0:00

Microsoft Unified Support busting your IT budget?

0:02

Well, you're not alone. US Cloud

0:04

is the global leader in third

0:06

-party Microsoft Enterprise Support, and they're

0:08

helping enterprises slash cost by 30 %

0:10

to 50 % versus Microsoft Unified, all

0:13

while getting faster response times from

0:15

domestic engineers. No offshoring, no

0:17

long waits, just expert support

0:19

you can count on, 24x7,

0:21

365. Trusted by 50

0:23

of the Fortune 500 companies and

0:25

offering senior -level engineers with an average

0:27

of 14 years experience with Microsoft

0:29

technologies, US Cloud is the smarter

0:31

alternative for premium Microsoft support without

0:33

the premium price. Ready to make

0:35

the switch? Book a call at

0:37

uscloud.com to get started today. That's

0:39

uscloud.com. Cloudcast Media

0:41

presents from the massive studios

0:43

in Raleigh, North Carolina. This

0:46

is the Cloudcast with Aaron Delb

0:48

and Brian Grace Lee, bringing you the

0:50

best of cloud computing from around

0:52

the world. Good

0:56

morning, good evening wherever you are and welcome back

0:58

to the cloudcast we are coming to you

1:00

live from the massive cloudcast studios here in Raleigh,

1:02

North Carolina Hope everybody is doing well Apologize

1:04

to those of you who are used to me

1:06

getting the show out first thing in the

1:08

morning In your inbox in your podcast feeder out

1:10

in YouTube on a Sunday morning had a

1:12

lot going on this weekend So get this one

1:14

out just a little later on a Sunday

1:16

morning, but apologize for that. Hope everybody's doing well.

1:18

Hope everybody Who wanted to get drafted in

1:20

the NFL this weekend got drafted. Hopefully those of

1:22

you that didn't want to get drafted, your

1:24

team picked the best player you could possibly get,

1:26

Future Hall of Famers and all those

1:28

sorts of things. So anyway, I hope everybody's

1:31

doing well. Spring is moving along very

1:33

quickly. We are almost into May. Hard to

1:35

believe the year is that far along.

1:37

But anyways, one of the things I've been

1:39

thinking about a lot lately is, you know, there's

1:41

a lot of buzz around AI agents. I know

1:43

we've talked about it a little bit on this

1:45

show. We've had a number of shows here recently.

