WEBVTT

NOTE Sentence-level transcript of https://www.youtube.com/watch?v=YZQsWVeN3rE

NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/YZQsWVeN3rE.html. A cue ends where the next begins, or 2 s after its last word.

s1
00:00:01.309 --> 00:00:03.309
[music]

s2
00:00:12.400 --> 00:00:15.720
Well, hopefully this is the one talk that's different from all the others today.

s3
00:00:15.720 --> 00:00:18.960
So, let's start with audience participation.

s4
00:00:18.960 --> 00:00:21.600
Who here has kids?

s5
00:00:22.000 --> 00:00:23.320
Oh.

s6
00:00:23.320 --> 00:00:24.920
Okay, there's a couple.

s7
00:00:24.920 --> 00:00:30.480
Um Who here kind of has a parent-child relationship with Claude?

s8
00:00:32.080 --> 00:00:33.480
Okay, there we go.

s9
00:00:33.480 --> 00:00:34.840
I have a kid.

s10
00:00:34.840 --> 00:00:36.160
This is Auto.

s11
00:00:36.160 --> 00:00:42.360
Uh he learns to drool this week, which surprisingly is something that these things don't come out of the box with.

s12
00:00:42.360 --> 00:00:45.480
He also discovered his fingers, but he doesn't know about his hands.

s13
00:00:45.480 --> 00:00:51.040
So, they're fascinating and they're in front of his face, but he doesn't know how they got there and he can't bring them back.

s14
00:00:51.040 --> 00:00:56.160
But anyway, I finished these slides this morning at 2:00 a.m. and he decided that playtime started at 5:00.

s15
00:00:56.160 --> 00:01:01.760
So, we're going to see how the rest of this talk goes, but clearly I need practice telling bedtime stories.

s16
00:01:01.760 --> 00:01:04.640
So, we're going to do one of those first.

s17
00:01:04.640 --> 00:01:10.800
Once upon a time, a go-to-market team set out on a simple quest.

s18
00:01:10.800 --> 00:01:13.000
They wanted to know one thing.

s19
00:01:13.000 --> 00:01:16.000
How their business was doing.

s20
00:01:16.000 --> 00:01:22.840
Between them and the answer lay one small, tidy pile of data.

s21
00:01:24.240 --> 00:01:29.000
And a dragon, because of course the best quests always have dragons.

s22
00:01:29.000 --> 00:01:31.720
And every quest also has guardians.

s23
00:01:31.720 --> 00:01:39.560
The first was large, worried, and had a great many questions about their security posture.

s24
00:01:39.600 --> 00:01:48.120
They defeated it in the way that important bosses are usually defeated, not with a sword, but with paperwork.

s25
00:01:49.400 --> 00:01:51.800
The next monster was worse.

s26
00:01:51.800 --> 00:01:58.400
It had many heads, and no two of them could agree on what a single word meant.

s27
00:02:04.840 --> 00:02:09.920
To beat it, our heroes had to accept a hard truth.

s28
00:02:09.920 --> 00:02:16.480
The systems they trusted had never actually promised to tell them the whole story.

s29
00:02:16.480 --> 00:02:28.080
So, they stopped believing any single one of them and started building something underneath all of them, a place where the data finally agreed with itself.

s30
00:02:31.120 --> 00:02:33.720
And then came the final boss.

s31
00:02:33.720 --> 00:02:35.160
It didn't attack.

s32
00:02:35.160 --> 00:02:37.480
It did something even worse.

s33
00:02:37.480 --> 00:02:39.959
It asked one question.

s34
00:02:39.959 --> 00:02:43.160
The hardest one of all.

s35
00:02:44.520 --> 00:02:48.080
Now, our heroes had learned a few things by then.

s36
00:02:48.080 --> 00:02:55.880
They'd seen what happens when you send a brand new adventurer in with no guidance at all.

s37
00:02:56.240 --> 00:02:58.160
And they realized the secret.

s38
00:02:58.160 --> 00:03:00.640
You don't need a genius adventurer.

s39
00:03:00.640 --> 00:03:03.480
You need a well-prepared one.

s40
00:03:03.480 --> 00:03:07.600
First, you write down how your world actually works.

s41
00:03:07.600 --> 00:03:11.320
Then you give every adventurer a knowledgeable guide.

s42
00:03:11.320 --> 00:03:13.160
They can ask for directions.

