Scott Brinker on How AI Will Reshape Martech in 2025

Justin: 00:00:14

welcome to rev ops FM, everyone.

2

: 00:00:16

Big day on the show today.

3

: 00:00:17

As we chat with someone who

really needs no introduction.

4

: 00:00:20

It is Scott Brinker, VP of

platform ecosystem at HubSpot

5

: 00:00:25

editor at chief martech.

6

: 00:00:26

com and creator of the famous marketing

technology landscape, super graphic.

7

: 00:00:32

Now Scott has just released a new

report called Martek for: 2025

8

: 00:00:36

And as you might expect, it's

all about how AI is reshaping the

9

: 00:00:40

marketing and Martek environment.

10

: 00:00:42

And it's just under a year ago

that we've done our first episode

11

: 00:00:44

on the impact of AI and marketing.

12

: 00:00:46

So I thought this is a great time to

check in, see how things have evolved in

13

: 00:00:50

the past 12 months, and how AI is going

to change your job in the years to come.

14

: 00:00:55

And Scott is gonna be our guide.

15

: 00:00:57

Scott, it is a pleasure

to have you on the show.

16

: 00:01:00

Scott Brinker: Well, thank you

so much for having me, uh, fun

17

: 00:01:02

topics looking forward to this

18

: 00:01:04

Justin: It is now Scott, I understand

your cocktail party trick is

19

: 00:01:06

that you can recite all 14,000.

20

: 00:01:08

Vendors from the Martech

landscape graphic by heart.

21

: 00:01:11

Is this true?

22

: 00:01:12

Scott Brinker: and I can even

draw their logos from memory.

23

: 00:01:16

Justin: It is, it is, it is an

amazing talent and my, my hat goes

24

: 00:01:19

off to you for maintaining that now

for how many years is it now that

25

: 00:01:22

you've been, producing that graphic?

26

: 00:01:24

Oh my

27

: 00:01:24

Scott Brinker: so it's

over a 14 year period.

28

: 00:01:26

We took one year off,

back in: 2012

29

: 00:01:30

Yeah, but yeah, it's just

been growing crazy ever since.

30

: 00:01:33

Justin: Well, it is become an amazing sort

of pillar of our, uh, of our ecosystem.

31

: 00:01:38

So thank you for the

work that you do there.

32

: 00:01:40

on this year's report on Martech

in: 2025

33

: 00:01:44

with the bottom line up front.

34

: 00:01:45

Like what are top one to two

messages you would like mops pros

35

: 00:01:48

to take away from this report?

36

: 00:01:50

Scott Brinker: Wow.

37

: 00:01:51

Well, I think at the end of the day,

we all know marketing is changing.

38

: 00:01:56

both at a technical level of like

the capabilities, you know, uh, the

39

: 00:02:00

products we use, what AI is making

possible for us as marketers to do.

40

: 00:02:05

but it's also very clearly shifting

the dynamics for how buyers are

41

: 00:02:10

going to want to engage with us.

42

: 00:02:11

you know, it's one of the things.

43

: 00:02:12

I mean, just to like pick some

random thing we could look at that

44

: 00:02:15

in previous years would have been

an entire year's worth of content.

45

: 00:02:19

Uh, and then this year we're like, Oh

yeah, by the way, it turns out that maybe

46

: 00:02:23

Google search, you know, is going to be

disrupted by these other channels in which

47

: 00:02:27

people are going to go and like, you know,

just directly get answers to questions.

48

: 00:02:30

We're like, yeah, that's one of

the like 15 or 20 major disruptions

49

: 00:02:34

that I'm looking at at the moment.

50

: 00:02:35

Thank you very much.

51

: 00:02:37

But I mean, you know,

again, I know this is.

52

: 00:02:41

I don't want to sound flippant about it

because I know it's a stressful thing

53

: 00:02:44

for people when you've got all these

changes and you're trying to keep up,

54

: 00:02:49

uh, and if there's any consolation,

it's that everyone's in the same boat.

55

: 00:02:52

Everyone is trying to keep

up with these changes.

56

: 00:02:55

Um, but I, but I always take this

optimistic view that these sorts of

57

: 00:03:00

disruptions are actually a good thing.

58

: 00:03:03

Gift to in particular marketers,

um, you know, marketing is

59

: 00:03:08

always about differentiation.

60

: 00:03:10

It's always about standing out from the

crowd, you know, and when you've ended

61

: 00:03:15

up in a market or an environment where

sort of everything is sort of converged

62

: 00:03:18

down to like a common playbook and common

channels and common approaches, it's

63

: 00:03:22

actually really, really hard to stand out.

64

: 00:03:25

But when you're in these moments

of disruption where things are

65

: 00:03:28

changing, I think those sorts of

marketers who are willing to be.

66

: 00:03:31

bit bold and like push out on the

frontier and experiment with it.

67

: 00:03:35

It becomes a really great opportunity

for them to differentiate.

68

: 00:03:39

So yes.

69

: 00:03:40

There's a whole bunch of ways in

which, you know, marketing and

70

: 00:03:42

MarTech is changing, but it's

actually a great opportunity.

71

: 00:03:46

Justin: One of the, risks concerns, I

don't know how we want to phrase it,

72

: 00:03:49

but one of the, one of the things that

I've felt and seen other people express.

73

: 00:03:53

Is, you know, if everyone's using

chat GPT to write their copy,

74

: 00:03:56

everyone kind of sounds the same.

75

: 00:03:58

so it's interesting that you,

you poke on it as like an

76

: 00:04:01

opportunity for differentiation.

77

: 00:04:02

How do you think companies can use AI,

to actually be different to stand out?

78

: 00:04:07

Scott Brinker: Yeah, well, I mean, I would

argue the, uh, yes, just using chat GPT

79

: 00:04:12

out of the box to write my blog articles

the same way everyone else is, that is

80

: 00:04:17

that same sort of pattern we've seen of

like, oh, well, I guess if you publish

81

: 00:04:21

a top 10 listicle on keyword X, you

know, I mean, there's always a formulaic

82

: 00:04:27

approach to these things that, you know,

sadly, yeah, a lot of, you know, sort of

83

: 00:04:32

Companies a lot of marketers like fall

into the trap of but very few of those

84

: 00:04:36

things like started out That way there was

someone who actually invented the listicle

85

: 00:04:42

who like hey Wow, this is actually a way

to like people this resonates with folks.

86

: 00:04:46

Oh, I could do a series

of these Andrew Chen who?

87

: 00:04:48

uh, has had a number of different roles,

you know, as a VC up to growth for Uber.

88

: 00:04:53

Uh, he had a phrase he had invented.

89

: 00:04:56

The Law of Shitty Clickthroughs.

90

: 00:04:58

Uh, you know, which is basically

this idea that, you know, marketing,

91

: 00:05:01

particularly in the digital space.

92

: 00:05:03

Is this continuous sequence

of, okay, a market or coming up

93

: 00:05:07

with some sort of novel way of

breaking through and engaging.

94

: 00:05:11

it works, like it's like this

incredible, like return because

95

: 00:05:14

it's not been done before.

96

: 00:05:16

And the word starts to leak out and

then other people start to do it.

97

: 00:05:19

And basically eventually over time, the

efficacy drops and it reverts to the mean

98

: 00:05:23

and you have to come up with another one.

99

: 00:05:25

And so, yeah, there's definitely

a lot of that with AI.

100

: 00:05:27

I think, you know, the folks

who are going to be creative.

101

: 00:05:32

With generative AI aren't just

saying like, Oh yeah, chat GPT,

102

: 00:05:37

write my blog article for me because.

103

: 00:05:39

Nobody's going to be

reading, those blog articles.

104

: 00:05:41

I mean, like if we have, if we

produce like a hundred times more,

105

: 00:05:45

you know, blog articles, uh, we have

not invented a hundred times more

106

: 00:05:49

human attention, you know, for people

to actually consume these things.