1:47

We've kind of dove into it. And,

1:50

you know, as certain things

1:52

are starting to solidify, because

1:54

we've already seen people building

1:56

AI capable things in all sorts of different

1:58

languages. There's lots of different frameworks for

2:00

building agents, but we're now starting

2:02

to see. Some things evolve

2:05

even to the last couple of weeks,

2:07

last couple of months. Things like MCP,

2:09

Model Context Protocol, Google just

2:11

came up with something called

2:13

A2A, which is another model

2:15

to, you know, tools, agent

2:18

to tools, agent to agent protocol for

2:20

doing this. So we're seeing a lot

2:22

going on from a technology in this

2:24

space. And I got to thinking

2:26

about it, I was reading a bunch of different

2:28

articles and different things and not all of them

2:30

technical, some of them just kind of architectural and

2:32

so forth. And it got me thinking

2:34

about AI agents. And so

2:36

what I did was in sort of

2:38

short form to help me kind of think

2:40

about what's going on in this space,

2:42

what are the things that companies and end

2:45

users are going to have to be

2:47

thinking about. I kind of made up five

2:49

topics, five questions that I'm thinking about

2:51

with AI agents. And I tried to put

2:53

them in. Kind of classic newspaper writing

2:55

for those of you kids newspapers were things

2:57

that used to come to your house

2:59

with all the news now It's just called

3:01

you know the internet, but you know

3:03

kind of classic, you know who what why

3:05

what what where when and how? Sort

3:08

of walking through these things and so what

3:10

I thought I would do is just kind of

3:12

go through those Kind of you know think

3:14

through some of the things that are going on

3:16

with this No, this isn't necessarily like a

3:18

have all the answers type of things, but it

3:20

is a you know as we are starting

3:23

to create things that

3:25

interact with other things in ways

3:27

in which, you know, we

3:29

use them to either augment or,

3:31

you know, replace or create

3:33

new experiences that are supposed to

3:35

look like humans doing certain

3:37

types of work. They got me

3:39

thinking about, you know, like how are we going

3:41

to kind of walk through these things? How are

3:43

we going to kind of deal with them? What

3:46

things have to live around them to be able

3:48

to provide protections and guardrails and all those sort

3:50

of things? So

3:52

anyways, I thought I'd kind of walk through some

3:54

of those today This won't be a real real long

3:56

show just kind of you know a set of

3:58

links that I found recently That I thought were sort

4:00

of interesting as well as just some questions that

4:02

I'm starting to ask myself and would love to kind

4:04

of hear people's feedback. You know, if you're beginning

4:06

to kind of go down the agent path, what are

4:08

you doing with it? What have you learned? What's

4:10

working well? What's not working well? All those sorts of

4:12

things. Feel free to shoot us a note show

4:14

at the cloudcast .net. We're always available there. Or if

4:16

you want to hit us up on one of the

4:18

social media channels, you can hit us up there

4:20

as well. All the links are at the bottom of

4:22

the show notes. So first

4:24

one I had is what I put

4:26

under who and what that one that

4:29

my question is or my kind of

4:31

topic I'm thinking about is You know,

4:33

how, how is an agent, an AI

4:35

agent represented in the context of a

4:37

person or a group or an organization?

4:39

Like, how are we going to think

4:41

about this thing which is going to

4:43

take autonomous actions? It's going to, you

4:45

know, do research. It's going to ask

4:47

questions. It's going to get back information.

4:50

It's going to do further sort of

4:52

prompt engineering and so forth. Do we

4:54

think about that as a person? Do

4:56

we think about that as a group?

4:59

Do we think about that as an organization? And

5:01

the reason I mentioned all those

5:03

things is, you know, that's going

5:05

to get us into stuff around,

5:08

you know, things like authentication or

5:10

authorization around permissions, around the output

5:12

of it, you know, kind of

5:14

attributing that back to something or

5:16

someone. And I'm curious, you know,

5:18

what, you know, what the industry is going

5:20

to start doing in terms of, are we

5:22

going to have agents that look, you know,

5:24

as if they're sort of individuals or we're

5:26

going to have agents that look as if

5:28

they are a group. you

5:31

know, are we going to see in situations

5:33

in which maybe an agent begins to create

5:35

some things and then the output of that

5:37

gets sort of fed up the chain to

5:39

a, you know, kind of a group of

5:41

agents or something like that. So, you know,

5:43

and this is where I'm kind of, I

5:45

think I'm thinking about it very hierarchically in

5:47

terms of the way that group, you know,

5:49

that people organize themselves within companies and so

5:51

forth. And I'm wondering like, is that going

5:53

to be a good way of thinking about

5:55

this or is that going to be sort

5:57

of an old way of thinking about it

5:59

that you know, there is a completely new

6:01

way of thinking about these things. So anyways,

6:03

that's the first thing that kind of came

6:05

to mind in terms of who is, you

6:07

know, do we think about this as if

6:09

it is, you

6:11

know, a pseudo individual,

6:14

a digital individual, or is there a

6:16

very different way to sort of

6:18

think about what an agent actually is?

6:21

Second thing I have my list sort

6:23

of falls in the, in the

6:25

what category is how will we audit

6:27

agent created data or content? Right

6:29

you know what we needed you know

6:31

what we need to do things

6:33

like labeling certain content as agent you

6:35

know created and the reason I'm

6:37

kind of going down this path is.