s43
00:03:13.160 --> 00:03:16.360
And then the adventurers start teaching each other.

s44
00:03:16.360 --> 00:03:22.360
So that no one ever has to start from anything and nothing again.

s45
00:03:23.360 --> 00:03:25.680
So, they gathered up everything they'd learned.

s46
00:03:25.680 --> 00:03:34.959
The permissions, the definitions, the knowledgeable guide, the habit of asking for a second opinion before they believed anything.

s47
00:03:34.959 --> 00:03:39.360
And they forged all of it into a single tool.

s48
00:03:39.360 --> 00:03:47.959
And when they had finally reached that tidy data pile, they found out that the dragon had never been the villain.

s49
00:03:52.000 --> 00:03:56.880
And so, the kingdom entered a golden age of growth and creativity.

s50
00:03:57.040 --> 00:03:59.320
The end.

s51
00:04:00.480 --> 00:04:01.120
Just kidding.

s52
00:04:01.120 --> 00:04:02.960
That's only where the bedtime story ends.

s53
00:04:02.960 --> 00:04:04.760
Uh the actual talk starts here.

s54
00:04:04.760 --> 00:04:08.640
Because every part of that story is a real problem.

s55
00:04:08.640 --> 00:04:12.240
And I want to show you how three of them actually get solved.

s56
00:04:12.240 --> 00:04:16.920
So, our agenda for today is I'm going to talk a bit about the age of agentic GTM.

s57
00:04:16.920 --> 00:04:26.919
And then we're going to walk through three actual examples that we either have worked through as a company recently or that we've built into the product that we sell to our customers.

s58
00:04:26.919 --> 00:04:33.880
So, briefly, to establish credibility here, I'm going to give you 1 minute about our company, Upside.

s59
00:04:33.880 --> 00:04:42.560
We are the data layer for the age of agentic go-to-market, which means that we start by helping you get go-to-market data ready for AI.

s60
00:04:42.560 --> 00:04:49.280
And that means knowing what actually happened with a data foundation and a dashboard you can use.

s61
00:04:49.280 --> 00:04:57.919
And then helping you answer things that used to be impossible at scale with things like deep research or multi-touch attribution model.

s62
00:04:57.960 --> 00:05:05.320
We do a lot to help you make AI correct about your business so that it's not hallucinating and giving you trust issues.

s63
00:05:05.320 --> 00:05:08.760
And then we make sure that this data is in all the places that you want to use it.

s64
00:05:08.760 --> 00:05:15.800
So, a lot of things you used to spend money to buy are actually things that you, the person closest to the problem, can now build.

s65
00:05:15.800 --> 00:05:21.160
So, with that established, this is the age of agentic go-to-market.

s66
00:05:21.160 --> 00:05:30.400
We have this interesting situation right now where go-to-market teams have not historically been the place where you'd find a high density of builders.

s67
00:05:30.400 --> 00:05:32.520
At least compared to product and engineering.

s68
00:05:32.520 --> 00:05:38.720
Obviously, there are exceptions to this, but in general, most marketers and sales people were never writing a ton of code.

s69
00:05:38.720 --> 00:05:46.880
And that's very much not to say they weren't aware of their problems and didn't have ideas of how they would solve them, but the historical toolbox here was basically

s70
00:05:46.880 --> 00:05:50.120
spreadsheets and PowerPoint slides.

s71
00:05:50.120 --> 00:05:52.560
And AI has changed that.

s72
00:05:52.560 --> 00:05:56.000
Cloud does for building basically what the bicycle did for mobility.

s73
00:05:56.000 --> 00:05:57.360
It makes it accessible.

s74
00:05:57.360 --> 00:06:09.240
And so spreadsheets and slides are no longer the default business tool that everyone has available, which means that a lot of people who historically had ideas and didn't consider themselves builders

s75
00:06:09.240 --> 00:06:11.960
now can be.

s76
00:06:11.960 --> 00:06:17.520
Before AI, I would have described myself as technical enough to be dangerous.

s77
00:06:17.520 --> 00:06:18.760
I can read code.

s78
00:06:18.760 --> 00:06:19.880
I can write code.

s79
00:06:19.880 --> 00:06:29.840
You probably wouldn't want my stuff in production generally if you could help it, but I'd be interested who in this room today considers yourselves an actual engineer.

s80
00:06:30.320 --> 00:06:31.400
Okay, awesome.

