0:00

Today on the AI Daily Brief, the

0:02

most in -demand agent use cases right

0:04

now. The AI Daily Brief is a

0:06

daily podcast and video about the most important news

0:08

and discussions in AI. To join the

0:10

conversation, follow the Discord link in our show notes. Hello,

0:18

friends. Spring Break Week continues with

0:20

another interview episode, and this time I

0:22

am once again joined by Neufar Gaspar to

0:24

discuss the agent use cases that we're

0:26

seeing come up most often. Nufar

0:29

is once again a brilliant AI analyst, product

0:31

and strategy leader who has consulted with

0:33

some of the biggest companies in the world

0:35

on AI strategy and who works with

0:37

us at Super Intelligent on our agent readiness

0:39

audits and our agent marketplace. Today

0:41

we're looking at some of the agent use

0:43

cases that are most in demand coming out

0:45

of both our audits and our agent marketplace

0:47

as a way to potentially help you understand

0:49

what agent uses are actually ready for prime

0:51

time and which still remain a little bit

0:53

farther away. All

0:56

right, Neufar, welcome back to the show. This

0:58

is not exactly a part two from

1:00

what we did before, but I think a

1:02

lot of people will connect the dots.

1:04

Before we were talking about what mistakes are

1:06

common that we're commonly seeing among organizations

1:08

as relates to AI broadly, but agents specifically.

1:11

Today we're talking about what agent use

1:13

cases are actually ready for prime time,

1:15

what agent use cases are being implemented

1:17

right now and people are finding success

1:19

with. And again, this comes back and

1:21

hearkens to sort of a theme from

1:23

that previous show as well, which is

1:25

how can we help organizations spend their time

1:27

more effectively in this in

1:29

this agent and AI transformation. So

1:32

let's dive in. We're going to talk

1:34

about some use cases specifically, obviously. But

1:36

I think that where we wanted to

1:38

start is talking about based on our

1:40

set of experiences, which includes, you know,

1:42

a huge number of conversations with different

1:44

types of companies, both in the context

1:46

of the super intelligent work with with

1:48

these agent readiness audits, but also in

1:50

terms of your independent practice and consulting.

1:52

What are we seeing on average at

1:54

the risk of maybe being overly reductive.

1:56

Where are orgs with agents on average,

1:58

at least in franchised organizations that we're

2:00

tending to deal with? Yeah. So

2:03

let's assume that there is some

2:05

bias because if they're very, very mature,

2:07

perhaps they will not come to

2:09

us. But everything that we've seen thus

2:11

far are companies that we categorize

2:13

as either agent initiation or agent exploration

2:15

phase, meaning that they're either just

2:17

starting to contemplate agents or maybe they've

2:20

started working on a handful of

2:22

agents or some agents ideas. But in

2:24

general, they're very early stages. And

2:26

like we talked about the seven common

2:28

mistakes, in many cases, there are

2:30

some are not ready that they have

2:32

a lot of work to do in

2:34

preparation for agents even before they

2:36

can introduce the first one. So they're

2:39

very, very early on. We are

2:41

seeing some organizations and we actually

2:43

encourage them in some cases that almost

2:45

have no AI adoption but are

2:47

looking into agents as the way for

2:49

them to bridge the gap of

2:51

how much behind they are. In

2:53

some cases, these are smaller organizations

2:55

where it makes more sense for them

2:58

to hire an agent versus hire

3:00

a human employee. And in other cases,

3:02

they will still have to do

3:04

the groundwork of getting agent ready before

3:06

they will be able to do

3:08

this bridging the gap with agents. Yeah,

3:11

they want to talk about also the advance. Absolutely.

3:13

Yeah, so I think maybe

3:15

a better way to frame this even than just

3:17

where orgs are on average is sort of what's the

3:19

band of organizations that we're seeing, the

3:21

common band from beginner to sort

3:24

of a little bit more advanced.

3:26

Yeah. So from what we've seen,

3:28

even the most advanced organizations, one

3:30

that already have agents in production,

3:32

we're only talking about, in many

3:34

cases, some handful of meaningful agents

3:36

in production. And I'm not talking

3:39

about the personal productivity that individuals

3:41

are creating for themselves. These are

3:43

not in my book, Meaningful Agents

3:45

in Production. And I'm not also

3:47

going to discuss the question of

3:49

whether they're custom GPT or other

3:51

such assistance or agents. Let's negate them

3:53

from the discussion. They do

3:55

have in production in many cases, off

3:57

-the -shelf kind of agents that they

4:00

built with vendors like AgentForce, Microsoft,

4:02

and others. The other cases

4:04

that we're seeing, and we'll talk more

4:06

about the actual use cases, but of

4:08

course customer support is by far the

4:10

most predominant category of where agents are

4:12

actually mature in production, and

4:14

a handful in some cases

4:16

of other supporting agents in

4:18

some very well -focused functions within

4:20

the company. So I

4:23

think that that bridges this into

4:25

maybe where organizations are not. So the

4:27

second thing that we wanted to

4:29

explore is, again, set up to what

4:31

agents are ready for prime time.