6:40

You know as we think about more and

6:42

more things that are being created as

6:44

a result of these agents going off and

6:46

doing whatever they're doing reasoning trying to

6:48

do certain tasks changing things together. You

6:51

know how will we sort of attribute

6:53

that to something right so for example

6:55

if I read an article on the

6:57

internet. For the most part, it's got

6:59

something attributed to it. You know, this

7:02

was the author. This was the group

7:04

that wrote this. Maybe you can figure

7:06

out if a grant or some academic

7:08

paper, like how was it funded, those

7:10

types of things. And I'm wondering, you

7:12

know, will we need to, in order

7:14

to help people have a sense of

7:16

trust or have a sense of understanding

7:18

of the information, will we need to,

7:20

you know, sort of attribute it to

7:22

something? There's plenty of stuff on the

7:24

internet that's just kind of randomly out

7:26

there. And again, maybe a lot of

7:28

that's bots as it is today. But

7:31

especially if we think about, hey,

7:33

the thing that was written, it may

7:35

come out as an opinion, but

7:38

it may try to present itself as

7:40

factual. Is it 100

7:42

% hallucination free? Is it

7:44

90 % hallucination free? I

7:47

just think about as we create things,

7:49

especially in a business context in which

7:51

other people might rely upon it. you

7:54

know, how do we go about

7:56

attributing it to where it came from

7:58

as opposed to just, it just

8:00

becomes something that goes into something else,

8:02

right? So, you know, even if

8:04

it's just simple as the way we

8:06

do attributes when we write reports,

8:08

you know, oh, this section was written

8:10

by this group or this is

8:12

the source of the information so that

8:14

if you want to, you know,

8:16

sort of validate it, it's there, right?

8:18

So will we... we do that? And

8:21

again, this sort of goes back to the first

8:23

question of like, how will we attribute that? Does that

8:25

get attributed to sort of a pseudo individual type

8:27

of user? Does it become a person? Does it become

8:29

a group? Does it become an organization? And

8:32

I think there's going to be, you

8:34

know, certain areas in which that needs

8:36

to be identified or at least labeled

8:38

in order for people to have trust

8:40

in it, to be able to use

8:42

it as a building block for something

8:44

else. So on and so forth. Okay.

8:47

Third thing I had on my list

8:49

and I think I got these who

8:51

what where when and who what I

8:53

don't have a where I'm not gonna

8:55

worry about the where cuz I think

8:57

the where is is You know, it's

8:59

the internet Third one on my list

9:01

was why and I guess my question

9:04

is you know why we choose an

9:06

agent as the implementation over another form

9:08

of working on a problem right so

9:11

you know, we're beginning to see some things come along and

9:13

I'll put some links in the show notes. You

9:15

know, we're seeing certain companies sort of

9:17

forward -looking companies who are saying, we,

9:21

you know, kind of top -down edict memo,

9:23

whatever it might be, kind

9:25

of pushing the company to say, look, we need to

9:27

be more aggressive in terms of how we're using

9:29

AI. And I know there's one, it's

9:31

either from Airbnb or Spotify up to look it up. I'll

9:33

get the link to it, put it in the show notes, where

9:36

essentially they're their manager said, hey, we are

9:38

not going to hire any new people, or

9:40

at least I'm going to put

9:43

a criteria on any hiring a new

9:45

people until you've identified that you

9:47

can't work through this with AI in

9:49

some way, shape, or form. In

9:51

essence, saying we're not going to

9:54

hire another net new person until

9:56

you've proven that the thing the

9:58

person would do is beyond the

10:00

scope of what you could do

10:02

with an AI agent or a

10:04

companion or something with you. And

10:07

I think it's gonna be very interesting

10:09

to sort of kind of go through that

10:11

because that's the type of You know

10:13

thinking that that I think does permeate a

10:15

lot of leadership sort of thinking around

10:17

stuff is you know How do we go

10:19

about how do we go about doing

10:21

this? How do we take advantage of it?

10:24

You know are there benefits to you

10:26

know having more agents than people or those

10:28

sorts of things but I'll be very

10:30

interested to sort of understand the criteria that

10:32

people have or companies have or groups

10:34

have for when when they decide that that

10:36

the problem scope should be sort of

10:38

immediately looked at from an, you know, engaging

10:41

AI versus, you know, again, sort of

10:43

an augmenting type of thing. So again, at

10:45

this, don't think about this as the

10:47

thing I talked about last week, which is

10:49

sort of augment versus replace, but this

10:51

is more like augment or additive, right?

10:53

So, you know, add an agent

10:56

as opposed to adding a person, right?

