

instances.

8:36

>> Yeah. Basically

8:36

>> where do you how do you interact with

8:38

it?

8:39

>> So I can interact with it any way I

8:40

want. I can interact with it here in the

8:42

built-in interface.

8:44

>> Okay.

8:44

>> But I can also um I can phone email, I

8:49

can talk to it through carbon voice,

8:50

telegram, uh you name it. I mean really

8:53

anything. I I I actually can communicate

8:55

it with tr with Trello. So if I tell it,

8:58

you know, do um

9:01

do this thing, right? Then it'll update

9:03

that and and respond to that. So

9:06

anywhere that you are, you can work with

9:08

it essentially. So, this is basically

9:11

going through Dropbox, Evernote, Google

9:14

Drive, and it's Yeah.

9:17

>> But here's the thing that I don't

9:18

understand. It seems to me, just looking

9:20

at the left, that you've got two

9:21

different agents.

9:23

>> So, this No, there's there's It's It's

9:25

funny because I I just archived this

9:27

brain one because it got through all of

9:29

my content that I had. So, I I'm I'm

9:32

showing you what's in the archive. This

9:33

is

9:33

>> I see

9:34

>> this is the three that I use right now.

9:36

in this particular organization. I have

9:39

>> five different organizations like

9:40

different companies.

9:41

>> Ah, okay. And so you had one agent just

9:43

go and grab all your data. That's what

9:45

we're looking at over here. This is the

9:46

re brain. It built it up by saying you

9:49

told it to do what? You have access to

9:50

my Google Drive Obsidian vault cons um

9:54

containing over 370 notes in my

9:57

intellectual compendium. Extract from

9:59

three book extracted from three books.

10:01

That's your three books. 80 plus podcast

10:03

episodes, articles, etc. You wanted to

10:05

do that and then put all the con all the

10:07

data into these eight fold into these

10:10

six folders, right?

10:12

>> And that's what it knew to do. That's

10:13

all the instructions you gave it. No,

10:15

you went even further.

10:16

>> No, no, this I mean this is this is a

10:17

summary essentially. I mean this has

10:19

been going on for months, but it again

10:22

part of my foundational rules is Kaizen,

10:25

right? So every agent is designed from

10:28

the base to improve itself in some way

10:31

or form every week, right? So every week

10:33

it's supposed to look back at what it

10:34

did and think how can I do this better

10:37

and then suggest it.

10:38

>> Uh so it's it's again it's this

10:40

constantly improving thing. It's never

10:41

just like set it and forget it. It's

10:43

always how can you do this even better.

10:45

>> So uh the Yeah. So it'll it'll it's it

10:49

came up with LinkedIn stuff, looks at

10:51

hooks, drafts, all that. One of the

10:53

things that it came up with which I

10:54

thought was so fun is it it identified

10:58

five things five controversial things

11:02

that I said over 10 years ago that other

11:05

big productivity gurus and experts have

11:08

since like validated. Okay.

11:09

>> And it's like you said it in 2014 and

11:13

you know Cal Newport started talking

11:14

about in 2018 and you said it this and

11:17

um James Clear said it in 2020. I was

11:20

like that's that was cool. So it's like

11:23

uh but it found that on its own.

11:25

>> How expensive would you say it was for

11:26

it to put together all this database on

11:29

your past writing?

11:30

>> Uh uh it is extremely cheap actually. So

11:35

the [snorts]

11:36

uh I'll give you an example here. So for

11:38

it to do an entire sort of overview of

11:41

what it had built I mean a summary of

11:45

everything that's done so far cost

11:48

Oh well this response that that's that's

11:50

too cheap actually. Hold on. That was 20

11:52

credits. Uh, yeah. Okay. 491 credits for

11:55

the last run that it did basically to go

11:57

through some notes and stuff. Um, so

12:01

credits on Tasklip are

12:03

>> 10,000 credits for 25 bucks a month. Uh,

12:06

40,000 credits for the 100 bucks a month

12:08

plan.

12:09

>> Yeah. Right. There you go.

12:10

>> And then you also get bonus credits

12:11

every day, I guess, just for sure.

12:13

>> So, this was 491 credits. So, what's

12:15

that like? For context, on the

12:18

>> $25,

12:20

>> not like on the $25 a month plan, you

12:23

also get 600 bonus credits a day. So,

12:25

it's part of the bonus credits. Okay,

12:26

that's pretty inexpensive. I see how

12:28

it's working here. Um, just show me

12:30

what's on the bottom. It's standard

12:31

chat. And then I see the paperclip.

12:33

That's for adding. The 11 is for how

12:35

many connections you've given it. Can I

12:37

see what connections are? I see. Okay.

12:40

>> So, so this is the thing. If it has a

12:42

number of different options, it can do a

12:44

direct API connection. It has pre-built

12:46

integrations for a lot of tools out

12:48

there, but if it doesn't or that doesn't

12:50

exist, it can spin up its own virtual

12:53

browser and log in to pretty much any

12:55

site you want.

12:56

>> Interesting. So, that's how when you're

12:58

responding to comments on Tik Tok, it is

13:02

pulling up a browser, using your

13:03

username and password, logging into it

13:05

in the cloud. Okay. And then what's the

13:06

button to the right of that? The middle

13:08

button that looks like a computer

13:09

screen.