4:33

There's some pretty distinct patterns in what

4:35

we're not seeing. I was on

4:37

a conversation with a group of chief

4:39

AI officers a couple of weeks

4:41

ago at this point, a

4:43

number from the finance industry, some

4:45

from pharmaceuticals. pretty big range of

4:47

different companies. And there was one

4:49

thing that was really clear. And

4:51

these are definitely more advanced, or

4:53

at least these are the most advanced

4:55

portions of their organizations. Maybe the organization

4:58

as a whole is in advance, but

5:00

these are people who are highly engaged,

5:02

highly enfranchised, really thinking about these things

5:04

all the time. They're the internal champions.

5:06

And it's very clear that where they

5:08

want to get eventually is agents involved

5:10

in core business

5:13

functioning, right? If they are

5:15

in insurance, they want

5:17

agents to actually be making

5:19

decisions beyond just sort

5:21

of providing algorithmic advice like

5:23

predictive analytics have done.

5:26

They want to reorganize their

5:28

whole companies around agent

5:30

and capacities. However, to

5:32

a person, none of them

5:34

really feel like agents

5:37

have the complexity, sophistication, duration

5:39

capability, to be

5:41

used for those specific purpose -built use

5:43

cases that are the very, very

5:46

core to their company. And so instead,

5:48

it seems like the place that

5:50

they're going is focusing on not things

5:52

that are unimportant, but just other

5:54

parts of the functioning of the company

5:56

that are, call it lower risk,

5:59

right? That still allow them to get

6:01

used to integrating agents into workflows, but

6:03

are on things like customer service,

6:05

marketing, sales. Is that something that you're

6:07

seeing as well? Yes,

6:09

but I'm not sure about the observation that the

6:11

technology is not ready for what they have

6:13

in mind. I think in many cases, the organization

6:15

is not ready. When it's from

6:18

culture, technology skills use

6:20

cases. In some cases,

6:22

they don't want to tackle

6:24

the most contradictory thing that

6:26

they can do to get

6:28

their employees basically to create

6:30

an uproar because they're seeing

6:32

the future. So that might

6:34

be another thing. There is also a

6:37

fear to take your core business.

6:39

offload it to AI because of

6:41

the potential pitfalls beyond the

6:43

employee. And then lastly,

6:45

technology perhaps is not ready in some cases,

6:47

but I'm not sure whether in all cases

6:49

this is indeed the case that the technology

6:51

is the biggest hurdle. Sure. So

6:53

basically it sounds like you're seeing

6:55

the same sort of inclination towards,

6:57

you know, orthogonal use cases rather

6:59

than, you know, core business function

7:01

use cases. You're just, you're not

7:04

sure that that's more technology or

7:06

the organization itself or some combination

7:08

thereof. Yeah. And in

7:10

many cases, we're saying that other

7:12

people are doing what other people

7:14

are doing. So there is like

7:16

a momentum here of automating the

7:18

support function. First, I think from

7:20

a business perspective, it makes sense

7:23

to dip your feet in the

7:25

water where it's more safe from

7:27

various perspectives and then go there.

7:29

Just yesterday, we had a conversation

7:31

with a company that is very

7:33

bold in their agentic approach, but

7:35

they're saying, let's get the efficiency

7:37

first. off the table and put

7:39

all of these agents that can free

7:42

some of the bandwidths of our employees.

7:44

And then let's tackle the core business

7:46

with agents, not because we think that

7:48

we can, but because we want to

7:50

have our employees have more bandwidths

7:52

to think about those core agents before

7:54

we dive head deep with those. Yep.