10:58

So not talking about replacing a

11:00

person, but we're talking about adding, you

11:02

know, being additive to it. So

11:04

I'll be very interesting to interested to

11:06

see some things get written about,

11:08

you know, why companies chose or when

11:10

companies chose to bring an agent

11:12

into the fold, right? What was the

11:14

thought process? Are there certain tasks

11:16

that just sort of automatically go into

11:18

that area? Are there certain things

11:20

that are kind of gray? So

11:22

that I'll be very,

11:24

very interested to see where

11:26

that goes. Next

11:29

one I had my list is what

11:31

I'll call when. I don't if I have

11:33

these in order, but when. And

11:36

I was thinking about this more and

11:38

more because as we think about agents, the

11:42

number of tasks that you could start having,

11:44

and I'm starting to see these little write

11:46

-ups about companies who have two or three

11:48

people and 40 agents doing things. And again,

11:50

some of those, I wonder if they're actually

11:52

valid or not, if they're just sort of

11:54

clickbait types of things. But in theory, agents

11:57

are nothing more than just computer

11:59

systems and they don't need sleep, they

12:01

don't need food, they just need

12:03

power and compute and internet connections. The

12:06

number that you could have working

12:09

on certain tasks could be a very

12:11

large number. And also,

12:13

depending on how we identify

12:15

them as people, group,

12:17

organization, whatever they might be,

12:20

it feels like there could

12:22

be a lot of granularity types of

12:24

things and permissions things that don't necessarily

12:26

align with, you know, that human way

12:28

of thinking about stuff. And again, this

12:30

is me trying to go through and

12:32

being like, I'm thinking about these

12:34

as, again, either augment, replace or additive to

12:36

humans. But I don't know that that's necessarily

12:39

the right way of doing that. I mean,

12:41

I'm kind of trying to explore that. But

12:43

I think about that in

12:45

the context of our authorization

12:48

systems is, you

12:50

know, How are they going to be

12:52

capable enough to do this? Because

12:54

a lot of our authorization systems sort

12:56

of authorize people to do certain

12:58

things, they get grouped into certain things,

13:00

and then we let other access

13:02

lists and permission lists and firewalls and

13:04

all sorts of things kind of

13:07

determine what a human can do, what

13:09

actions can they take based on

13:11

this login criteria, this ID, that type

13:13

of thing. And, you know, I

13:15

know we've had some things on the

13:17

show and we talked to the

13:19

folks off Z about this. But

13:22

I do think it's going to

13:24

be very interesting to sort of see

13:26

what has to evolve from the perspective

13:28

of AI agents and how granular they

13:30

get or how broad they get as

13:32

to, you know, are the current authorization

13:34

systems or, you know, ID management systems

13:36

or whatever it might be, are

13:39

they up to speed as to,

13:41

you know, what what's going to be

13:43

needed for this and where are

13:45

going to be those places where systems

13:47

can either get exploited or maybe

13:49

there's so much going on in terms

13:52

of agent to agent communication or

13:54

agent to service or agent to tools,

13:56

whatever that might be, that we

13:58

overwhelm the systems and we're going to

14:00

need something that scales better. So

14:02

I do wonder if the systems

14:04

we have in place today or some

14:07

of the technology that's pretty prevalent

14:09

today will be viable or whether or

14:11

not we're going to need something

14:13

that is very, very different, just because,

14:15

again, I think we

14:17

think about this, or maybe it's just

14:19

me thinking about this. We think about it

14:21

in terms of kind of like people, but

14:24

I don't know that we've understood

14:26

sort of the layers that agents could

14:28

potentially create in terms of new

14:30

capabilities that we don't necessarily map very

14:32

well to our authentication systems. And

14:34

maybe we'll dive into this with Brandon

14:36

a little bit because he'll be

14:38

on cloud news the month this month

14:40

and that's sort of one of

14:42

his areas of specialty is ID stuff.