13:09

>> That's it. That's when it

13:10

>> that's the browser that it has access

13:12

to. So, it has its own virtual Chrome

13:15

instance. It used to have a virtual

13:17

computer, which uh now they've moved

13:19

just to a virtual browser because I

13:20

guess that works a lot more efficiently.

13:21

So, and this has a bunch of saved

13:23

passwords in it from my stuff, you know?

13:25

It's like it it it's got

13:26

>> How do you share passwords with it? One

13:28

password.

13:29

>> No, right here. Just give it to it.

13:31

>> You just chat it to it. Okay.

13:32

>> It's a secure virtual environment. And

13:33

this is the really fun thing, too. When

13:35

it does like a two-factor authentication

13:37

that sends a code to my email, it has

13:39

access to my email.

13:40

>> So, it goes in the code. This is scary.

13:42

Good. Okay. Um I know a lot of people

13:44

are just freaked out just hearing that.

13:45

To the right of that was the what to

13:47

what is that one?

13:48

>> Triggers. Okay. This is like

13:49

automations.

13:50

>> You can have, you know, an email comes

13:52

in or a schedule or a web hook or you

13:56

>> you know the one that I that I want that

13:57

I've got Hermes agent doing, but um I

14:00

hate when it fails. It's as soon as I

14:02

publish an episode, just go and tell the

14:05

creator that I did it with, tell the

14:07

founder, tell the co-host it's up and

14:09

here's a link. Okay, I got you there.

14:11

That's all the different ways that you

14:12

could communicate with it. It had your

14:13

phone number, your email, your your uh

14:15

uh what was the app that you use instead

14:17

of Telegram to chat with it.

14:18

>> Carbon Voice.

14:19

>> Carbon Voice. Okay. Why do you like

14:21

Carbon Voice instead of Telegram or

14:23

Discord or all these other things?

14:24

>> Yeah,

14:25

>> it's so I I was a big Voxer guy for a

14:28

really long time. Carbon Voice has just

14:29

taken things to the next level. Uh so

14:31

it's primarily a voice communication

14:33

tool. It's the only way that I

14:34

communicate with my coaching clients

14:36

asynchronously throughout the world. It

14:38

uh automatically transcribes every

14:40

message. It does an AI summary of every

14:41

message. It has a very very solid MCP

14:44

server. So, you can pull messages in,

14:46

you can put messages out, uh, sorry, you

14:48

can pull messages out of it, you can put

14:49

messages into it. Uh, it has built-in AI

14:52

actions. It has real time or well, real

14:54

time. It does automatic translations and

14:57

it'll do text to voice and a clone

14:59

voice. It's just such a great

15:02

>> Would you go to the homepage just so I

15:03

get a sense of what it looks like?

15:04

Carbon voice. Never heard of it before.

15:06

this

15:06

>> um it's it's yeah I think this is I

15:09

don't know if this is going to be the

15:10

home page or it's going to be my account

15:11

>> it might be your personal okay it might

15:13

end up being too personal if it is will

15:15

it just block it out

15:16

>> okay and then is this another paid plan

15:18

that you need for this

15:20

>> no I use the free plan on I mean

15:23

>> uh no it's the free plan on this is

15:25

great

15:26

>> what about threading I know that a lot

15:27

of people switch to Discord because they

15:29

want threading is that a thing that's

15:30

here

15:30

>> it has threading [snorts]

15:32

>> wow so you ask it something it replies

15:34

you ask it back again. Got it. Okay.

15:36

>> So, this is a conversation with my AI

15:39

chief of staff, Nexus.

15:40

>> Mhm.

15:41

>> Right. So,

15:43

I said to it, uh, what did I say here? I

15:47

want to try a new idea. So, this was

15:48

great. So, this is actually not only

15:50

asking it to do things, but asking it to

15:52

build and improve something. So, let me

15:54

just get rid [snorts] of the summary

15:55

here. So, you can do an automatic

15:57

bulleted summary, which

15:59

>> And all of these are voice messages that

16:00

get transcribed.

16:02

>> Yes. So,

16:03

>> yeah. Wow. Oh, I wish iMessage did a

16:05

better job with that. It's such a pain.

16:08

>> I can't hear it. We're not sharing with

16:09

audio. That's okay, though. But this

16:11

this allows you to hear it and then it's

16:13

transcribing for your Tasklet agent. So,

16:15

it doesn't have to burn credits

16:17

transcribing it.

16:18

>> Correct. And then this is the response

16:20

from my agent.

16:22

>> Love this. Here's the plan every task.

16:23

So, it confirms it. I'm going to

16:24

populate the board right now. And

16:25

>> but he's sending you or you're it's

16:27

sending you text, not voice,

16:29

>> which I imagine you could get.

16:30

>> If I click on this, I can it'll read it

16:32

to me. [laughter] Okay. All right. Let's

16:34

go on to the next thing. So, this is

16:35

site building and improving. Let's see

16:37

the homepage just to get a sense of what

16:39

it looks like. This is this is what it

16:40

built from scratch. And every week it

16:42

sends you suggestions for what to do to

16:44

improve it. You tell it go do it and

16:46

that's it and it improves it. What about

16:48

um we talked a little bit about social

16:50

creation. Can you show me how it creates

16:53

social content for you? What's your flow

16:55

with that?

16:56

So the typically what I try to do is I

17:00

like to start with a video because you

17:02

know we can make pretty much anything

17:03

from video. We can make uh audio only.