107

: 00:05:53

In fact, if anything, there's sort of

the indication now that, you know, the

108

: 00:05:56

recipients on the other side here, they're

kind of savvy on this, you know, they're

109

: 00:06:00

leaning into like, Hey, you know, this

huge stream of emails that you sent me.

110

: 00:06:04

Can you just summarize that in

a few bullet points for me and

111

: 00:06:07

let me know if there's anything

particular I should pay attention to.

112

: 00:06:10

So I think the creative marketers are

looking at generative AI or AI more

113

: 00:06:14

broadly, not about just how do we churn

out more of the same kind of articles,

114

: 00:06:19

but how do we do something different?

115

: 00:06:21

Justin: on that note, your

survey suggested like at

116

: 00:06:23

least 80 percent of marketers.

117

: 00:06:25

If I interpreted the results correctly

at least 80 percent of marketers are

118

: 00:06:28

using AI in some fashion, and it looked

like at least half of them were using AI.

119

: 00:06:31

Daily or weekly, which I think is huge,

like compared to where we were a year

120

: 00:06:36

ago, where it felt like people were

still like, what do we do with this?

121

: 00:06:39

And what is this thing?

122

: 00:06:40

but drilling down on on that creativity?

123

: 00:06:43

do you have a sense?

124

: 00:06:44

Could be database could be

intuition based, like the

125

: 00:06:47

impact of this AI adoption?

126

: 00:06:49

Is it mainly efficiency right now?

127

: 00:06:51

Or are there marketers out there that

you've seen who are actually being able

128

: 00:06:54

to produce better marketing as a result

of doing this than they otherwise could?

129

: 00:06:58

Scott Brinker: there's a little

bit of a blending between the

130

: 00:07:01

efficiency and the innovation side.

131

: 00:07:03

for instance, like, when I think

about it this way, okay, so like the

132

: 00:07:06

efficiency would just be like, okay,

well, we can do this thing faster

133

: 00:07:10

or we can do it less expensive.

134

: 00:07:12

Not bad.

135

: 00:07:13

but here's where it's sort of like plays

into the degree to which you're able to

136

: 00:07:17

turn that from an efficiency into a true

innovation thing is it starts with like.

137

: 00:07:24

The questions, the hypotheses, what

are the experiments that I want to run,

138

: 00:07:27

you know, well, historically, you know,

for marketers who weren't necessarily

139

: 00:07:31

data scientists or data engineers,

like being able to go deeper into their

140

: 00:07:35

data, you know, to answer some of these

questions, to get data, to feed some

141

: 00:07:39

of their hypotheses, very often involve

the cycle of filing tickets and grab

142

: 00:07:43

an analyst, track that down, you know,

and so there's a whole set of some

143

: 00:07:46

of these, uh, you know, new AI tools.

144

: 00:07:49

many of them are early, but

they're developing very quickly.

145

: 00:07:52

Uh, that is democratizing the ability

for marketers to be able to ask more

146

: 00:07:57

questions of their data and have it

come back with answers immediately.

147

: 00:08:01

So we're taking out, you know, that

whole like multi day or in some case,

148

: 00:08:05

multi week, you know, analyst cycle.

149

: 00:08:07

Now you could say, that's

the thing about efficiency.

150

: 00:08:10

It is, you know, but to me,

marketers, like one of their.

151

: 00:08:16

is their imagination.

152

: 00:08:17

That's just like constant, like coming up

with questions and ideas, you know, the

153

: 00:08:21

vast majority of which we've kind of been

trained to just like, let go of, because

154

: 00:08:26

we're like, oh yeah, no, I just can't

take too much work to get that answer.

155

: 00:08:29

So, you know, I don't care, you know,

but as we sort of, you know, uh, reduce

156

: 00:08:33

the number of things that feel out

of reach to be able to like ask more

157

: 00:08:37

and more questions and get answers

and like chase those ideas down, I

158

: 00:08:41

think that actually helps unlock.

159

: 00:08:43

New creativity and new ideas.

160

: 00:08:46

Same thing when we talk about the actual

implementation, when we compress the time

161

: 00:08:51

and costs to be able to produce things,

This isn't just about saying like, Oh, I

162

: 00:08:55

was going to produce 10 things, you know,

and I was going to spend a week on it.

163

: 00:09:00

Now I can produce 10 things

and I can do it in two days.

164

: 00:09:04

Um, Oh, okay, great.

165

: 00:09:06

I'm more efficient.

166

: 00:09:07

You know, it's really like, okay, now for

that extra three days, like what do I do?

167

: 00:09:12

Can I produce more things?

168

: 00:09:14

And it's that produce more things,

which is where you get that.

169

: 00:09:17

inflection from like, okay, wait a second.

170

: 00:09:19

This isn't necessarily just about

efficiency because as long as you're

171

: 00:09:23

not saying, well, I'm just going to

produce more of the same sort of thing,

172

: 00:09:26

but you're gonna be like, oh, well,

let me try some other experiments.

173

: 00:09:29

Let me like, you I've got a little

bit of time, I've got this wow idea.

174

: 00:09:34

let me create this and put that out there.

175

: 00:09:38

So harnessing that time is another way.

176

: 00:09:42

, And a third one is, and we've seen

this direction headed for a while with

177

: 00:09:48

these sort of no-code tools, , but

now generative AI is bringing.

178

: 00:09:53

a sort of no-code capability

to so many disciplines.

179

: 00:09:56

You know, I mean, you know,

with, uh, you know, images.

180

: 00:09:59

We're starting to see some really

cool stuff emerge here in video

181

: 00:10:01

creation and things like this.

182

: 00:10:03

it ex it democratizes ability

for people who have an idea?

183

: 00:10:08

To actually like,

instantiate it, make it real.

184

: 00:10:11

you know, and even if you want to be

modest about it and say like, Oh, well,

185

: 00:10:15

you know, maybe some of this stuff isn't

ready to go from idea to, you know,

186

: 00:10:18

production ready final version, if it's

to go from idea to something, that's

187

: 00:10:23

a pretty darn good representation of

what it could be, you know, to be able

188

: 00:10:29

to get reactions to that, you know,

and use that as a guide for developing

189

: 00:10:33

the production ready version one.

190

: 00:10:36

Again, is this an efficiency thing?

191

: 00:10:38

Yes, because I didn't necessarily need

the full team to even like put together

192

: 00:10:42

a prototype concept, but it's that

empowerment to create these prototype

193

: 00:10:46

ideas, you know, just pretty much at

the speed of thought, that is a huge

194

: 00:10:51

unlock from an innovation perspective.

195

: 00:10:53

Oh my goodness.

196

: 00:10:54

It's like fuel for the imagination.

197

: 00:10:57

Justin: And you've multiplied your

ability to iterate to your point,

198

: 00:11:00

to try more ideas or to spend that

time doing deep human work that maybe

199

: 00:11:05

you usually would not have time for

in a typically overloaded calendar.

200

: 00:11:09

Scott Brinker: Yeah.

201

: 00:11:10

You know what I mean?

202

: 00:11:10

You talk about like what you could

do with the additional hours in the

203

: 00:11:13

day and certainly more experiments

and more ideas, you know, but for the

204

: 00:11:17

human work, how about spending more

time just talking with customers, you

205

: 00:11:21

know, I mean, we know as marketers,

like that's a big part of our job.

206

: 00:11:26

I.

207

: 00:11:26

You know, a lot of marketers who

don't get a lot of time actually being

208

: 00:11:30

able to talk with customers to be

able to have that sort of like human

209

: 00:11:34

connection and human insight that

they feed in to their imagination.

210

: 00:11:38

So I think you're right, like that

sort of human work, boy, if we

211

: 00:11:41

could get 10, 20 percent more of

our week, uh, you know, allocated

212

: 00:11:45

to that, that would be a huge gift.