7:57

No, it totally makes sense. So, okay,

7:59

then let's move to the meat of

8:01

this conversation, which is what things people

8:03

are doing right now? What are we

8:05

seeing most commonly in terms of the

8:07

agentic use cases that are being deployed,

8:09

that are ready for production, that are

8:11

actually yielding results for companies? Yeah,

8:13

so I can name a few and

8:15

then add whatever I'm missing. But one

8:18

thing that is very straightforward and easy

8:20

to use is basically to use agents

8:22

that others have built. What are those

8:24

are agents for coding that are probably

8:26

one of the most mature, I think

8:28

literally every coding platform now offer an

8:30

agent. Some of them are better than

8:32

others. Some of them are agent native

8:35

versus others that are just introducing agent

8:37

almost as an afterthought. But those are

8:39

getting some good momentum and some positive

8:41

feedback. my

8:43

personal favorite deep research agents that

8:45

are doing amazing job and

8:47

we're also seeing some companies basically

8:49

creating their own version of

8:51

deep research so that it can

8:53

work internally or in their

8:55

own terms and these are some

8:57

very good use cases. The

8:59

other probably simplest thing that people

9:01

are doing is augmenting the

9:04

classical like Zapier or make automations

9:06

with more agentic capabilities, whether

9:08

it's for planning, some open -ended

9:10

tasks that currently agents can do

9:12

and beforehand they couldn't, or

9:14

augmenting them with more like

9:16

NLP -based interactions of text

9:18

and speech, but just adding

9:20

them to the existing flows

9:22

concretely or metaphorically by creating

9:24

similar automations using other tools. Let's

9:27

actually pause I want I want to break

9:29

these apart a little bit because there's there's

9:31

a lot to dig into here So let's

9:33

talk about the augmented automation a little bit

9:36

because they're frankly the least interesting of these

9:38

things to me They're sort of

9:40

the they're a very obvious starting point,

9:42

but they are this is the

9:44

area where people love to debate,

9:46

are these things really agents or not? Like, what

9:48

should we call automations? What should we call agents?

9:50

I'm well on the record with this one that

9:52

I think people should give up the ghost

9:54

and agents are close enough and actually people's understanding

9:56

of that term is directionally correct and they should

9:58

just be fine with it. But it feels

10:01

like this is an area

10:03

where there are certain types of

10:05

tasks that are just, they're

10:07

so begging to be automated more.

10:09

And it's really just figuring

10:11

out these sort of very slight

10:13

customization improvements for existing attempts

10:15

at automation that are, it feels

10:17

likely that these things are,

10:19

you know, going to be completely

10:21

boring wrote and normalized, you

10:23

know, inside of a very short

10:25

period of time. Galileo called

10:28

these the digital assembly line in

10:30

KPMG's taco framework. They called these the

10:32

taskers, right? Think things that are

10:34

very, very specific. I mean,

10:36

how much are, how much are organizations

10:38

getting fired up about these things

10:40

versus they're already in the kind of

10:42

table stakes column? Yeah,

10:44

in most cases I think those

10:46

will be like stuff that the

10:49

employees will individually create for themselves

10:51

and thereby they will not move

10:53

any needle. In some cases though

10:55

you are seeing some business processes

10:57

that even using this method can

10:59

be automated significantly more than what

11:01

they've been doing thus far and

11:03

in those cases you might be

11:05

able to see even a higher

11:07

use cases that are implemented using

11:09

a not very complicated technology. Interesting.

11:12

So basically, there's a risk at

11:14

undervaluing the simplicity just because it

11:16

is simple in terms of its

11:19

potential business impact. Exactly. Yeah.

11:22

Let's talk about deep research encoding for

11:24

a second. Because for my money, I

11:27

think that these might

11:29

be the two agentic augmentations,

11:31

however, agent categories,

11:34

that to me, it is

11:37

very hard now to

11:39

justify doing things that you used

11:41

to do without them, without them today.

11:44

I think that the capabilities of

11:46

research tools are, it's very

11:48

hard for me to see how

11:50

people who, I don't do a

11:52

single thing that involves any sort

11:54

of research without using these tools at

11:56

this point, right? And in general,

11:59

my workflow often involves cross -checking two

12:01

to three to four of these to see

12:03

how they come up with things, right? you

12:06

know, three different versions of

12:08

deep research running at any given

12:10

time. Now, obviously... We're just scratching

12:12

the surface of deep research possibilities because

12:14

the versions that we're using are

12:16

sort of very, you know, individually designed

12:18

agents that don't have access to

12:20

proprietary knowledge bases or, you know, and

12:22

obviously what enterprises are thinking about

12:24

is how to plug those into, to

12:26

other data sources. But it feels

12:28

completely like not, not in six months, not in

12:30

12 months right now. If you are, are

12:32

doing anything that involves sort of research or strategy

12:34

and not using those tools, I tend to

12:36

think you're behind. And I think that it were

12:38

pretty, pretty much in a similar spot when.

12:40

comes to coding. Now, coding is

12:42

interesting because there

12:45

is I think an ironic or

12:47

surprising at least amount of

12:49

intransigence when it comes to adoption

12:51

of the sort of coding

12:53

agent tools and vibe coding tools

12:55

and things like that among

12:57

enterprise developers. And to some

12:59

extent there are pieces of it that

13:01

are understandable, right? Like the first

13:03

generation of these tools that are becoming

13:06

popular, right? The IDEs, the cursor

13:08

and windsurf, the specific text code tools

13:10

like Bolt and lovable, they're absolutely

13:12

optimized right now for an individual's sort

13:14

of developer experience as opposed to integrating

13:16

and interacting with these massive legacy

13:18

code bases that have thousands of people

13:20

working on them and a guy

13:22

who might be typing on them the

13:24

next day someone different is using

13:26

that code. But it still feels basically

13:28

criminal at this point to not

13:30

be taking advantage of these sort of

13:32

new efficiencies of these coding tools.