s81
00:06:31.400 --> 00:06:33.120
Thank you for being brave.

s82
00:06:33.120 --> 00:06:39.400
Who would tell themselves that they are also technical enough to be dangerous?

s83
00:06:40.160 --> 00:06:41.120
All right.

s84
00:06:41.120 --> 00:06:44.040
And who doesn't read code at all?

s85
00:06:44.640 --> 00:06:47.440
Okay, thank you for being brave in the back.

s86
00:06:47.440 --> 00:06:51.720
Uh those categories should have been mutually exclusive collectively exhaustive.

s87
00:06:51.720 --> 00:07:03.840
So if anybody didn't raise your hand, I recognize that there's no free coffee here and I am the first to talk after lunch, but the thing about AI is it makes all of us technical enough to be dangerous.

s88
00:07:03.840 --> 00:07:15.160
What AI has done for those second two groups is basically provided you with an infinite supply of valedictorian interns with computer science degrees who can help you implement all of the solutions

s89
00:07:15.160 --> 00:07:19.360
that you probably know better than any engineer that you'd work with on a different team.

s90
00:07:19.360 --> 00:07:26.480
And that incredibly speeds up iteration, and it also kind of makes me feel like I'm Harry Potter walking around with a magic wand for the first time.

s91
00:07:26.480 --> 00:07:35.160
But essentially, when you set interns loose on problems led by non-technical people to build things, you're going to get some interesting results.

s92
00:07:35.160 --> 00:07:39.800
And a couple of years ago, I remember that everyone was talking about the AI hallucination problem.

s93
00:07:39.800 --> 00:07:44.920
Somehow that big word seems to not come up as often anymore, but I think we have its older sibling now.

s94
00:07:44.920 --> 00:07:46.840
It's a trust problem.

s95
00:07:46.840 --> 00:07:52.480
And if you ask Claude to do something like report on revenue, it doesn't say, "I'm not sure."

s96
00:07:52.480 --> 00:07:54.280
It says, "Here you go."

s97
00:07:54.280 --> 00:07:58.040
And it gives you a wrong answer that looks exactly like being right.

s98
00:07:58.040 --> 00:08:12.200
So, a lot of the talks this week I think have been around the bleeding edge of technical optimizations, and a lot of the influencer fluff you see on LinkedIn and Twitter tells you about the magic improvements if you, you know, do your prompt with exactly the right incantation.

s99
00:08:12.200 --> 00:08:17.520
I think both of those things might well be true, but I don't generally have time to do either of them.

s100
00:08:17.520 --> 00:08:27.960
And in practice, I find that the most practical thing for working with AI is something that we all know pretty intuitively anyway, because we have a lot of experience

s101
00:08:27.960 --> 00:08:30.800
working with other people.

s102
00:08:30.800 --> 00:08:33.599
And in that framing, establishing trust actually isn't new either.

s103
00:08:33.599 --> 00:08:36.120
We already know a lot about how to do this for people.

s104
00:08:36.120 --> 00:08:41.640
So, the main thesis of the talk today is actually, when in doubt, manage your agents like other humans.

s105
00:08:41.640 --> 00:08:46.440
If you were helping a team of humans figure out how to do something, great.

s106
00:08:46.440 --> 00:08:48.880
Do that with your AI agents, too.

s107
00:08:48.880 --> 00:08:52.520
And if you take nothing else from this talk, this is my one practical tip.

s108
00:08:52.520 --> 00:08:55.760
Use commander's intent when you prompt.

s109
00:08:55.760 --> 00:09:05.520
This is something that comes out of the armed forces doctrine, but basically tell your agents why you want them to do something, and they will do it a lot better.

s110
00:09:05.520 --> 00:09:07.360
And by the way, this works for humans as well.

s111
00:09:07.360 --> 00:09:12.560
People don't generally like being micromanaged, so give them commander's intent and you'll get great outcomes.

s112
00:09:12.560 --> 00:09:21.680
But, beware because the agents have been trained on material from other humans, and so they like to micromanage themselves, and that's not usually what you want.

s113
00:09:21.680 --> 00:09:24.800
Don't tell Claude to improve itself, you'll get micromanagement.

s114
00:09:24.800 --> 00:09:28.640
You have to pull it back and say, "Remember, we're talking about the why."

s115
00:09:28.640 --> 00:09:29.920
Anyway, onto the examples.