14:46

And the last one I had is, I'm

14:49

going to be very interested in

14:51

when we have agents that are

14:53

focused on different things, right? And

14:55

we've already seen certain things happen,

14:57

right? We've seen chatbots and we've

14:59

seen co -pilots and we've seen

15:01

things that sort of float around

15:03

what an agent could sort of

15:05

mean is, well, we have agents

15:07

that are kind of specialized. And

15:09

I assume the answer to this

15:11

is, yes, absolutely, we'll have agents

15:13

that are specialized. But I'll be

15:16

very interested to see what happens

15:18

when we have agents who are

15:20

specialized in human interacting type of

15:22

things, i .e., and maybe this will

15:24

be the dividing line. Maybe those

15:26

will be more chatbot type of

15:28

things. But I can imagine the

15:30

agents, again, even if

15:32

they're just doing... customer service, you

15:35

know, creating agents that have a

15:37

level of empathy and understanding of,

15:39

you know, what do we want

15:41

our experience to be with humans

15:43

versus the things that are going

15:46

to be, you know, agent to

15:48

agent or agent to system, you

15:50

know, service type of things in

15:52

which we're trying to drive a

15:54

certain level of efficiency or sort

15:56

of deterministicness to the end result.

15:59

And I'll be very interested to see, you

16:01

know, how those two sort

16:04

of different characteristic trade, an

16:06

agent talking to another computing

16:08

service versus an agent interacting

16:10

with a human being or

16:12

a set of humans. You

16:14

know, how they're able to

16:16

be tuned, adjusted, taught

16:18

about the concepts of, you know, this

16:21

is the brand, this is the experience

16:23

we want to try and have versus,

16:25

I need you to get me the

16:27

very best answer with the least amount

16:29

of, you know, risk or the greatest

16:31

probability of success. you know, in

16:33

a, in a, you know, computer to computer type

16:35

of communication. So anyways,

16:37

um, yeah, that was, that

16:39

was the last one I sort of had

16:42

on my list, uh, in terms of

16:44

I'll be very interested to see how that

16:46

evolves, right? How it, how it plays

16:48

itself out, uh, in terms of, um, you

16:50

know, how will we understand those, those

16:52

things in terms of framework? Um,

16:54

will those sort of be binary types of questions

16:56

like, Hey, this is a human. So these

16:58

are the, these are the three or four. core

17:00

things you want to sort of focus the

17:02

agent around. This is going to be non -human,

17:04

know, you're going to focus on these types of

17:06

things. Like, are we going to be able

17:09

to make it simple enough that it can be

17:11

kind of a checkbox of, you know, empathy, friendliness,

17:14

you know, patience, all

17:16

those sorts of things

17:18

versus, you know, deterministic,

17:21

when at all costs, get the lowest answer, you

17:23

know, get the best outcome, all those types

17:25

of things. Like, I'll be interested to see how

17:27

that goes along. So anyways, I think those

17:29

are my five things that I had. How

17:32

will we represent them? How will we

17:34

audit them? How

17:36

will we determine when to use

17:38

one versus the other? What are our

17:40

off systems going to look like? And

17:42

then how will we build kind of

17:44

fit for purpose types of concepts around

17:46

these agents? So anyways, just a

17:49

few thoughts I had. I'll be curious

17:51

what types of things you all are thinking

17:53

about in terms of AI agents, whether it's

17:55

at a technology level or at a You

17:57

know system level or you know kind of

17:59

a again kind of an interaction level so anyways

18:01

that. I'll wrap it up again sort of

18:03

said sort of a short Short show just a

18:05

few things I've been thinking about I put

18:07

some links in the the show that you can

18:09

read a little bit More about all these

18:12

things That have you know kind

18:14

of in my mind lately, but with that I'll wrap

18:16

it up Thank you all for listening. Thanks for

18:18

telling a friend. Hope your spring is going well Hope

18:20

things are getting warmer if you have questions again,

18:22

Feel free to shoot us an email at at the

18:24

cloudcast .net. Hit us up on any of the social

18:26

channels that we've got at the link at the

18:28

bottom of the show notes With that

18:30

I'll wrap it up Another cloud of the month

18:32

coming here pretty briefly Those always are know very

18:34

very popular people love those So we've that coming

18:37

up soon and with that I'll wrap it up

18:39

We'll talk to you next week Thank

18:41

you for listening to the

18:43

Cloudcast. Please visit the cloudcast net

18:45

to find more shows show

18:48

notes videos and everything social media

18:55

You