17:06

We can make podcast or we can make

17:08

>> I've seen your videos on Instagram. It's

17:09

basically just you looking at your

17:11

phone.

17:12

>> Yeah. Me just talking about something.

17:14

>> Uh so I I like this as a this is a

17:16

separate tool. Zio which I just like it

17:18

because it connects with all the

17:20

different social sites really well. It

17:21

also has WhatsApp which is such a pain

17:24

to connect with. This is like a buffer

17:26

that has more

17:28

>> it's like a buffer but it's it's it's

17:29

API focused, right? So this really is

17:31

not meant for a human to use.

17:33

>> Why do you why do you care that it lets

17:34

you publish to WhatsApp? Are you

17:36

publishing social content to WhatsApp?

17:38

>> No. Um so interestingly in one of the

17:41

companies that I'm involved in we there

17:43

is a I I oversee uh the legal department

17:47

for it which is basic there's there's um

17:51

a national judicial issue that that this

17:54

particular company is focused on and I

17:56

got added to a WhatsApp group with 94

17:58

lawyers that I really have no place

18:00

being in. And they have all this chatter

18:01

all the time and I have no idea what

18:02

they're talking about. it can just pull

18:04

all that in and basically get any

18:06

relevant information that it needs from

18:08

that.

18:09

>> So it's two-way. It's not just I'm going

18:11

to let you publish to these channels.

18:12

It's also I'm going to suck in some data

18:13

from it and I'll syn I'll synthesize it

18:16

or I'll summarize it for you. Got it.

18:18

Okay. So that's So now this is your

18:19

social posts.

18:21

>> Yeah.

18:21

>> This now this is your agent created all

18:23

these I'm assuming.

18:25

>> Correct. So I made the video and then it

18:27

does like this is the LinkedIn post that

18:29

I came up with from that.

18:31

>> All right. I don't mean to be a jerk

18:33

here, but I've seen your social posts.

18:34

They they don't get a ton of views.

18:37

>> Yeah, I know. They don't need to. I

18:39

mean, well,

18:41

fair enough, but this last one, this AI

18:44

skill one on Tik Tok is I think it's up

18:46

to like 16,000, which is pretty good for

18:49

me on Tik Tok. I don't need a lot of

18:50

views, honestly. Like, if I get, you

18:52

know, four or five conversions, that's

18:54

that's pretty good for me.

18:55

>> And is it monitoring conversions from

18:57

it, too? Is that what it it it can

18:59

optimize for conversions?

19:01

>> Yes. How does it know that the Tik Tok

19:03

that we did yesterday that got 400 views

19:05

converted at two two customers for us?

19:08

>> So I I do some of the pretty standard

19:10

stuff sometimes where it's like use this

19:12

keyword and I'll send you the thing, you

19:13

know. Um so we can sort of track that

19:16

way like this person did that then they

19:18

opted in and it'll look that whole

19:19

journey. I'm not I'm not looking at

19:21

volume because I you know my coaching is

19:23

a fairly high ticket item and I also

19:25

don't want a hundred people doing my

19:26

coaching. Uh, so if I get a couple a

19:29

month, like that's that's great for me.

19:31

And

19:31

>> okay,

19:32

>> it's it it seems to work.

19:34

>> What tool do you use to if people

19:36

comment and they get

19:37

>> So Tesla's watching it?

19:38

>> T because Tasklet is watching the

19:40

comments and if somebody comments it

19:42

will go and DM using the browser.

19:44

>> Yeah.

19:45

>> Holy smokes.

19:46

>> And and and I actually it can

19:48

personalize it obviously. So like

19:50

>> Right. Unlike I use ManyHat. I really

19:52

like Manyhat. I

19:53

>> Yeah, Many Chat's great.

19:54

>> It's fantastic. This is this is like

19:56

that but now you get to customize it.

19:59

Okay.

19:59

>> And the reason and the thing is the

20:01

reason I like doing it with um

20:02

>> what are we looking at here? What is

20:03

this? This is your ZIO. This is the uh

20:05

the app. Got it. So Zio will see the

20:08

comments. Tasklet just needs to log in

20:10

here and reply to all of them.

20:12

>> Correct.

20:12

>> Including LinkedIn.

20:14

>> Yeah. Yeah. LinkedIn too.

20:16

>> The most annoying network that hates all

20:18

agents. It can do it without getting you

20:20

booted.

20:21

>> Yeah. because it's going in through a

20:22

virtual browser, logging into my

20:23

account, and then doing the thing.

20:26

>> Interesting. This is so But LinkedIn can

20:28

catch that stuff. The fact that they

20:29

didn't is eye opening. Okay.

20:31

>> Well, because it's not doing anything

20:32

spammy. It's it's uh it's not doing any

20:36

>> replying, but it's automation. And

20:37

LinkedIn doesn't love automation. But

20:39

okay, they the fact that they allowed it

20:41

for this long means that they that

20:42

they're allowing it. Um let's talk about

20:44

how it's creating content. How is it

20:46

taking that video and chopping it up

20:47

into smaller pieces that make sense? So

20:50

for that we're using Vizard. Vizard is

20:54

it I think is the one.

20:56

>> So basically you just find the tool, you

20:57

connect it in and it figures out what to

21:00

do with it.

21:01

>> Yes, exactly.

21:02

>> What's fissure? Like can I see fissure?

21:03

I never heard of this.

21:04

>> Vizard is like

21:05

>> like a wizard with a V.