13:34

I mean, we at Super have, we

13:36

had to let developers go who

13:38

wouldn't get with a picture basically to

13:40

change their processes around them. Is

13:43

there at what point do organizations

13:46

just start to mandate that these

13:48

are now the way that you

13:50

do things is, you know, if

13:52

you are not agentically augmented in

13:54

these categories, you're just you're just

13:56

out. You're too far behind. Yeah,

13:59

you know, you're very savvy in

14:01

that. But when we talk to

14:03

organizations, there are so much behind

14:05

you. And a lot of that

14:07

comes from what you're saying that it's very

14:09

optimized for the individual user. But another

14:11

part of it, which is probably the lowest

14:13

hanging fruit, they just don't know the

14:16

breadth of possibilities that these tools can take

14:18

you. Because deep research, for example, people

14:20

get stuck on the deep research part and

14:22

forget that it's just like a deep

14:24

reasoning agent that could literally get you any

14:26

complex task done much better. than any

14:28

other AI tool. So they

14:31

understand that. Similarly with coding, like when

14:33

I'm talking to engineers of... on

14:35

the stack, low level, hardware, high

14:37

level, in many cases, they just didn't

14:39

spend enough time to understand all the

14:41

use cases that they can do with

14:44

them, and that's why they're not using.

14:46

So I agree with you that they

14:48

should, but they need to spend some

14:50

more time to get comfortable with the

14:52

existing tools, and then they can augment

14:54

them and modify them to be like

14:56

enterprise ready for their needs. Today's

15:00

episode is brought to you by Plump. If

15:02

you're building agentic workflows for clients or colleagues,

15:04

it's time to take another look at Plum. Plum

15:07

is where AI experts create, deploy, manage,

15:09

and monetize complex automations. With features

15:11

like one -click updates that reach all

15:13

your subscribers, user -level variables for personalization,

15:15

and the ability to protect your prompts

15:17

and workflow IP, it's the best

15:19

place to grow your AI automation practice.

15:21

Serve twice the clients in half the

15:23

time with Plum. Sign up today

15:26

at useplum.com, that's U -S -E -P -L -U -M

15:28

-B dot com forward slash N -L -W.

15:31

Today's episode is brought to you by

15:33

Vanta. Vanta is a trust

15:35

management platform that helps businesses automate security

15:37

and compliance, enabling them to demonstrate

15:39

strong security practices and scale. In

15:42

today's business landscape, businesses can't just

15:44

claim security, they have to

15:46

prove it. Achieving compliance with a

15:48

framework like SOC2, ISO 2701,

15:50

HIPAA, GDPR, and more is how

15:53

businesses can demonstrate strong security

15:55

practices. And we see how much this

15:57

matters every time we connect enterprises with

15:59

agent services providers at Superintelligent. Many of

16:01

these compliance frameworks are simply

16:03

not negotiable for enterprises. The

16:05

problem is that navigating security and compliance

16:07

is time consuming and complicated. It

16:10

can take months of work and use up valuable time

16:12

and resources. Venta makes it easy

16:14

and faster by automating compliance across 35 plus

16:16

frameworks. It gets you audit ready in weeks

16:18

instead of months and saves you up to

16:20

85 % of associated costs. In fact,

16:22

a recent IDC white paper found

16:24

that Vanta customers achieved $535 ,000 per

16:26

year in benefits, and the platform pays

16:28

for itself in just three months. The

16:31

proof is in the numbers. More than 10

16:33

,000 global companies trust Vanta, including

16:35

Atlassian, Quora, and more. For a

16:37

limited time, listeners get $1 ,000

16:39

off at Vanta.com slash NLW. That's

16:42

V -A -N -T -A dot com

16:44

slash NLW for $1 ,000 off.

16:48

Hey listeners, want a supercharger business

16:50

with AI? In In

16:52

our fast-paced world, having a

16:54

solid AI plan can make

16:56

all the difference. Enabling organizations

16:58

to create new value, grow,

17:00

and stay ahead of the

17:03

competition is what it's all

17:05

about. KPMG is here to help

17:07

you create an AI strategy that really

17:09

works. Don't wait, now's the time

17:12

to get ahead. Check out real

17:14

stories from KPMG of how AI

17:16

is driving success with its clients

17:18

at KPMG. US. US slash AI.