213

: 00:11:47

Justin: You know, one of the

first things in your report is

214

: 00:11:51

a graphic of the hype cycle.

215

: 00:11:53

I will include a link to the report so

people can can look on this, but you know,

216

: 00:11:56

it has that peak of inflated expectations,

that trough of disillusionment

217

: 00:12:01

and the, gradually increasing up

to the plateau of productivity.

218

: 00:12:05

And I can feel riding that over

the past year from like, whoa, AI,

219

: 00:12:09

like AI, like AI SDRs, they suck.

220

: 00:12:13

And then gradually I find myself now

you mentioned about disrupting Google.

221

: 00:12:18

I've started to ask chat GPT

things instead of Googling them.

222

: 00:12:21

Like it's actually, I

don't do it on purpose.

223

: 00:12:23

It's just, this is a more efficient

way to get my information or, I have

224

: 00:12:27

grok on my phone and yesterday there

was a battery with some words in

225

: 00:12:30

French and I didn't understand it

and I just took a picture and said,

226

: 00:12:32

can you translate this text for me?

227

: 00:12:34

Translated the text like it's

starting to work its way in.

228

: 00:12:38

So like, where do you think

we are on that on that curve?

229

: 00:12:41

Where would you peg us?

230

: 00:12:42

You know, obviously everyone's at

different places personally, but

231

: 00:12:45

as a collective, what do you think?

232

: 00:12:51

Scott Brinker: two reasons.

233

: 00:12:51

Um, one is I'm convinced there isn't

actually one hype curve here, but

234

: 00:12:57

like there's a multitude of different

hype curves all at different stages.

235

: 00:13:01

I mean, again, like with, um.

236

: 00:13:03

I mean, just take some of the stuff here

with like open AI, you know, where are

237

: 00:13:06

we at as far as our comfort level and

using chat GPT, to help us brainstorm or

238

: 00:13:12

help us like edit some of our writing.

239

: 00:13:15

I think we're actually now pretty far into

the, you know, plateau of productivity.

240

: 00:13:19

Most people are pretty

comfortable with it.

241

: 00:13:20

It's pretty good at that.

242

: 00:13:22

Where are we on using

it as a search engine?

243

: 00:13:25

I don't know.

244

: 00:13:25

Probably at this exact moment, uh,

given the announcements, you know,

245

: 00:13:28

uh, happening, you know, right

now, uh, we're probably at the peak

246

: 00:13:31

of hype of like, Oh my goodness,

it's going to disrupt everything.

247

: 00:13:35

There'll be a trough, you know, where

we had on the, you know, ability from

248

: 00:13:39

open AI with things like soar and, you

know, there's video creation stuff.

249

: 00:13:42

Probably very early on in

the technology triggers.

250

: 00:13:46

And so it's, it's interesting because

for one thing, there's just multiple

251

: 00:13:49

hype curves at different stages.

252

: 00:13:51

but the other thing is, this is

the thing about the hype curve.

253

: 00:13:54

I've, I've just noticed over the years.

254

: 00:13:56

And again, kudos to.

255

: 00:13:57

Gartner for like identifying this pattern

and like articulating it, yeah, in such

256

: 00:14:02

a excellent way, but it used to be,

I mean, I've got some gray hair here.

257

: 00:14:06

It used to be that these hype curves

were something that played out over.

258

: 00:14:11

Between a 5 to 10 year period, you

know, in fact, this is how like

259

: 00:14:14

the tech industry was structured.

260

: 00:14:16

This is how venture capital

was built to do this, slowly,

261

: 00:14:23

You

262

: 00:14:25

know,

263

: 00:14:30

well, maybe it's more like, you know,

4 to 6 years, things like in marketing,

264

: 00:14:35

uh, the customer data platform,

you know, You know, revolution.

265

: 00:14:38

I feel like that thing like Barely got

started in like: 2015

266

: 00:14:43

and now we're already at the thing

of like, well, CDPs, are they, they

267

: 00:14:47

a thing now, or is this all the day?

268

: 00:14:48

And you're like, wait a second.

269

: 00:14:49

I, this thing was like

barely getting started.

270

: 00:14:51

Like what, um, now we're on to some

next new, uh, you know, architecture.

271

: 00:14:56

It feels like an AI compressed that

hype curve down to something that

272

: 00:15:00

really is more about months than years.

273

: 00:15:02

I kept feeling like, you know, even in

the first year or so of chat GPT, you'd

274

: 00:15:06

be like, okay, well, yeah, this thing is

really great, but it's data is, uh, you

275

: 00:15:11

know, two years old and then they connect

it up to the web and you're like, okay,

276

: 00:15:14

all right, well, I guess that's all.

277

: 00:15:15

Yeah.

278

: 00:15:16

It's really good, but it

can't do math problems.

279

: 00:15:19

Uh, you know, oh, well now we connected

and it can run little Python programs and.

280

: 00:15:23

Calculate that and you're like, okay,

well, like everything I like thought was

281

: 00:15:27

this barrier that would be this multi

year thing kind of vanishes after a couple

282

: 00:15:31

months, you know, and so I think it just

makes it really hard to say like, yeah,

283

: 00:15:35

the hype curve is still a thing, but where

are we and how quickly will we move along?

284

: 00:15:41

Boy, I don't know.

285

: 00:15:43

Justin: you mentioned the

book, the innovators dilemma, a

286

: 00:15:45

number of times in your report.

287

: 00:15:47

And in that book, uh, the author

talks about how he studied the

288

: 00:15:51

disk drive industry and Because of

this like fruit fly effect where

289

: 00:15:55

it was a very fast moving industry.

290

: 00:15:56

So you could observe the life cycle.

291

: 00:15:58

And I just, he's passed away

obviously, but he was here looking

292

: 00:16:02

at, at these very compressed cycles

that we're going through now, like

293

: 00:16:06

it is like watching life almost on 1.

294

: 00:16:09

5 X.

295

: 00:16:09

You know, the way you can with a

YouTube video where you see things

296

: 00:16:12

moving very quickly, businesses

starting trends coming and going.

297

: 00:16:16

Scott Brinker: Yeah, no.

298

: 00:16:17

And such a brilliant, Clay

Christianson, uh, you know,

299

: 00:16:20

the whole innovative dilemma.

300

: 00:16:21

but this thing about disruptive

innovation, boy, you just

301

: 00:16:24

see it this perfectly thing.

302

: 00:16:25

Like, his big, insight there

was, you know, so often these

303

: 00:16:28

disruptions would happen.

304

: 00:16:30

From below, where like it would end up

empowering a set of people to do things

305

: 00:16:35

who quite frankly, wasn't cost effective

for them to do things the old way.

306

: 00:16:39

Like the great example would

be like, you know, coding.

307

: 00:16:41

Okay, well, if I needed to build

a software program, you know.

308

: 00:16:45

Not too long ago.

309

: 00:16:46

All right.

310

: 00:16:47

Well, I've got to hire multiple engineers.

311

: 00:16:49

I'm going to need this.

312

: 00:16:49

You know, it's like, oh, okay.

313

: 00:16:51

So if I want to do a simple little

program to like, you know, go fetch me

314

: 00:16:55

some stock market data and I'm like,

no, it's just not worth doing that, you

315

: 00:16:59

know, and then you have these disruptive

innovations that come along that.

316

: 00:17:03

They're maybe not yet at the

level where they can do the sorts

317

: 00:17:07

use Silence.

318

: 00:17:12

cases, they work friggin beautifully, you

know, and they unlock that capability for

319

: 00:17:17

all these people who before, like, didn't

even have the option, to get things done.

320

: 00:17:22

So Big fan of, Christiansen.

321

: 00:17:24

This is his time.

322

: 00:17:26

Justin: Yeah, no, it's.

323

: 00:17:28

Scott Brinker: above, he's like, yep.