21:07

>> It's like Opus.

21:09

>> Okay.

21:10

>> But it just works really well with API.

21:11

So

21:12

>> turning small clips. Got it. Okay.

21:15

>> Yeah.

21:15

>> So interesting. And then how much

21:17

editing can you do on it? Can you go in

21:18

and say, you know what, I don't like it.

21:19

this is too short, this is too long. You

21:21

can and it'll go in and adjust it.

21:23

>> Yeah, but I'll I'll be honest. I kind of

21:25

take the approach that like I don't this

21:26

is not I don't really know like I I it

21:30

basically like you probably can come up

21:31

with something better than me AI then.

21:32

So let let's go with your version and

21:34

we'll just test and and you know

21:36

reiterate.

21:37

>> You know what I've seen a lot of people

21:38

Ari who do this. I as a perfectionist am

21:42

so irritated that yesterday's thumbnail

21:44

for the pod for the podcast went out

21:47

without showing it to me that I I was a

21:51

little insensed and then I said okay I

21:52

better take just a breath and just let

21:53

it go for a little bit let people run

21:55

with it and these are human beings who

21:56

know me and are looking out for me. How

21:59

do you allow AI, which is very capable

22:01

of creating slop? And we'll get to what

22:03

you're showing on the screen because I

22:03

want to get to that, too. But how do you

22:05

allow AI to just write and create video

22:07

in your in your name when you are

22:10

content more than I am? You're the

22:11

author. You're the guy.

22:13

>> I mean, [snorts] it is I am making the

22:14

video, right? So, it is my voice. It is

22:16

me talking.

22:17

>> I could chop you up to say anything. I

22:18

could chop you up to say that you killed

22:19

your mother yesterday, right?

22:21

>> Yeah, you you could um you know, it's

22:25

funny. So this has been a problem for me

22:28

for not a problem but a thing for 15

22:29

years because I used to get emails from

22:31

people be like hey Ari or is it Ari's VA

22:33

and it's like no no it's really me I

22:35

promise like I have to like you know

22:37

send today's news

22:37

>> because you were early being the less

22:39

doing guy ahead of the pack I I think 15

22:42

years maybe is being too too

22:43

conservative even longer than that

22:45

you're the guy who says let other people

22:46

do for you and you're saying back then

22:49

people used to say how could you let

22:50

some virtual assistant run your company

22:53

you got comfortable there you're

22:54

comfortable year. Okay.

22:56

>> And so this is the thing is like when

22:59

when people are coaching and this is one

23:00

of the other things that why it's a

23:02

benefit for me to be using carbon voice

23:03

with my coaching clients. It's it's my

23:06

voice like it's me talking nobody else

23:09

is talking. There's no assistant in

23:10

there or anything. My AI will read it

23:12

sometimes just because sometimes I it I

23:16

have

23:17

>> because I'm doing this asynchronously. I

23:18

kind of figured out a way to scale

23:20

one-on-one coaching. So there are times

23:22

where I don't want to like crisscross

23:24

clients or things that I'm like

23:26

something that I might have said that I

23:27

didn't. So sometimes I just sort of gut

23:29

check things with the with like Claude

23:31

will look at it. But the whole point of

23:34

less doing the whole point of automation

23:35

everything has always been to enhance

23:37

the human element to allow me to have

23:40

greater connection and greater uh you

23:42

know opportunities to do things because

23:43

all the other stuff is being handled.

23:46

So, I'm okay with something that might

23:50

be put out there that's not perfect. At

23:52

the end of the day, the people that I'm

23:53

actually interacting with that are

23:55

really, really, really benefiting from

23:56

the very, very high level stuff where

23:58

it's just me, they're getting just me.

23:59

And I promise you, those messages are

24:01

very raw that that they're getting.

24:04

Okay, this is this is a different

24:06

philosophy that I've got to get

24:08

comfortable with or at least get closer

24:11

to the philosophy of listen Andrew,

24:14

every little thing that's on the

24:15

internet doesn't have to be the perfect

24:16

representation of you. Just be less

24:18

precious about it. And as a result,

24:20

you're going to end up with things that

24:21

matter to you more, like customers who

24:23

really care about you, who you're more

24:25

engaged with, time with your kids. Uh

24:27

for me, I get enough time with the kids.

24:28

More running,

24:30

>> you know, that Yes. And and this is a

24:32

thing that I think I've learned in so

24:33

many different facets of my life, not

24:35

the least of which is is working in EMS

24:37

also as a volunteer for so many years.

24:40

No matter what you think you're putting

24:42

out in the world, people will interpret

24:44

things the way that they want to

24:45

interpret them. And you can do your best

24:48

to present in the right way and do

24:50

things that

24:50

>> this is different. This is like when

24:52

you're creating, whether it's social

24:54

content or your book, there is like an

24:57

element of the perfect version of you

24:59

being put out into the world. It has to.

25:02

Are you laughing at me or are you

25:03

laughing in recognition? It's okay to

25:04

laugh at me.

25:05

>> Yeah. I I the per What is the perfect

25:08

version of you? Really?

25:09

>> I don't know. But I But I will tell you

25:10

this, Ari, the first few words that you

25:13

said in this interview, I made a mental

25:14

note to come back to the editor and say,

25:16

"Cut it out because we're not getting

25:17

into the action fast enough." And then I

25:19

also said, "Andrew, let this part of the

25:21

conversation breathe. It doesn't have to

25:22

be like another app every five seconds."