17:20

Again, back to the show. I

17:23

do think that you're right that so with deep

17:25

research, the terminology, the name

17:27

actually, even though it's. sort of

17:29

been universally adopted, right? It's called

17:31

deep research for both open AI

17:33

and Gemini. It's called deep search

17:35

for GROC. It potentially distracts people

17:37

a little bit from the full

17:39

set of possible use cases by

17:41

being called research, right? I also

17:43

think that by virtue of being

17:45

embedded in these other tools, as

17:48

opposed to a standalone thing that

17:50

was introduced as a breakout kind

17:52

of standalone thing, it's perhaps being

17:54

underappreciated in terms of just how differentiated

17:56

it is. We at Superintelligent

17:58

have been among the companies, and

18:00

I've talked to lots of people who have had this

18:02

experience, who spent a bunch of time trying to

18:04

build our own kind of system to Wire

18:06

a bunch of things together only to just see

18:08

if we could run it through deep research

18:10

and have it produce way better results And you

18:12

know that we need we need some like,

18:14

you know 90 letter German word for just do

18:16

it with deep research instead of trying to

18:19

build it yourself But I but I think that

18:21

there is still we're just scratching the surface

18:23

of those use cases and it's gonna take some

18:25

amount of time of Diffusion of people sharing

18:27

their specific uses of deep research for it to

18:29

fully embrace, you know on the on the

18:31

coding side I think that you're right. I think

18:33

that it's going to change rapidly. I think

18:35

that, you know, what, what, what I'm seeing is

18:37

you're starting to see more discussion

18:39

even within the enterprise around what things

18:41

these tools can be used for

18:43

right now, right? So, uh, you don't

18:45

want to use it as your,

18:47

as your sort of primary coding environment,

18:49

but you should deploy these things

18:51

for refactoring or whatever it is, right?

18:53

That you started to be more

18:55

discreet about it. You're also seeing just

18:57

an absolute flood of companies race

19:00

in to try to fill the current

19:02

gaps in. capability and new challenges

19:04

that these tools are arising from, right?

19:06

So you're seeing on the consumer side, you're

19:08

seeing companies that are coming in to try

19:10

to make it easier to go from, okay,

19:12

I've got this code base that I don't

19:14

understand. How do I actually, you know, make

19:16

it live on the internet and do things?

19:18

There are companies that are coming in and

19:20

doing that. On the enterprise side, you're absolutely

19:22

seeing companies that are trying to come in

19:24

and start to maximize for those enterprise use

19:26

cases, even though they're more complex. So I

19:28

think that that's going to change. pretty quickly. Anything

19:31

else on those before we move to sort

19:33

of like the big 800 pound gorilla

19:35

in terms of capabilities or things that people

19:37

are doing now with customer support? The

19:40

only thing that we need for Super

19:42

is to have the deep research on

19:44

API. So if someone in the decision

19:46

-making process can create a very good

19:48

API -able deep research, the better. Yeah.

19:51

Dear OpenAI, I know you some of you

19:53

guys are listening. Please let me know when

19:55

the API is coming. All right,

19:57

so let's talk about customer

19:59

employee support as you know that I

20:01

think probably the area that that is most

20:03

discussed when it comes to agents Yeah, so

20:05

we talked about it before that customer

20:07

support is probably the most mature agentic

20:10

use case out there But there is

20:12

an abundance of flavors for a customer

20:14

support right we're talking from all

20:16

the way from a simple,

20:18

very like FAQ kind of

20:20

an agent, all the way

20:22

to the very impressive, completely

20:24

autonomous end -to -end customer support

20:26

agents. We're also talking about

20:28

other flavors of that that

20:30

can be agents that are

20:32

helping to upsell or cross

20:34

-sell your product because they

20:36

identify opportunities. So we're starting

20:38

to see these implementations. We're

20:40

also seeing similar notions in

20:42

internal employee support, whether it's

20:44

IT support, HR support, legal

20:47

support, payroll, basically everything

20:49

that requires someone to answer

20:51

questions in various capacities. These

20:53

are perfect agents. And

20:55

you can even kind of extend those

20:57

to other types of support in the

20:59

broad sense of the term, whether it's

21:01

to help with employee learning and development

21:03

or on board new employees. In

21:05

many cases, these are the most

21:07

prime time ready agents that we have

21:09

out there. And lastly, and you

21:11

will probably claim that that's a category

21:13

on its own is everything related

21:16

to outbound communication using various voice agents

21:18

to create more sales or maybe

21:20

reach out to candidates that we

21:22

want to hire and so on.