324

: 00:17:31

, Justin: but part of that dynamic that

he describes about the innovation coming

325

: 00:17:34

from above you, you start to address it

a little bit in, the different segments

326

: 00:17:39

and, maybe you can walk us through them

a little bit, but the, the segments

327

: 00:17:42

that you kind of spell out, are these

indie tools, you know, bubbling up

328

: 00:17:46

Using AI innovating very rapidly.

329

: 00:17:48

These challenger platforms that

are kind of, threatening the big

330

: 00:17:51

incumbents and consolidating a number

of different abilities, the incumbents.

331

: 00:17:54

who are, you know, quickly

trying to add a eyes as fast as

332

: 00:17:58

they can onto their platforms.

333

: 00:18:00

and then, uh, won't even touch

services as a software yet.

334

: 00:18:02

But then you also talk as you just

address this, this custom apps idea

335

: 00:18:05

that actually I can just build it

myself now with chat GPT, help maybe

336

: 00:18:10

amend or add anything you'd like to

to my explanation of that landscape.

337

: 00:18:14

And then perhaps where are you

seeing the most fruitful innovation

338

: 00:18:17

right now in all of those segments?

339

: 00:18:19

Scott Brinker: well, I would start by

saying, you know, in any, in any, uh, in

340

: 00:18:21

any, uh, in any, uh, Period of disruption.

341

: 00:18:23

Uh, we're used to thinking of sort of the

battle between startups and incumbents.

342

: 00:18:27

you know, and one of the reasons why

we made a distinction and sort of the

343

: 00:18:31

startup segment between this idea of

these indie tools versus the challenger

344

: 00:18:34

platforms is because so many of these

new AI tools that have popped up.

345

: 00:18:42

actually looking to displace,

the major incumbent platforms.

346

: 00:18:45

In many cases, they now

integrate with those platforms.

347

: 00:18:48

They're complimentary to those platforms.

348

: 00:18:51

and so it's a little bit like, oh, I

mean, again, not to say there won't be

349

: 00:18:55

overlap and exchange back and forth,

you know, in many ways, there will be.

350

: 00:18:58

The complimentary nature there seems to

exceed the competitive nature, at least

351

: 00:19:02

at this stage of the game versus there

are a set of challenger platforms that are

352

: 00:19:07

very much looking at this inflection point

of like, Oh yeah, we're gonna, we're,

353

: 00:19:11

we're bringing down Salesforce, you know?

354

: 00:19:13

and that's a hard.

355

: 00:19:14

That's a hard hill to climb, but,

356

: 00:19:16

Justin: Yep.

357

: 00:19:16

Scott Brinker: know, this is

the nature of the tech industry.

358

: 00:19:19

Um, some, some of those folks are

actually going to climb that hill and,

359

: 00:19:22

you know, disrupt the incumbents they're

going after, you know, so that's the,

360

: 00:19:26

you know, that's what a central, the

commercial, uh, MarTech landscape, uh,

361

: 00:19:31

you know, where we're seeing those groups.

362

: 00:19:33

The thing about the custom software

is people Because AI is really, I

363

: 00:19:40

mean, already the barriers to entry

in creating software, the cost of

364

: 00:19:44

creating software was already like

plummeting, you know, given the cloud,

365

: 00:19:47

given frameworks, given, you know,

AWS services of every flavor you can.

366

: 00:19:52

I mean, it was already like headed

down, like more people could do it than

367

: 00:19:55

ever, but now with these AI co pilots.

368

: 00:19:58

And increasingly, you might talk a

bit about how AI agents are able to

369

: 00:20:03

even create code without you knowing

that they were creating code for you.

370

: 00:20:07

Um, you know, it's just further

accelerated this ability for like

371

: 00:20:10

companies to be able to create

more and more software themselves.

372

: 00:20:15

will that be a total replacement

to those commercial products?

373

: 00:20:19

my sense is in the, Foreseeable future.

374

: 00:20:22

Probably not.

375

: 00:20:23

Um, you know, with due

credit, uh, what was it?

376

: 00:20:26

Klarna, you know, ripping out Salesforce

and Workday and all these other ones,

377

: 00:20:30

but you know, they're, they're in a

league of their own, you know, but what

378

: 00:20:34

I think of it is, is a more of a, again,

another one of these complimentary

379

: 00:20:39

components, where like, okay, yeah,

there are certain capabilities on the

380

: 00:20:42

market that it wouldn't make sense for

me to like reinvent the wheel, right?

381

: 00:20:46

Like, you know, I don't need

to write my own email server.

382

: 00:20:48

At this point, that's not a comparative

advantage, uh, you know, for most

383

: 00:20:52

companies, um, but when it comes to how

I want to design a particular customer

384

: 00:20:57

experience or the way in which I want to

optimize a particular workflow, from quote

385

: 00:21:02

to cash, you know, that actually might

be something that I would want to custom

386

: 00:21:07

develop the way in which that works, the

way in which that experience is delivered.

387

: 00:21:11

Now I can do that on top of

those commercial platforms,

388

: 00:21:15

but I'm now leveraging AI.

389

: 00:21:16

To be able to accelerate the

customization that I build on top of it.

390

: 00:21:20

Justin: you know, incumbents have often in

my experience With adding that innovation

391

: 00:21:27

and, mean, there's many potential reasons.

392

: 00:21:29

Part of it is maybe they just are

not as laser focused on that one

393

: 00:21:32

problem, but it's always seemed

counterintuitive to me that like.

394

: 00:21:36

You know, two developers in a bedroom

somewhere can produce something that is

395

: 00:21:41

a bit closer to the pulse of the market

than some of these big legacy platforms

396

: 00:21:44

that I've owned and used with all these

resources who just seem to miss the mark.

397

: 00:21:49

And some of that, you know, you've

mentioned, you know, they have

398

: 00:21:52

backwards compatibility, dependence

on existing revenue structures.

399

: 00:21:56

There's a lot of the

innovators dilemma stuff.

400

: 00:21:58

And then part of it, to me, it almost

feels like a mindset thing where somehow

401

: 00:22:02

a dysfunction creeps in at larger

scale where they seem unable of, of

402

: 00:22:07

the focus or of the creative leaps that

are required, to innovate at the level

403

: 00:22:12

that sort of the young blood can, I

wonder, do you share that experience?

404

: 00:22:17

I'll say as a notable exception,

I'm not just saying, cause you're

405

: 00:22:19

on, you're on the show, but, but,

but HubSpot almost feels to me it

406

: 00:22:22

to be an exception to that rule.

407

: 00:22:24

And I don't know how they've.

408

: 00:22:24

continue to innovate the way that they

seem to have been able to do, at the

409

: 00:22:29

scale that they're at, but a lot of other

companies, and I come from the Marketo

410

: 00:22:32

ecosystem, they have not been able to do

that, like Eloqua has not been able to,

411

: 00:22:36

you know, a lot of these players, um,

what is your take on this issue about

412

: 00:22:39

innovation within legacy incumbents

and how can it flourish as a possible?

413

: 00:22:43

Scott Brinker: there's a quote and

I wish I could remember who said it.

414

: 00:22:45

Um, I believe it was a

at Andreessen Horowitz.

415

: 00:22:50

that it's like, uh, it's a race between,

will the startups get distribution

416

: 00:22:55

before the incumbents get innovation?

417

: 00:22:57

I think this time is a

little bit different.

418

: 00:23:00

you know, for a couple, of reasons.

419

: 00:23:03

First of all, is this pattern of.

420

: 00:23:06

dilemma like disruption from below.

421

: 00:23:08

Everyone's now sort of read those books.

422

: 00:23:10

Um, you have a whole generation,

you know, of, uh, executives at

423

: 00:23:15

these incumbent companies that

perhaps a number of them actually

424

: 00:23:19

were the disruptors, you know, Okay.

425

: 00:23:23

uh,

426

: 00:23:30

A

427

: 00:23:34

little

428

: 00:23:47

bit slower, just, you know, it'd be

sort of easier to say like, yeah, yeah,

429

: 00:23:50

all right, we'll get to that next year.