s116
00:09:29.920 --> 00:09:34.760
So, the first one here is I needed to redo our entire website recently.

s117
00:09:34.760 --> 00:09:46.880
And at my previous company, which was 600 people and I led the product marketing team, this would have meant several product marketers, a designer, a couple of consulting PMs, a web team, and it would have taken 2 months.

s118
00:09:46.880 --> 00:09:53.600
And so, in general, with AI, sometimes you just kind of have to give it a go to see if it's improved enough from the last time you tried.

s119
00:09:53.600 --> 00:09:59.600
And so, I did that, you know, YOLO mode, here are the sources, go read everything, please give me my website.

s120
00:09:59.600 --> 00:10:04.840
And it was obviously a fail, even with using a nice plan mode in Claude first.

s121
00:10:04.840 --> 00:10:08.160
I'm sure this will work eventually, but it doesn't today.

s122
00:10:08.160 --> 00:10:12.120
So, you have to do some scaffolding.

s123
00:10:12.120 --> 00:10:19.040
And by scaffolding, what I mean is you have to tell it what to know about your business.

s124
00:10:19.040 --> 00:10:23.080
So, I maintain this list of anchor assets for our entire company.

s125
00:10:23.080 --> 00:10:31.960
I think documentation is something that humans generally should have used anyway, but that was often something that was also generally delegated and forgotten.

s126
00:10:31.960 --> 00:10:41.760
Anyway, I maintain these anchor assets, and one of the really important ones here is the product capabilities reference, because on the website, you want to talk about your product.

s127
00:10:41.760 --> 00:10:46.240
And this is something that I actually did use AI to compile.

s128
00:10:46.240 --> 00:10:54.000
So, this wasn't me sitting here and typing all of these out, but if you pull up one of these cards, you'll see that it tells you what does the product capability

s129
00:10:54.000 --> 00:11:00.760
do, why it matters for a bunch of the personas, which are also defined in these anchor assets.

s130
00:11:00.760 --> 00:11:04.200
And then, it also shows the track record of how Claude found all of this.

s131
00:11:04.200 --> 00:11:07.760
So, these are citations from across every system that it's connected to.

s132
00:11:07.760 --> 00:11:13.800
I can follow them back if I need more details, but that helps me know that it didn't hallucinate the important parts.

s133
00:11:13.800 --> 00:11:24.080
And if you look back at what we had on this website, that actually ladders pretty much directly up to what was on the homepage, and then if I were to go into these product pages, it also ladders up there.

s134
00:11:24.080 --> 00:11:29.160
So, the step here was define the structure first, and then turn Claude loose.

s135
00:11:29.160 --> 00:11:32.000
Don't try and YOLO it from the beginning.

s136
00:11:32.000 --> 00:11:39.640
Uh if this kind of thinking is new to you on product capability references, by the way, this I think is probably the best book on how to do it.

s137
00:11:39.640 --> 00:11:45.480
There's also a nice summary here that somebody wrote up online, particularly steps five to six.

s138
00:11:45.480 --> 00:11:51.080
I think are the important ones for this, but the whole book is worth reading, honestly.

s139
00:11:51.120 --> 00:11:59.040
Um this ticker is staring me in the corner of the space, so I'm going to give you five more seconds to scan that and then we're moving on.

s140
00:11:59.560 --> 00:12:05.600
Two other important things are general D Sloop scale, there are a ton of these floating around, and also consult Persona Bench.

s141
00:12:05.600 --> 00:12:12.920
I have a set of agents that inhabits each of the important personas and I can have them review something on demand and give their opinions.

s142
00:12:12.920 --> 00:12:15.520
But the second one is our radiant librarian.

s143
00:12:15.520 --> 00:12:22.120
So this is something that actually exists inside our product, and I'm going to show you how it works with a diagram.

s144
00:12:22.120 --> 00:12:23.480
So this is the librarian.

s145
00:12:23.480 --> 00:12:28.160
If you, the user, have a question, "How much pipeline did we create in Q1?"

s146
00:12:28.160 --> 00:12:31.560
Then your agent is happily going to go off and say, "Oh, okay.

s147
00:12:31.560 --> 00:12:33.960
Well, quarter means, you know, January to March.

s148
00:12:33.960 --> 00:12:35.839
I'll just look at created date."

s149
00:12:35.839 --> 00:12:38.240
And that is probably not correct.

s150
00:12:38.240 --> 00:12:42.400
So what we actually have it do is consult the librarian first.