21:25

Yeah, I do think that those are, I think

21:27

there's a couple different distinct categories there. I

21:29

think that the part of the

21:31

reason though that you might want

21:33

to connect them is that all of

21:35

these have a common

21:37

thread of talking in air

21:40

quotes to a person. or

21:42

finding out some information about them or from

21:44

them, and then

21:46

integrating that with some pre -existing set

21:48

of information. And

21:50

this is just all the versions of

21:53

that AI is really good at

21:55

right now. And so

21:57

you're seeing this Cambrian

21:59

explosion of basically every type

22:01

of that interaction. that AI can

22:03

do. So let's talk about voice

22:05

agents for a minute. I think

22:07

that part of why voice agents

22:09

are such a hot category is

22:11

that this is a capability that

22:14

is really useful right now. I

22:16

mean, this is something that we

22:18

observed. Like part of where our

22:20

voice agent interviewing came

22:22

from was observing that.

22:25

in other areas voice agents actually were

22:27

doing a pretty good job right so i

22:29

was looking over at the hiring space

22:31

where companies were already deploying voice agents to

22:33

do you know initial screening interviews and

22:36

things like that and they were working pretty

22:38

well. The voice capabilities are good. Advanced

22:40

voice mode had come out, so latency was

22:42

better. All of these things had kind

22:44

of come online as capabilities. And

22:46

we had the thought, well, maybe you could

22:48

basically turn that into a consultant whose job

22:50

is just to sort of ask the right

22:52

questions and grab a bunch of information. And

22:54

it turns out that a hundred other

22:57

startups or a thousand other startups are

22:59

basically going through that same process

23:01

of thinking through every other version of

23:03

asking people questions. So you have

23:05

voice agents for market research. You have

23:07

voice agents. for, you name it,

23:09

right now they're coming out. And so

23:11

I think that it's hard to

23:13

call voice agents aren't so much a

23:15

category as much as sort of

23:17

a common underlying technology capability that begets

23:20

lots of categories. But I think

23:22

that as companies are thinking about where

23:24

they could be getting value from

23:26

agents right now. It is

23:28

not unreasonable to ask what are

23:30

current functions that involve us

23:32

talking to people, you know, literally

23:34

talking to people and would any of them be

23:37

well suited for, you know, one of the copious

23:39

number of voice agents that are out there or,

23:41

you know, rolling our own version of that. And

23:43

there is so much value if

23:45

you combine the voice agent with

23:47

the deep research or deep analysis

23:49

agents, then you get even a

23:51

full blown consultant that is autonomous,

23:54

basically. I mean, that's, we call the

23:56

sort of underlying technology behind, or at least I

23:58

call it, I don't know if you'd call it

24:00

this, but I call the underlying that

24:02

we use for the agent readiness audit

24:04

an agent consultant engine, because that's what

24:06

it feels like, right? It's job is

24:08

to ask the right question, obviously with

24:10

us, you know, helping kind of give

24:12

it some initial ideas about what the

24:14

right questions are. And then, you know,

24:16

do come up with and do analysis

24:18

on the basis of some particular goals

24:21

and particular knowledge, and which, you know,

24:23

starts to look very prox... what consultants

24:25

do. I think it's worth mentioning

24:27

briefly too before we sort of

24:29

broaden out again the sales agent

24:31

use case. This is one

24:33

to me that feels very

24:35

much again like... I don't know

24:37

that I've ever run across, or at

24:39

least in the last six months, run

24:42

across any sort of sales organization that

24:44

couldn't take advantage of the sales agent,

24:46

the sales type, SDR type agents that

24:48

are available right now. Now, that's not

24:50

to say that they are, you can

24:52

just grab one off the shelf and

24:54

it's instantly good to go. There is

24:57

more work than I think people might

24:59

imagine or might want when it comes

25:01

to getting their SDR agents up

25:03

and running. However, Sales

25:06

is an area

25:08

where there is no risk,

25:10

I don't believe whatsoever, of

25:12

there's always more leads. You always

25:14

want more potentials. If a sales agent,

25:16

like a human sales agent or

25:19

sales representative had access to an agent

25:21

that could get them 10 ,000 times

25:23

the number of leads, they would

25:25

be nothing but thrilled because ultimately, more

25:27

and more is the goal. And

25:29

so I think that one, from

25:31

an internal change management

25:33

perspective. Sales is a really good area where

25:35

it seems highly unlikely to me that

25:38

we're going to see big cuts in the

25:40

sales organization because of agents. We're going

25:42

to be straight, not in efficiency AI, but

25:44

opportunity AI, where it's just how much

25:46

more can we do? How much faster can

25:48

we grow? How much bigger can we

25:50

get? And I think that that's going to

25:52

be a very useful bridge as, or

25:54

employees try to figure out what management's goals

25:56

are as it relates to agents. But

25:58

two, they're also kind of an area

26:00

where we're starting to get a little bit of

26:02

a preview. of the future. you know, Lindy's

26:05

swarms came out recently. And

26:07

swarms are basically agents that beget other

26:09

agents, at least in Lindy's case, where

26:11

they start to do parallel processing, right?