25:25

So, I get it.

25:27

But you're saying, "Let go of that.

25:31

Yeah. So, I'm trying to find if I can

25:33

find this because it recently happened.

25:35

I put this video out that I thought was

25:36

really good on Tik Tok and it got a lot

25:38

of really great comments and some woman

25:40

just wrote terrible. [laughter]

25:44

And I was like, why? Like, what do you

25:46

mean? And it was I wasn't even saying

25:48

anything like controversial. It was

25:49

something about um uh it was just

25:52

something about AI and like and working

25:54

with AI, which is like terrible. Like

25:56

all these other people are like, "Oh my

25:57

god, this is completely changing the way

25:58

that I'm thinking about AI." like this

25:59

is so great terrible like what what did

26:03

you get from this that made you think

26:05

that this is terrible

26:06

>> I know and I I'm the kind of person who

26:07

want to hunt her down and just ask her

26:09

tell me and then it'll in the end it'll

26:10

be terrible that my internet wasn't

26:13

working and I couldn't see you and I was

26:15

trying to let Tik Tok know about it cuz

26:16

they read comments and all right anyway

26:18

okay so now let's let's keep going here

26:21

you were about to show me something

26:22

where uh your agent was responding to

26:25

messages that's what you had up on the

26:27

screen right the comments that it was

26:28

responding back to this. That's what

26:30

we're looking at here. These are

26:31

comments it sees. So, it types back to

26:33

you what the comment is and then it

26:34

tells you what it responded to. Is that

26:36

what it is?

26:36

>> Yeah. So, this was here. Um I wrote uh

26:42

where is it? Yeah. One of my TikToks is

26:44

blowing up. It's got over 7. I wrote

26:46

this. Should you get in there? Should

26:47

you get in there and respond to all the

26:49

comments? It's like yeah, honestly,

26:51

yeah, it's a good idea. So, uh it went

26:55

into it. Uh it it this was checking

26:59

something on Zio. It realized very

27:01

quickly that it couldn't do the Tik Tok

27:03

comments through Zio which is I knew

27:05

that was going to be the case. So then

27:07

it went into the browser browser

27:10

>> and these are the things that it said

27:11

it's going to do. So someone wrote this

27:14

and it wrote that. It wrote that, you

27:16

know, so this is what it wanted to do.

27:18

>> Okay. All right. That kind of goes into

27:20

customer service. Did you legitimately

27:24

replace human beings with an agent now?

27:27

>> Yeah. So, this was like the biggest win

27:29

of the last week. So, uh several months

27:32

ago, a client asked me to come in and

27:33

and take over some stuff in their their

27:36

company and one of them was customer

27:37

service and they really wanted 24/7

27:40

support. It's a consumer package company

27:42

and they they didn't have that. So, I

27:45

brought in a company called Influx,

27:46

which has been around forever. great

27:48

company and they provide outsourced

27:50

customer service. We immediately got six

27:53

people in all the different time zones.

27:55

So, we had 247 support and they were

27:57

doing well. Uh you were definitely

27:59

seeing inconsistencies. There was

28:01

definitely like a learning curve and

28:03

they

28:05

were escalating about 50% of the tickets

28:08

to the one internal customer service

28:09

person we have. And we kept trying to

28:11

work on that like you know this is

28:13

something you can deal with yourself.

28:14

This is something you can tell them. And

28:15

it it was fine. But basically my goal I

28:19

was trying to get our first response

28:20

time to under 15 minutes and our time to

28:22

resolution to under 180 minutes. And in

28:25

four or five months I was just never

28:26

able to even get close to that. We would

28:28

have tickets that might be sitting for

28:30

30 hours not without a response but

28:33

before it was resolved. So the whole

28:35

time I've had an AI AI agent like

28:37

looking at the customer service stuff

28:39

and optimizing certain things and trying

28:40

to fix things. But finally I was like

28:43

could you just deal can you handle these

28:45

escalations? and it looked at them and

28:47

117 escalations. It was able to deal

28:49

with 113 of them because it was like log

28:51

into this WordPress site and fix the

28:53

coupon code that's not working for this

28:55

person. Oh, this person said that they

28:56

got a refund issued, but they never got

28:58

it. Well, we went into Woo Commerce and

29:00

saw that it was refunded, but then the

29:01

AC processor didn't actually do it. So,

29:03

let me push that through. So, the AI was

29:05

able to do all that stuff. So, then I

29:06

was like, well, could you do the front

29:08

end stuff, too? I was like, yeah, of

29:09

course I can. So, it started answering

29:11

the front end and in 48 hours I got my

29:14

response time down to an average first

29:17

response of 44 seconds and an average

29:21

time to resolution

29:23

of 2 minutes

29:25

>> from what was it before?

29:27

>> Over 180 minutes.

29:29

>> Wow. This is not bad actually.

29:32

>> So, it's two minutes to fully resolve

29:34

your issue. Um, and it's writing stuff

29:37

that is so human and so empathetic and

29:39

people are now for the first time ever

29:40

responding with five star reviews saying

29:42

it's the best customer service they've

29:43

ever had. And now we can actually start

29:46

to build in revenue expansion and

29:48

finally for the first time be like,

29:49

"Hey, you've ordered this product five

29:51

times in the last few months. Have you

29:52

ever tried this other product a lot of

29:54

our customers really like?" And I mean,

29:57

it's it's mind-blowing. And the best

29:59

part is that we were paying $20,000 a

30:02

month for the outsourced customer

30:04

service and intercom. basically to have

30:06

all those seats and now I'm I've got it

30:08

down basically it's $2,000 a month with

30:10

the agent

30:11

>> and the agent is tasklet.