430

: 00:23:52

We've got, you know, the main

thing we need to ship this

431

: 00:23:54

year, you know, there's just.

432

: 00:23:56

There's just no illusion of that.

433

: 00:23:58

You know, everyone basically looks

at that and says like, yeah, this

434

: 00:24:01

is, this is changing the game.

435

: 00:24:02

This is going to change everything, you

know, about how software actually works.

436

: 00:24:07

And so given that, you know, forceful

momentum, you know, and then, yeah,

437

: 00:24:12

the fact that a lot of the leaders

that these companies do recognize that.

438

: 00:24:16

Yeah, you know, if, if we lose this,

we're going to lose it because we let, you

439

: 00:24:20

know, someone disrupt us, uh, you know,

from below on this I think companies are

440

: 00:24:25

like attacking this with much greater

ferocity, than historically they have.

441

: 00:24:30

Now, that being said, there are still

like, there are structural advantages

442

: 00:24:34

and structural disadvantages, you

know, that large companies have.

443

: 00:24:38

and I don't think there's

a silver bullet to it.

444

: 00:24:40

But I think we're going to see through

this next cycle of the next five

445

: 00:24:44

years, which ones of the incumbents

actually successfully made the

446

: 00:24:48

transitions and which ones didn't.

447

: 00:24:50

And they're going to become the, you

know, business school case studies of

448

: 00:24:53

like, okay, here's how you fend this off.

449

: 00:24:57

I mean, at the end of the day, it's, you

know, we started with Christiansen, we

450

: 00:25:00

end with Christiansen, it's like, you

have to be willing to disrupt yourself.

451

: 00:25:05

you know, and I mean, I'll, uh, you know,

give a shout out here to, um, competitor.

452

: 00:25:10

you know, I think, Benioff, you

know, and basically taking all of

453

: 00:25:14

Dreamforce, being like, nah, it's

not Salesforce, it's AgentForce, you

454

: 00:25:18

know, my God, was that a ballsy move.

455

: 00:25:22

now, you know, how that plays

out, how it delivers on it.

456

: 00:25:25

All right.

457

: 00:25:25

You know, I'm not, not for me to comment

at this point, you know, but I mean,

458

: 00:25:29

that's sort of like willingness to

like disrupt, you know, all of your

459

: 00:25:34

existing plans, all of your existing

narrative, you know, go to market.

460

: 00:25:38

That's the sort of thing that I think

Christensen would be like impressed and

461

: 00:25:41

be like, okay, yeah, no, not shy about

disrupting the previous model, uh, you

462

: 00:25:48

know, to like make it to the next wave.

463

: 00:25:50

Yeah.

464

: 00:25:50

Justin: And just to close the loop on your

point, the latest edition of Innovators

465

: 00:25:55

Dilemma has a forward by Benioff.

466

: 00:25:57

So, um,

467

: 00:25:59

Scott Brinker: See?

468

: 00:25:59

Okay, yeah.

469

: 00:26:00

Justin: So, just to underscore your

point about the latest, you know, the

470

: 00:26:04

disruptors who have internalized the

message and, like, maybe we've seen

471

: 00:26:07

sort of the end of history then of

the, um, Of that pattern repeating,

472

: 00:26:10

and in terms of the actual implementation,

like I think we all saw in: 2024

473

: 00:26:14

need to get our AI features out the door.

474

: 00:26:16

Like, what are we going to do?

475

: 00:26:17

You know, almost like a, solution

in search of a problem in a way.

476

: 00:26:20

Like we, we have AI, we need to

figure out what we're going to do

477

: 00:26:22

with it and how we're going to.

478

: 00:26:24

we want and be able to

deliver that to users.

479

: 00:26:25

Um, and so, and so, but we have

some, we have some questions

480

: 00:26:26

that we haven't answered.

481

: 00:26:27

So Mike, you're welcome to

jump in and get to them.

482

: 00:26:28

and some of them.

483

: 00:26:29

I've been really interesting.

484

: 00:26:30

Some of them, a few of

them have been good.

485

: 00:26:31

Some of them have been kind of

like, Oh, I see what you're trying

486

: 00:26:34

to do, but, but not so much.

487

: 00:26:35

One of the things I've observed just

personally is a bit of a last mile effect.

488

: 00:26:40

Like, uh, we are recording this,

uh, using the tool to script.

489

: 00:26:42

It's a great tool.

490

: 00:26:43

I use it to, to record my podcast.

491

: 00:26:46

They have a feature just as an example,

like remove umza nos, like, Oh, amazing.

492

: 00:26:50

This takes me hours.

493

: 00:26:51

I just pushed this

button, remove umza nos.

494

: 00:26:53

And it does a pretty good job,

but then I don't know, 5 percent

495

: 00:26:56

of the time, It clips a word.

496

: 00:26:58

It makes it weird.

497

: 00:27:00

So either like you're okay with

that and you're willing to tolerate

498

: 00:27:02

that or you still have to review

the whole thing anyways and so it's

499

: 00:27:05

almost that last mile is enough to

like, it's just not quite there.

500

: 00:27:09

Do you think is this, a fly in the

ointment and we will outgrow it or do

501

: 00:27:12

you think there is something deeper

there that companies need to address to

502

: 00:27:15

be successful as AI software companies?

503

: 00:27:18

Scott Brinker: Yeah, and I think

this is one where it's probably

504

: 00:27:21

incredibly use case dependent,

uh, you know, on what the both.

505

: 00:27:27

Um, what is it?

506

: 00:27:28

It's a question of both like how

much tolerance is there for error.

507

: 00:27:33

Um, and then.

508

: 00:27:35

You know, what is the, uh, speed

by which asymptotically you're like

509

: 00:27:40

reducing the error down to where

it falls below that threshold, you

510

: 00:27:46

know, that you're concerned about.

511

: 00:27:47

Um, I do think video, video,

audio, any of these things that

512

: 00:27:54

are like human representations.

513

: 00:27:55

I mean, they have the phrase for

this, like the uncanny valley.

514

: 00:27:58

Um, You know, we see

515

: 00:28:01

Silence.

516

: 00:28:16

I mean, a technological perspective,

it's a frigging wonder, it's still

517

: 00:28:21

pretty clear, like, nah, nah, this

is, this is not a real person.

518

: 00:28:25

And even if you weren't super

paying attention to that, just

519

: 00:28:27

something would seem off, you know?

520

: 00:28:30

Will we get past that?

521

: 00:28:34

Kind of suspect we will.

522

: 00:28:35

and I think in the meantime, you know,

a lot of those things, yeah, like

523

: 00:28:39

again, everyone has their own choices.

524

: 00:28:41

Like I wouldn't use them, you

know, for production purposes.

525

: 00:28:44

you know, but am I comfortable using it to

like, you know, like say clean up certain

526

: 00:28:51

amounts of like audio background noise.

527

: 00:28:54

Yeah.

528

: 00:28:54

It seems to do like in

general, like a phenomenal job.

529

: 00:28:57

Justin: I think, I think you're right.

530

: 00:28:58

I think, no, no, I think you're right.

531

: 00:29:00

And it's, it's like, I can't

remember if this was in your report

532

: 00:29:03

or something else I was listening

to, but it's like, it doesn't work.

533

: 00:29:05

It doesn't work.

534

: 00:29:05

It doesn't work.

535

: 00:29:06

And all of a sudden it

works and it's normal.

536

: 00:29:08

It's just like, well, yes, of

course this would be the way.

537

: 00:29:10

And I think that'll, like you

said, with the, the ums and ahs

538

: 00:29:13

or something like that, it'll,

Eventually go below that threshold of

539

: 00:29:18

acceptability of, of error tolerance.

540

: 00:29:21

And, and then life will, you

know, will never be the same.

541

: 00:29:24

Let's talk a little bit about, agents.