26:13

So instead of it being an agent

26:15

that's doing, you know, data enrichment

26:17

around a particular lead at a time,

26:19

it's, you know, an agent that's fragmented itself

26:21

into a hundred different agents that's doing

26:23

data enrichment across a whole set of lookalikes

26:26

all at the same time. And it's

26:28

all just efficiency. It's how much more can

26:30

it get done in a given

26:32

period of time? And Again, organizations are

26:34

just starting to put these systems online. I

26:36

think that there's still a lot of work, a

26:38

lot of customization that's necessary. But

26:40

I do think that to the extent that, again,

26:42

companies are looking for a place where they

26:44

can dive in right now, get

26:46

their hands dirty, and probably

26:49

get some pretty clear ROI

26:51

from agents. Sales and SDR

26:53

type agents are a pretty good place to look. Yeah,

26:56

I agree and also probably

26:58

out like personalized outreach marketing

27:00

is the same methodology. Yep.

27:02

Yeah, I think it's sort of

27:04

the same bucket. I will say

27:06

that I I am less convinced

27:08

around marketing content in general. I

27:10

still think that there is a

27:12

fairly meaningful gap in quality between sort

27:14

of like copy that's going to come from

27:16

agents right now and copy that comes

27:19

from people. Not because agents are bad at

27:21

writing or anything, but this is just

27:23

it's an area that involves so much taste

27:25

and so much agency and so much.

27:27

you know, knowledge and experience that you don't

27:29

even realize that you have, but you

27:31

happen to notice that, you know, people responded

27:33

to this one word in a tweet

27:35

one time in a way that made you

27:37

never want to use that word again.

27:39

You know, whatever. It's the closest where there's

27:41

actually still a big meaningful gap. I think

27:44

that that'll change over time. I again, think

27:46

that swarms are going to be part of

27:48

the answer where we run lots and lots

27:50

of scenario planning, maybe testing. Yeah, exactly. War

27:52

game type campaigns with marketing. But let's talk

27:54

then about let's zoom back out for just

27:56

a minute. as we close out here and

27:58

talk about, given what

28:01

organizations aren't doing, what

28:03

the challenges are, and then what they

28:05

are doing, what should orgs do? What

28:07

should they be thinking about right now?

28:09

What constitutes a reasonable, a realistic, a

28:12

successful, agentic approach in this particular moment

28:14

based on where we actually are? All

28:16

right. So first, let's make sure

28:18

that they overall improve their agent

28:21

readiness across the board. We talked

28:23

about a little bit when we

28:25

talked about the seven mistakes, but

28:27

get your docs in order, your

28:29

culture, your strategy, have the skills

28:31

in place, have your tech stack

28:33

ready, have your agent infrastructure ready. That's

28:36

like the first and foremost thing

28:38

that you should do. Anything to add

28:40

there? I agree. I think

28:42

that there's a temptation when it comes

28:44

to agents. to think that

28:46

what we should be doing, the entirety of what

28:48

we should be doing is picking an agent and

28:50

deploying it. And I think that that's a big

28:52

part of it, but there is so much infrastructure,

28:54

new capabilities, new thinking that needs to be done

28:56

around it. And a lot of that work actually

28:58

can be done more successfully even than some of

29:00

these deployments in the here and now. Yeah.