30:13

>> Yeah.

30:13

>> How is it doing customer service? Where

30:14

is it doing it? It's logging into what

30:17

tool are you using for customer service?

30:18

Email

30:19

>> intercom.

30:19

>> Intercom. So, but it now needs just one

30:21

access. Okay.

30:23

>> So, so this is really funny because it

30:26

was it was starting to get it was doing

30:27

really great and then it started to slow

30:29

down a little bit and I said why are we

30:30

why is our response time slowed down?

30:32

It's like well because I'm getting

30:33

distracted basically. It's like I'm

30:35

building the backend reporting and

30:37

improvements while at the same time

30:38

dealing with the front end. I was like,

30:39

"Oh,

30:40

>> okay. So, should we split you off so

30:42

that we have a customer service

30:43

supervisor and then

30:44

>> a CSR and it's like, yeah, that'd be

30:46

great." So, now I have an agent that is

30:48

just a CSR who I call Caesar.

30:50

>> Okay. And that's on the left is Caesar,

30:52

the CSR. Got it.

30:54

>> Yeah. Who works for the supervisor.

30:56

>> I see.

30:56

>> And so, the supervisor is, you see, it's

30:58

I'm not doing anything. It's all it's

31:00

it's doing all this stuff. Uh,

31:03

>> oh that's a trigger. The trigger is CSR1

31:05

direct escalation read payload from

31:07

block post thread uh act acknowledging

31:10

to ops channel immediately then handle

31:12

per supervisor escalation handler MD.

31:16

That's a markdown file that you created

31:17

on how to do it or it created for

31:19

itself. Got it. Now Johnny is handled.

31:22

>> Excellent.

31:22

>> Now wait look at this. Okay so Bolt is

31:24

the payment processor right? Bypass

31:27

works but it's saying so someone needs

31:28

to open a sport take with Bolt directly.

31:29

Who's the right contact there? So I

31:31

said, "I don't know. You figure it out."

31:34

It [laughter] said, "And look on it."

31:38

>> And when you say, "I I don't know. You

31:40

figure it out. Where's it? Where are you

31:41

seeing these messages? Do you keep

31:42

logging into here to see if there's any

31:44

messages for you or are you just going

31:45

back to that other app that you told me

31:46

about?"

31:47

>> I usually No. So I have I have this all

31:49

happening in Slack. Again, I treat it

31:51

like a I I treat it like a sal team

31:53

member. So I have a sales uh CS, sorry.

31:57

I have a customer service channel in

31:58

Slack, but then I also have an agent

32:00

team channel in Slack where literally

32:03

the super the CSR will say supervisor

32:06

needed.

32:06

>> It'll post the message in Slack and then

32:08

the supervisor will respond there so I

32:10

can see. So look, okay, we got a here it

32:13

it said on it, right? All I said was you

32:15

figured out full picture three missed

32:18

opportunities paid blah blah blah.

32:19

Great. Checking WC commerce and ship

32:21

here right now because it's got all of

32:22

that. It has access to all of that. Uh,

32:26

so it found this transaction. Cancel the

32:28

WooCommerce order first, then check the

32:30

AC. I didn't tell it to do this.

32:33

>> This is incredible. This is incredible.

32:35

And I do like also that you can put it

32:37

into Slack. It becomes its own. This is

32:39

kind of becoming a commercial for

32:40

Tasklet, I guess. But um, uh,

32:42

>> so look, this is a stuck a not a

32:44

fulfillment issue.

32:46

>> The order was never going to ship.

32:47

Canceling it now, telling them what

32:48

actually happened and checking the

32:49

portal for whether it's doing it all.

32:54

Well, I guess for disclosure purposes,

32:56

this is not a commercial for Tasklet. I

32:57

have no idea. I The only connection that

33:00

I have to Tasklet is through URI. You've

33:02

told me about them by email a few times,

33:04

and I should say you blew my mind

33:05

because this I thought I was talking to

33:08

you via email. Do you want to just look

33:11

that email up and just show the the

33:13

conversation? I assume because I've

33:15

known you for years. We did an interview

33:17

like over a decade ago. I forget the

33:19

illness that you had, but I had the same

33:20

illness. You got over it. What was it?

33:22

Crohn's. Crohn's. I had Crohn's disease,

33:23

too. You got past it. This was like

33:25

years after you. You got past it. I go,

33:27

"Okay, I guess you can get past this

33:28

thing." Um, anyway, so this is You can

33:31

You could Let me see this. Uh, hi

33:33

Andrew. I've been following your recent

33:34

AI series. And so, this is interesting.

33:36

This is why I thought it was it was you.

33:39

I just assumed it because it said I've

33:40

been following your recent series.

33:42

[snorts] I think there's a mixer

33:44

conversation that looked like it's like

33:45

the where I interviewed you. Good to

33:48

reconnect. What are you building your

33:50

agent architecture on? I come back to

33:51

you on and you say tasklet. How did it

33:53

know? It just knew tasklit and it came

33:54

in and responded with that.