542

: 00:29:27

This was a concept that I, I kind

of thought that I understood.

543

: 00:29:30

And then as I read this section,

was the area where I realized my, my

544

: 00:29:33

understanding was quite superficial.

545

: 00:29:36

In terms of like, Oh, I actually didn't

really understand what we meant by that.

546

: 00:29:40

And about this idea of a large

action model and the real

547

: 00:29:43

implications of an agent.

548

: 00:29:45

So maybe assuming there's probably other

listeners that are in that position as

549

: 00:29:48

well, do you walk us through this and

what it means and what the implications

550

: 00:29:52

are maybe for how we will use them?

551

: 00:29:54

Scott Brinker: Yeah, happy to do that

with the caveat that, um, you know,

552

: 00:29:57

if you ask 100 experts for their

definition of an agent, probably get

553

: 00:30:02

at least 120 different, explanations.

554

: 00:30:04

I think the oversimplified version

is almost like an automation.

555

: 00:30:10

this ability to simply say like,

I'm going to make a request,

556

: 00:30:13

of this app, software agent.

557

: 00:30:16

It's going to go off.

558

: 00:30:16

It's going to perform that request,

come back and deliver it to me.

559

: 00:30:19

I, I think one of the examples I

pointed to here was, Dharma Shaw,

560

: 00:30:23

uh, you know, co founder of HubSpot.

561

: 00:30:25

He's, uh, created as one of his many

projects that he goes off and does.

562

: 00:30:29

Uh, I've been the custom wonder of

this fellow, agent AI, which is a

563

: 00:30:33

whole bunch of these, like very.

564

: 00:30:34

Purpose built, you know, agents

that they go off and they

565

: 00:30:37

do one thing and they do it.

566

: 00:30:39

I think if you ask others, you know,

what makes something an agent is the

567

: 00:30:44

ability for it to have almost a little

bit more of an internalization of how it

568

: 00:30:50

goes about making decisions, you know,

so you give it a higher level goal.

569

: 00:30:55

starts to reason about, okay, well,

how would I accomplish that goal?

570

: 00:30:59

I would break it down into

these different steps.

571

: 00:31:01

I would take this first step

based on the results of that.

572

: 00:31:04

I would adjust, move to the second step.

573

: 00:31:06

you know, and I think this is where

people get most excited because as you

574

: 00:31:12

start looking at agents through that

lens, you know, it's not just about

575

: 00:31:16

things that are almost like these.

576

: 00:31:18

You know, uh, very purpose specific tools,

but they start to become things that I'm

577

: 00:31:23

not quite sure I'm ready to call them

digital coworkers, you know, uh, but they

578

: 00:31:27

become like, yeah, these more autonomous,

support capabilities, for what we do, But

579

: 00:31:33

I think this is where things get really

interesting is because, okay, we're, we're

580

: 00:31:38

used to starting to have this interaction

with chat GPT, just from a conversational

581

: 00:31:43

perspective, you know, and particularly

with some of the latest frontier models

582

: 00:31:46

on these LLMs, you can actually see them

do the reasoning and you can ask them

583

: 00:31:50

and you're like, yeah, I could kind of

figure out a plan for this, the, the

584

: 00:31:55

leap from that conversational AI to truly

being an agent is like, okay, well, now

585

: 00:32:01

I've empowered this thing to take action.

586

: 00:32:02

Action out in the world.

587

: 00:32:04

and some of the most recent, announcements

from folks like Anthropic and OpenAI are

588

: 00:32:08

about, you know, giving these, LLMs, the

ability to, uh, I think what Anthropic

589

: 00:32:13

calls it like computer use, it's this

idea that you can start to try and like

590

: 00:32:18

operate some of your software for you.

591

: 00:32:20

I think that stuff's really interesting.

592

: 00:32:23

to me, when I look at the greatest

opportunity for agents to be put into

593

: 00:32:29

production, it's less about having them

work with software the way we humans

594

: 00:32:36

would work with software of like, Oh

yes, uh, navigate to this menu choice

595

: 00:32:41

and drop down here and pick this and

fill out these three form fields.

596

: 00:32:46

Not that it is impossible, but it's

frankly, Boy, a really circuitous

597

: 00:32:50

way, you know, for a computer to

interact with other computer systems

598

: 00:32:55

to be able to get things done.

599

: 00:32:56

The way in which do that is we

do it through APIs, you know, and

600

: 00:33:00

this is the thing that is very

exciting for me is this rise of the

601

: 00:33:05

AI intelligence to create agents.

602

: 00:33:09

Is happening at the same time that

we've been on a good streak for

603

: 00:33:14

the past decade of more and more

software is now built with open APIs.

604

: 00:33:19

You know, this was a function of we ended

up with these heterogeneous tech stacks.

605

: 00:33:24

People needed to

integrate things together.

606

: 00:33:26

How do you integrate it together?

607

: 00:33:27

Well, you have APIs that

can talk to each other.

608

: 00:33:29

We have all software categories

integration platform as a service, you

609

: 00:33:33

know, build around that capability.

610

: 00:33:35

Um, And without going out on my soapbox

of why APIs are such an amazing thing

611

: 00:33:41

for, you know, I wish more software

vendors did a better job with more APIs.

612

: 00:33:45

They're at least headed

in the right direction.

613

: 00:33:47

I think having that converge with this

AI intelligence for agents to then

614

: 00:33:52

be able to take actions through those

APIs is that combination is going to

615

: 00:33:57

just unlock a tremendous amount of.

616

: 00:34:01

Justin: Is the agent, an AI of a kind

of different nature than what we're

617

: 00:34:05

getting when we're chatting with chat

GPT, or is it more just it's like

618

: 00:34:09

chat GPT with lots of extra context

empowered to take action through

619

: 00:34:14

APIs or through access to interfaces?

620

: 00:34:16

Scott Brinker: I think the thing

that takes it one step further is the

621

: 00:34:20

ability for it to be a feedback loop.

622

: 00:34:22

I mean, like, if you have an agent,

you know, it's not just about

623

: 00:34:26

like, oh, it comes to a plan.

624

: 00:34:28

It can kind of have done that already.

625

: 00:34:30

Alright, now the next step

was, okay, it can take action.

626

: 00:34:32

Action on that plan.

627

: 00:34:33

It can call, you know, these

APIs or primitive sense can

628

: 00:34:38

like manipulate the computer.

629

: 00:34:39

UI for you.

630

: 00:34:41

but then the step beyond that, that

gets really interesting is it can

631

: 00:34:45

then feedback into its decisioning.

632

: 00:34:49

What happened as a result?

633

: 00:34:51

Oh, this is the result I got.

634

: 00:34:53

That was not what I expected.

635

: 00:34:55

Okay, let me adjust my plan and I'm

going to try this other thing or, Hey,

636

: 00:34:58

wow, this looks like this is a problem.

637

: 00:35:00

I want to flag this and create an

exception, you know, for someone,

638

: 00:35:04

you know, that sort of feedback loop.

639

: 00:35:06

And again, this is all very early

with people experimenting with stuff,

640

: 00:35:10

but I think that's what starts to

make this really powerful because.

641

: 00:35:14

You know, we've, again, over the past

10 years, automation has been one

642

: 00:35:20

of those themes that's made its way

through pretty much every facet of

643

: 00:35:24

the tech stack, but certainly in rev

ops, marketing ops, sales ops, right?

644

: 00:35:29

I mean, marketing automation platforms,

the very name of these products, you know,

645

: 00:35:33

are about automation at the center of it.

646

: 00:35:35

but a lot of it's been this very

deterministic rules based automation.

647

: 00:35:39

where things get dicey is when

something goes wrong or it

648

: 00:35:43

doesn't work quite as expected.

649

: 00:35:45

Usually then we've had to sort of

like surface that up to then the

650

: 00:35:48

human administrator to figure it out.