29:03

And then the second part and the

29:05

most important at least in the

29:07

context of this conversation is to really

29:09

have a prioritized list of egenic

29:11

use cases. And you should probably have

29:14

a list that is much more

29:16

comprehensive than what you can probably physically

29:18

do or from a feasibility perspective

29:20

do right now. But if you start

29:22

having this list and then you

29:24

go at it one by one, you

29:26

will probably be much better off

29:28

than just selecting one kind of opportunistic

29:30

one and going from there. I

29:33

will encourage you is to have

29:35

this list very much prioritized, not just

29:37

by value, because in many cases,

29:39

like we talked before, the highest value

29:41

agents are the ones that you

29:43

will probably not want to tackle at

29:45

the beginning, but some kind of

29:47

a weighted prioritization between the value, the

29:49

feasibility, and basically the cost. And

29:52

then once you have this prioritized

29:54

list, you can go and start executing

29:56

them. And we can help you

29:58

figure out some of the indications for

30:00

what might be a good use. So

30:04

first of all, in terms of a good

30:06

use case, a good place to

30:08

start is to see what others

30:10

in a similar industry are doing. I

30:13

want to provide one caveat here, like

30:15

we talked before, that sometimes that can create

30:17

a bias of you overdoing again and

30:19

again what others have already done. And in

30:21

many cases, that's not the only thing

30:23

that you can do. The other

30:25

thing that you can try and look for

30:27

are use cases in your business that might

30:29

be a good candidate for agents. and I

30:31

have like a few pointers to provide

30:34

you in order to identify a good

30:36

use case for agents. Some of them

30:38

are more trivial like first and above

30:40

all, it should be something that is

30:42

being already done on

30:44

a computer, digital use cases

30:46

is good for agent. Second,

30:49

think about use cases where you need

30:51

a lot of specialization, but the humans

30:53

are a bottleneck. So your agent's worms

30:55

are a good example. Ideally,

30:57

you would have so many sales persons

30:59

that you will outreach to anyone and

31:01

everyone and follow every lead, but

31:03

you just don't have enough sales

31:05

people. So that's an example of

31:07

a specialization. Another example, very classical

31:10

one, is legal stuff, an

31:12

agent that goes over contracts. Often you

31:14

just don't have access to enough

31:16

lawyers, or it's very expensive. So that

31:18

will be a good example of

31:20

a use case that require personalization. Other

31:22

places that are very good candidates

31:24

are cases where you need a

31:26

lot of availability 24 -7, so of

31:28

course customer support, but that also

31:31

covers all the employee support or

31:33

think of any cases where perhaps

31:35

you want to do better by

31:37

your customers, but you currently can't,

31:39

either because you're a small company

31:41

or just don't have these services

31:43

and consider agentifying them. Another

31:45

thing that I can offer is

31:47

cases where you need to have a

31:49

lot of personalization. And you don't

31:51

have the demand power to personalize, so

31:53

marketing we can debate, but there

31:55

are other cases where you

31:57

want to handle each and every

31:59

individual separately. And then

32:01

a few additional things that I

32:03

can think about is what about cases

32:05

where the more data you have,

32:07

the better the agent will behave. These

32:10

are often places where when I am

32:12

sending you to create your infrastructure, these are

32:14

the places where you should consider. So

32:16

these are also relevant places.

32:19

Another thing that I'm often asking

32:21

our customers and you always say,

32:24

don't just ask about that, but

32:26

that's important is cases where people

32:28

are disliking what they do for work

32:30

because it's repetitive, tedious, and

32:32

they would have liked to do other stuff. And

32:35

last two that I always kind of look

32:37

for in a good agent use case is

32:39

where the process is well defined. The business

32:41

process is very clear. There is a very

32:43

clear set of policies by which decisions should

32:45

be made. And lastly, and

32:47

connected to that is when we can

32:49

measure the output of agents. And that's

32:51

why perhaps coding is such a good

32:53

use case for that, because we can

32:55

measure whether the code is functional or

32:57

not. But think of other

33:00

places. If you're able to differentiate between a

33:02

good and bad outcome, that might be a

33:04

very good use case for agent. What

33:06

type of sort of next steps should

33:08

people be thinking in terms of should they

33:10

be, you know, getting together committees to

33:12

start making decisions differently? Should they just be

33:15

throwing themselves into a first test case?

33:17

How much should they be focused on, you

33:19

know, the infrastructure build out versus just

33:21

actually getting their feet wet with agents right

33:23

now based on what's available? So

33:25

it's going to be, and it depends kind

33:27

of an answer, if they had all the

33:30

resources in the world, so all of them

33:32

create a list of use cases, invest in

33:34

infrastructure, and simultaneously start working on pilots of

33:36

agents, because that's the best way to learn

33:38

and augment both the infrastructure and the use

33:40

cases. If they don't have

33:42

all the resources in the world, I

33:44

would probably be very opportunistic and

33:46

say deploy your first agent and learn

33:48

from that. That's my take.

33:51

Do you believe something else? I

33:53

know I'm with you. I think that

33:55

there is no substitute for the hands -on

33:57

learning that comes by actually getting in

33:59

there and understanding capabilities. I

34:01

also think that it naturally is

34:03

going to be get You're

34:05

going to figure out all the things that

34:07

you don't have in place as you go

34:09

try to actually deploy an agent, right? Your

34:11

if your data is not ready, you're going

34:13

to have to deal with that to get

34:15

that agent ready. If you start to run

34:17

into issues of decision making, maybe that's that's

34:19

sort of what prompts you to think about

34:21

kind of guardrails and, you know, governance more

34:23

broadly. So I kind of think that by

34:25

going after actually doing the thing, right, starting

34:27

by starting, you're likely to have all the

34:29

other pieces come along with it. Yeah,

34:32

I agree. and end up overhauling your

34:34

entire tech stack. Yeah, exactly.

34:36

And get ready for a lot of change

34:38

in a very short period of time. Nufal,

34:41

always awesome to have you on the show. Thank

34:43

you for this. We'll come back and do another, you

34:45

know, what people are doing with agents in six

34:47

months. I expect it'll be very, very different than what

34:49

we're talking about today.