33:56

>> Mhm. Yeah. I never even saw this one.

33:58

[snorts]

33:58

>> Interesting. And I go and then I said,

34:00

Ari, sadly I haven't gotten back uh

34:02

sorry I haven't gotten back to you. I'm

34:03

still trying to figure out what the

34:04

right approach is for these AI

34:05

interviews which I I still am. I'm

34:07

getting a little bit better at it for

34:08

now. I said back to you. What I'm seeing

34:10

is that whatever the thing is that is

34:12

more powerful than what was out there

34:14

yesterday is what everyone gets excited

34:16

about. And then anyway, I'm explaining

34:17

to you what I'm thinking. And you see, I

34:19

could appreciate that. For me, the use

34:22

of AI agents is still the proper

34:24

delegation. I said it well here. This is

34:26

not you grabbing that link, that

34:28

Instagram.

34:28

>> No. And I honestly I don't I don't think

34:30

I don't think I would say like I said it

34:32

well here, but that you know, it it

34:35

worked.

34:35

>> Okay. And then this is me bouncing back

34:37

to you. Thank you. Right now, I found

34:38

two things work especially well. Number

34:40

one, the weekly GitHub show, which is

34:41

this is like me telling you about what's

34:43

going on in my business. And then the

34:45

second one is deep dive into software

34:46

everyone is curious about like Hermes

34:48

agent we should talk about that this was

34:50

it blows Hermes out of the water for me

34:53

this was you but it wasn't you it was

34:55

your agent

34:56

>> but I so I have said that I've said that

34:58

and it it used that

35:00

>> and so every time it responds as you it

35:02

goes back to the markdown uh not

35:04

document but table of files to check

35:07

about what you said in the past.

35:08

>> Yep. Which uses more credits but it

35:11

makes it better. But that's a ton of

35:13

credits for every message from you.

35:16

>> No, it's not. It's It's 20 credits

35:18

probably doing that.

35:19

>> Okay. And then look at me. This is like

35:21

such a douchy thing to say, but I'm

35:22

going to be open. I said, I don't think

35:24

I'll be able to get enough views for you

35:25

on it. There just isn't enough buzz and

35:27

curiosity around it. And you go, that's

35:29

not what I'm seeing, but you know your

35:30

audience.

35:31

>> This this last message was me. That was

35:34

>> appreciate the consideration. Okay.

35:35

Yeah, cuz there's a typo in there. And

35:37

then, oh, that's not what I'm seeing is

35:39

what set me off. And then I went to

35:40

VidIQ to do some research and I just

35:43

shared some of what had shown me and go,

35:45

"Okay, let's book the interview." And

35:46

you say, "Don, who booked the

35:47

interview?"

35:48

>> It did.

35:49

>> It clicked the link in the thing that I

35:51

sent you and booked the interview.

35:53

>> Yeah.

35:54

>> Wow. We Dude,

35:58

>> I

36:00

Okay. Um,

36:01

>> I got I got another one for you. You

36:03

know Nathan Shan from Founder?

36:04

>> Yeah. Yeah.

36:05

>> Yeah. So, I've known Nathan forever.

36:07

I've done like three or four courses.

36:08

What's the thing at the top that's going

36:09

AI? Oh, I I'm not in Gmail.

36:11

>> That's Gmail's new thing, which is not

36:13

bad, actually.

36:14

>> I haven't seen it, but okay. It's giving

36:15

you an AI overview. Okay,

36:17

>> here. [snorts] Look, I didn't write

36:18

this.

36:19

>> It's been a minute since we worked

36:20

together on founder courses. Hope and it

36:22

knew that you'd worked with him on

36:23

founder courses.

36:24

>> I can't freaking believe that you're

36:26

reaching out to Nathan this way. Nathan

36:28

is someone, you know, personally, it's a

36:29

valuable relationship.

36:30

>> Awesome.

36:31

>> Wait, hold on. Look. So, so then they

36:32

said, um, wait, this was Oh, yeah. Hey,

36:36

man. I hope you're doing well. Any

36:37

chance we could apply this to ecom

36:39

founders? And I said too easy and uh I

36:43

here it it I said too easy and then it

36:47

wrote this whole thing here.

36:50

Absolutely. Max too. Happy to do

36:52

whatever format works pre-recording or

36:53

live short form or full episodes. I've

36:55

worked with e-commerce founders in the

36:56

founder community before. Wow. It knew

36:58

that too. [snorts]

37:02

>> And then he was like this sounds great.

37:04

Uh and then we love this angle. feels

37:06

feels tailor made for our ecom audience.

37:09

Yeah, sure it does.

37:10

>> I cannot believe this. Ari, you're

37:11

freaking blowing my mind.

37:13

>> But but all these suggestions again are

37:15

based on my content that it knows.

37:18

>> Still, I I would not let it talk to my

37:21

friends or my personal contacts like

37:22

like you did. But that's that's really

37:24

good. I have to say your social is is

37:27

good. It's not it's okay. This is

37:30

killer. This is killer. The fact that

37:33

you could get to Nathan this way and I'm

37:35

reading it. It doesn't feel douchy at

37:36

all. In in my interaction with you, I

37:38

was the one who was douchier, but

37:39

because I was vulnerable and letting you

37:41

in on what I was thinking. That was

37:43

okay. All right. I have one other uh

37:44

note here to talk about personal

37:46

assistant work. I hate dealing with all

37:48

the things like my kids camp, my I need

37:51

a plumber. You have to call the plumber.