651

: 00:35:51

to allow these AI agents to be

in a position where they can kind

652

: 00:35:55

of figure it out for themselves.

653

: 00:35:57

Boy, that's a huge breakthrough.

654

: 00:35:59

Justin: Maybe if you'll indulge me in

a thought experiment, because this is,

655

: 00:36:02

it's just a really interesting thing.

656

: 00:36:03

something in marketing operations

that is A common task, you know,

657

: 00:36:06

building out some kind of campaign or

program or email within a marketing

658

: 00:36:10

automation platform, whether that's

Marketo or HubSpot or something else,

659

: 00:36:13

quite often we have internal service

resources or teams that do that.

660

: 00:36:18

So a marketer says, Hey, I want an email.

661

: 00:36:19

They do a ticket, someone on the marketing

ops team will go and build that for them.

662

: 00:36:23

what would be involved if we wanted to

make an agent to replace that, which

663

: 00:36:26

it sounds like we could, whether the

agent was using APIs, which many of

664

: 00:36:29

these tasks are possible through APIs

or had to do it through the interface.

665

: 00:36:33

Like, how do you even get started

if someone wanted to do that?

666

: 00:36:44

Scott Brinker: Like, is that a

set of steps that an agent could

667

: 00:36:47

actually, you know, take over?

668

: 00:36:49

All right.

669

: 00:36:49

Well, so I've got to like, you know,

let, is it getting full copy or is

670

: 00:36:53

it only getting sort of a framework?

671

: 00:36:55

Does it adjust it?

672

: 00:36:55

Does it check it up against,

uh, you know, brand voice?

673

: 00:36:58

Okay.

674

: 00:36:59

Now, like, do I think about

like, okay, who am I doing

675

: 00:37:02

for like audience selection?

676

: 00:37:04

You know, who's the list?

677

: 00:37:05

What should be the criteria for that?

678

: 00:37:07

Um, okay.

679

: 00:37:08

Now I've, you know, got my email,

like, you know, what's the right

680

: 00:37:11

delivery schedule, you know, for

these folks, or there are other

681

: 00:37:14

competing, you know, messages.

682

: 00:37:16

How do we make sure, you know, we've

got the right throttling on that.

683

: 00:37:19

I don't know.

684

: 00:37:20

Maybe there's things of like a check

of like the accessibility, you know,

685

: 00:37:23

there's like a whole series of things

that you could imagine us doing that

686

: 00:37:27

with more traditional automation.

687

: 00:37:30

It's just.

688

: 00:37:30

Some of these things were so qualitative

in how they needed to be evaluated

689

: 00:37:35

and like traditional, you know,

machine learning, automation AI just

690

: 00:37:39

wasn't quite able to capture that as

well as, you know, these LLMs have

691

: 00:37:44

almost like the mirror opposite.

692

: 00:37:46

You know, they're particularly

good at some of those like

693

: 00:37:48

squishy, qualitative things.

694

: 00:37:50

I think if you mirror those two things,

it wouldn't all surprise me, uh, you know,

695

: 00:37:54

to have more and more of these agents

show up that can, yeah, kind of like take

696

: 00:37:58

a ticket from a marketer, basically roll

together, you know, the whole campaign.

697

: 00:38:03

Presumably for at least some time

here, then present it to a human who

698

: 00:38:08

will review it, make sure, like, okay,

yeah, this is, you know, what I wanted,

699

: 00:38:11

but if that reduces, the work and the

time, involved in putting one of those

700

: 00:38:15

campaigns together from again, like three

days to three hours, a huge, huge leap.

701

: 00:38:23

Justin: the fascinating thing about

that for me is like every tool

702

: 00:38:29

is like putting their own little.

703

: 00:38:31

AI features in place.

704

: 00:38:32

Like I use Zapier and you can go and

build his app the traditional way,

705

: 00:38:35

like drag your little things in and

configure them where you can type

706

: 00:38:39

something in and be like, and do it.

707

: 00:38:41

And it's doing an okay job of doing that.

708

: 00:38:43

But quite often as a professional

in that I'm maybe more inclined to

709

: 00:38:47

just do it myself a lot of the time.

710

: 00:38:49

But if every tool has those little

things, you're still needing a human

711

: 00:38:52

there to stitch it all together.

712

: 00:38:54

But now you have this agent that is

kind of outside of all those systems,

713

: 00:38:57

but can interact with all of them.

714

: 00:38:58

It's trained on your brand and company

specific policies and procedures and

715

: 00:39:04

can chat with the marketer on your

behalf and then go out and execute

716

: 00:39:07

and take feedback and make changes

like that's fairly revolutionary.

717

: 00:39:11

Like, are we, are we, do you

think that all the systems

718

: 00:39:13

are in place to do that today?

719

: 00:39:15

Or are we still like a step or two away

from an agent being that autonomous?

720

: 00:39:19

Scott Brinker: Yeah, I think a

lot of the ingredients are there.

721

: 00:39:22

, again, this is one of these things

where, just like we were talking earlier

722

: 00:39:25

here about, you know, the uncanny

valley of, you know, uh, AI video

723

: 00:39:29

editing and generation and whatnot.

724

: 00:39:31

you know, it's particularly when

you're talking about like actually

725

: 00:39:33

running these operations, you know,

for multi billion or multi billion

726

: 00:39:37

dollar, marketing organization,

your tolerance for error in there.

727

: 00:39:43

it's probably such that we're probably

a little bit of ways, you know, from

728

: 00:39:46

wanting to let, you know, the AI

agents run that stuff autonomously.

729

: 00:39:49

They've got some work to do to prove that,

the reliability is going to be there.

730

: 00:39:52

But I think directionally,

yeah, it makes a ton of sense.

731

: 00:39:56

Yeah, I mean, you know, it's an

interesting question as to will this

732

: 00:40:01

be a different layer of the tech stack?

733

: 00:40:06

I think one of the things that's

really Interesting right now is there's

734

: 00:40:10

not one agent platform out there.

735

: 00:40:14

There's not like there's

one home for agents.

736

: 00:40:17

pretty much that every

tool seems to be developing

737

: 00:40:20

capabilities for its own agents.

738

: 00:40:22

There's a standalone agents, the LLMs

themselves are creating agents, you know?

739

: 00:40:28

And so I think one of the most interesting

things is going to be like, okay, yeah,

740

: 00:40:31

we've got all these like different.

741

: 00:40:33

Agents out there, what's the, what's the

orchestration, will that be the existing,

742

: 00:40:40

uh, in common platforms, you know, that

have served as the orchestrators and

743

: 00:40:43

sort of current go to market automation,

orchestration, uh, Uh, I'm certain

744

: 00:40:49

everyone who is an incumbent there,

uh, wants that to be the case, you

745

: 00:40:53

Justin: Yeah.

746

: 00:40:53

Scott Brinker: but I think it

would also be again to, you

747

: 00:40:55

know, disruptive innovation.

748

: 00:40:57

Like, yeah, are there other places

where people might be trying to, like,

749

: 00:41:01

serve as the ultimate orchestrator,

of these different agent capabilities?

750

: 00:41:05

Personally, I would keep an eye on that.

751

: 00:41:08

Justin: I mean, so to an extent, you've

just asked yourself the question that I

752

: 00:41:11

wanted to ask next, which is, you know,

you, you referred to the importance of

753

: 00:41:15

this orchestration and you kind of called

it this coordinating centers of gravity.

754

: 00:41:20

I thought that was a really good phrase.

755

: 00:41:21

In this big ops environment that provides

cohesion and governance to the plethora

756

: 00:41:26

of apps and agents and automations.

757

: 00:41:28

And what does that look like?

758

: 00:41:30

is it just a workado on steroids that

kind of is able to provide the API

759

: 00:41:35

pathways for agents and other apps

that are on the periphery to interact

760

: 00:41:40

with each other and to take action.

761

: 00:41:42

Is it something else that we

don't really understand yet

762

: 00:41:44

or haven't conceived of yet?

763

: 00:41:45

What will it be?

764

: 00:41:46

Scott Brinker: Yeah.

765

: 00:41:47

Uh, and this would be one of the things

I will hesitate to make a prediction.

766

: 00:41:53

because I think there's old model,

you know, the way in which we would

767

: 00:41:56

historically have thought of this

is something that is a very clear

768

: 00:42:00

conductor, to the orchestra, uh, your

example of like, you know, workado,

769

: 00:42:04

you know, as a version of that, I

think what's interesting as a parallel

770

: 00:42:08

is to consider what has happened

if the data layer inside companies.

771

: 00:42:14

so for a while, what we had was we had

all these little mini databases attached

772

: 00:42:18

to every single individual different app

that lived in their own little silos.

773

: 00:42:21

you know, now with the, More and

more of those silos opened up.

774

: 00:42:26

We started to, in particular with things

like cloud data warehouses and lighthouses

775

: 00:42:29

and things like that, be able to get, you

know, data, you know, flowing more freely.

776

: 00:42:34

but then the question became like,

okay, we've now got all this data being

777

: 00:42:38

generated or being used by different

things throughout our organization,

778

: 00:42:41

but how is this being orchestrated?

779

: 00:42:44

you've now seen a multiple different

architectures, of how people

780

: 00:42:49

think about dealing with that.

781

: 00:42:50

You know, do we do it entirely

through one central authority?

782

: 00:42:53

Do we do it through this concept of

data products, you know, where different

783

: 00:42:57

teams, like, you know, manage things

as a product, you know, interfacing to

784

: 00:43:01

others, how much do you allow things

to be federated, you know, uh, where

785

: 00:43:05

do you allow there to actually be.

786

: 00:43:07

redundancies or, desyncs, because

in the grand scheme of things,

787

: 00:43:11

what you're losing, you know, from

that redundancy is more than gain

788

: 00:43:15

for, you know, by the efficiency

or the speed on some other level.

789

: 00:43:18

I suspect it's going to be, if I had to

take a guess, it would be something more

790

: 00:43:25

like that where, boy, this ability at the.

791

: 00:43:29

Services layer, not the data

layer, but the services layer.

792

: 00:43:33

it is going to be a free flowing mechanism

where like, hey, it's all in the cloud.

793

: 00:43:38

Anything can call anything else.

794

: 00:43:40

So it's going to be then about the

architectures we put around that.

795

: 00:43:43

Will some of them be centralized?

796

: 00:43:45

Yeah, I think those will

be the simplest models.

797

: 00:43:49

But could you imagine one where

people start to actually design

798

: 00:43:53

them as decentralized, systems?

799

: 00:43:56

Yeah.

800

: 00:43:57

Uh, could you could see that?

801

: 00:43:59

In fact, actually, again, like the

larger, larger you get as a company,

802

: 00:44:02

it's almost like, I mean, we see this

in the data products, you know, side,

803

: 00:44:05

it's just at some level, you know, a

hundred percent centralized solution

804

: 00:44:09

just becomes incredibly unwieldy, you

know, at this massive enterprise scale.

805

: 00:44:15

Justin: un, unmaintainable.

806

: 00:44:16

And to your point about like digital

coworker, I had this, this chat

807

: 00:44:19

with another founder who's working

on an AI product, like an AI that

808

: 00:44:24

has all the same context as you,

you know, that reads every email.

809

: 00:44:28

That is in your project management system

that's getting all the tickets that's

810

: 00:44:33

getting the meeting notes or is attending

the meeting could become could take

811

: 00:44:36

your job or could become this incredible

resource to sort of dialogue with, like,

812

: 00:44:42

all right, what are we going to do?

813

: 00:44:43

You know, like this, this kind of

always on partner thought partner

814

: 00:44:46

working through problems with you,

like every person could have one.

815

: 00:44:50

how much does that actually

amplify the ability?

816

: 00:44:52

Of a person and it's, I mean, it's hard to

think of that being centrally coordinated.

817

: 00:44:56

That's kind of like your, your AI

mini me in a way, sort of following

818

: 00:44:59

you around and partnering with you.

819

: 00:45:01

but that, sort of surrealistic kind

of futuristic idea doesn't actually

820

: 00:45:05

seem that far fetched now, given

everything we've talked about.

821

: 00:45:08

Scott Brinker: No, in fact, actually, this

is, I would argue, one of the use cases

822

: 00:45:12

I talk to who are now doing a lot with

AI, like it's one of the most reliable

823

: 00:45:18

use cases is having these AI again, like

whether it's feeding it all of your own

824

: 00:45:23

stuff, or quite frankly, having it feed.

825

: 00:45:26

Things of like, oh, well, this is

everything I've ever received from

826

: 00:45:29

my boss, you know, you know, and

being able to, like, have these

827

: 00:45:33

sorts of dialogues, you know, it's

like a Socratic method, almost

828

: 00:45:36

Justin: Yes.

829

: 00:45:37

Scott Brinker: uh, you know, just how

it changes our ability to look at things

830

: 00:45:41

from, like, a different perspective.

831

: 00:45:44

because it's both got that combination

of, uh, we won't say it's omniscient,

832

: 00:45:49

uh, you know, yet, but it's like,

boy, it, it, it does have more data

833

: 00:45:53

than pretty much any other human you

would, you know, talk to about that.

834

: 00:45:56

but also again, in an environment

here where there's, there's no

835

: 00:45:59

emotion on the other side, there's no

judgment, there's no, you know, like.

836

: 00:46:03

Privacy thing of like, yeah, now the AI

is going to go like, you know, whisper

837

: 00:46:07

behind my back about that conversation.

838

: 00:46:09

I mean, it just, it's this amazing

thing, I think, to give people

839

: 00:46:12

a lot of freedom to just like

creatively dialogue and explore.

840

: 00:46:16

Different ways of looking at

things, coming up with new ideas.

841

: 00:46:19

Justin: It really does feel to me that

we were creeping closer and closer to

842

: 00:46:22

having like the onboard ships, computer

of Star Trek, just available to us

843

: 00:46:27

where it's like, you're talking to

them, like computer, do this, do that.

844

: 00:46:31

Help me solve this problem,

become this, and an assistant.

845

: 00:46:35

do you ever sort of take a step like

sometimes when I think about these

846

: 00:46:37

things, I'm like, this is like, yes,

it could be weird and dystopian and

847

: 00:46:41

there could be bad outcomes, but

it's also just incredible like that.

848

: 00:46:43

This stuff exists right now.

849

: 00:46:45

Like what?

850

: 00:46:46

What timeline did we wander into that?

851

: 00:46:48

This is happening.

852

: 00:46:49

It is so weird.

853

: 00:46:50

Scott Brinker: Yeah.

854

: 00:46:50

you know, again, probably, probably

reaching too far back into the archives,

855

: 00:46:55

but you know, a very early Star Trek

movie was a Star Trek four where they

856

: 00:47:00

travel in time to like what at the time

was, you know,: 1980

857

: 00:47:05

and was it Scotty, you know, he

has to interact with one of the

858

: 00:47:08

computers and he's like trying

to talk to it at the computer.

859

: 00:47:10

You know, the guy hands him the mouse.

860

: 00:47:11

He's like talking to the mouse,

like, this is a huge, big laugh line.

861

: 00:47:15

Like, well, of course.

862

: 00:47:16

Justin: I think that's probably all

that we have time for today, but Scott,

863

: 00:47:19

thank you so much for chatting with me.

864

: 00:47:22

Super interesting.

865

: 00:47:23

Super exciting.

866

: 00:47:23

Again, I encourage everyone to read

the full report, which is obviously

867

: 00:47:26

a lot of detail that we didn't get

to cover, but thank you for, uh, for

868

: 00:47:29

putting that out and being such a great

resource to the marketing ops community.