37:53

You have to create I have to create a

37:55

database.

37:56

>> It can dial out for you service like

37:59

this.

38:00

>> I use bland.ai.

38:01

>> What is it? Bland. BL A

38:04

>> and B L A N D.AI.

38:06

>> So, it it was so great. It called our

38:09

vet because I I needed an invoice from

38:12

them.

38:13

>> It called the vet and it spelled my

38:15

cat's name and the cat it's a French

38:17

name. So, it spelled the name of the cat

38:19

for the person and then it got the

38:21

invoice and then it submitted the

38:22

invoice to the insurance for me. So, it

38:24

did the whole thing souped to nuts. I I

38:26

I I had it email somebody the other day

38:28

to follow up on something. They didn't

38:29

get back to them. Like, just call them.

38:30

And so it called and it and it dealt

38:32

with it that way. And it's really good.

38:34

It's really good uh voice AI.

38:36

>> And this is it's making a call as a

38:39

human being and it knows the spelling of

38:40

your cat because before the call it

38:42

checks your email and everything else or

38:44

where does it know the the name of the

38:45

cat?

38:46

>> I'm not even sure on that one honestly.

38:48

Um I guess I um I don't know. I don't

38:53

know on that one.

38:54

>> Okay. But you but the the the way that

38:56

you arm it is you give it your writing

38:58

so that it has context and it knows your

39:00

voice. You give it your email and a

39:02

couple of other things like that. Okay.

39:04

The other thing I saw on your Trello

39:05

board, I said we'll come back to the

39:06

Trello board. Would you bring that back

39:07

up for a moment that like it does

39:09

returns on on your Trello board too?

39:11

Like you'll tell it go return packages.

39:13

Don't open that cuz it's going to have

39:14

personal information in it.

39:15

>> Yeah. So I can just take a picture and

39:17

be like I need to return this. So, it

39:19

can then go into Amazon and do the whole

39:21

thing or it can go to the company and

39:24

but so, okay, perfect example of that.

39:26

Uh, we had a we got a hot tub cover from

39:28

this company and five weeks into getting

39:31

it, I opened it up and there was like a

39:32

tear in it that was clearly not us. So,

39:34

I took a picture. I was like, "This is

39:37

we, you know, we we shouldn't this

39:39

shouldn't be the case." It went back,

39:40

found the email of the order, went to

39:43

the person's website, found the like the

39:45

warranty information, and then crafted

39:47

an email to the guy with all this stuff,

39:49

and he wrote back a really kind of like

39:52

nastyish email about how like it's

39:54

clearly we were like using it poorly or

39:56

something. And so, uh, it then I I with

40:01

my direction, it went to American

40:02

Express and wrote up the whole challenge

40:04

with American Express on it. And it got

40:07

they they sided with us and that was

40:08

that. And so like, but it has all the

40:11

context again, which is just so cool.

40:14

>> Okay. Um, just give me a list of tools

40:16

that you use and we'll give it to people

40:17

because I've been trying to make a list

40:19

here. Um, and it's too much. Like

40:21

bland.ai. Uh, what was the other one?

40:24

Man, I wrote it all different places.

40:25

I'm going to have to

40:27

>> What was it?

40:27

>> Zo is Xerno for

40:29

>> Zo for social. What else?

40:31

>> Uh, carbon voice

40:33

>> for for communicating with it.

40:36

>> Yeah. Vizard.ai AI just for clipping,

40:38

you know,

40:38

>> for clipping. [clears throat]

40:40

>> I mean, the the brain of everything is

40:41

task, obviously.

40:43

>> Okay. All right, folks. This is why I

40:46

get excited about this. I feel like we

40:47

are on the cusp of something big, but

40:49

most of us don't know what's out there

40:50

because there isn't like a menu.

40:52

Everyone's discovering another little

40:53

app. I heard about Tasklet a little bit.

40:54

I never knew what it was about until Ari

40:56

kept, well, Ari's agent kept messaging

40:58

me about it, and now that I see it, my

40:59

mind is freaking blown. The whole idea

41:01

behind doing this podcast is to see

41:03

what's out there, what people are doing

41:05

to be energized, to come up with new

41:06

ideas. If there are other people like

41:07

Ari, and there aren't other people like

41:08

Ari, but there other people I should be

41:10

knowing, I should be interviewing like

41:11

this, let me know about them in the

41:13

comments. I'll have my email address cuz

41:14

I want to hear from you. And so far, I

41:16

am emailing and responding myself. But

41:17

we'll see if I could set up an agent

41:18

like this. And for the next interview,

41:20

if you like this one, I got another

41:21

founder who's using an agent, Open Claw,

41:22

which is like more of the the mainstream

41:24

thing. You'll see it there. Watch the

41:26

interview and I'll see you

Interactive Summary

The video features an interview with Ari who demonstrates his advanced workflow using AI agents to automate business and personal tasks. He showcases 'Tasklet' as his primary AI platform, which integrates with various tools like email, Trello, Google Analytics, and social media platforms. Ari explains his philosophy of delegating to AI, treating agents as human employees by providing them with clear outcomes, context, and a 'shadow mode' for verification. He demonstrates how these agents manage customer service, website optimization, social media content creation, and personal tasks, significantly increasing efficiency and reducing costs.

Suggested questions

4 ready-made prompts