s151
00:12:42.400 --> 00:12:50.680
And the librarian has access to documentation and the library of knowledge items about your company and the schema of prior failed queries.

s152
00:12:50.680 --> 00:12:55.800
And so it basically gives your agent a just-in-time memory of all the important things.

s153
00:12:55.800 --> 00:12:59.480
So we know that the fiscal year here is actually February through April.

s154
00:12:59.480 --> 00:13:03.760
And you pipeline means things that only meet me uh stage two or later.

s155
00:13:03.760 --> 00:13:10.240
And then you get a nice trustworthy answer with citations back rather than something that you discovered for the first time.

s156
00:13:10.240 --> 00:13:17.800
I was going to do a live demo of this, but I think we're going to skip that for the purposes of time and move on to example number three,

s157
00:13:17.800 --> 00:13:21.560
which is what we call our jury and judge workflow.

s158
00:13:21.560 --> 00:13:30.120
So we started this company to solve multi-touch attribution, which is kind of like the holy grail of anything in go-to-market.

s159
00:13:30.120 --> 00:13:32.800
And it took us 2 years to get anywhere close.

s160
00:13:32.800 --> 00:13:39.280
But, along the way, we built this AI-native data layer and then discovered that that was actually what everybody needed for everything else.

s161
00:13:39.280 --> 00:13:48.640
And when we came back to multi-touch attribution, enabled by Opus, we actually get really interesting results that are actually pretty impressive.

s162
00:13:48.640 --> 00:13:58.680
So, we're going to look at how this workflow works because there's this classi- classification of challenges in go-to-market where there is no empirically correct answer.

s163
00:13:58.680 --> 00:14:01.200
And we have a model for how we deal with that in the real world.

s164
00:14:01.200 --> 00:14:04.240
It usually involves a trial by a jury of your peers.

s165
00:14:04.240 --> 00:14:11.320
So, here we're doing a very similar thing where you might say, "I'd like attribution on the Acme deal."

s166
00:14:11.320 --> 00:14:15.800
And your agent is not going to go off and immediately answer this on their own.

s167
00:14:15.800 --> 00:14:28.560
They are going to spin up a team of independent analysts who all look at the data independently and come up with an evidence-cited opinion for what they think the attribution credit of that deal should be.

s168
00:14:28.560 --> 00:14:38.920
And none of these are necessarily correct, but they are independent research that then goes back to the consensus judge who says, "Oh, I'm not treating these as fact.

s169
00:14:38.920 --> 00:14:46.600
I'm treating them as input, and I'm going to weigh the reasoning quality of each of these analysts, and then I'm going to help you come up with the final version here.

s170
00:14:46.600 --> 00:14:51.360
And if there's not enough consensus, then I'll escalate and expand the jury.

s171
00:14:51.360 --> 00:14:53.839
But, my job is not to do research on my own.

s172
00:14:53.839 --> 00:15:01.880
It's to say, "Here's the final result based on a whole team of independent analysts doing a really good job of research."

s173
00:15:01.880 --> 00:15:04.880
And this is something that we see in the real world.

s174
00:15:04.880 --> 00:15:06.440
We would do it with humans, too.

s175
00:15:06.440 --> 00:15:15.240
Turns out multiple researchers with somebody who helps at the end is better than a single person kind of perseverating on that forever.

s176
00:15:15.240 --> 00:15:17.960
So, thus ends the three examples.

s177
00:15:17.960 --> 00:15:20.400
But, I have a bonus.

s178
00:15:20.400 --> 00:15:27.160
So, bonus is right here, agent tears.

s179
00:15:27.160 --> 00:15:31.360
In my experience, you can't fix stupid.

s180
00:15:31.360 --> 00:15:37.640
So, this basically means friends don't let friends use really bad harnesses or low intelligent models for important work.

s181
00:15:37.640 --> 00:15:43.480
An example of this is Slackbot released its MCP client functionality a couple weeks ago.

s182
00:15:43.480 --> 00:15:47.760
So, you can now hook Slackbot up and it will talk to MCPs and I thought, "Oh, that's awesome."

s183
00:15:47.760 --> 00:15:57.480
Now, everyone who has Slack can suddenly consult the Upside Librarian and I was so excited to try it and it turns out that this is horrifically stupid.

s184
00:15:57.480 --> 00:16:13.920
So, basically, the general case here is any AI product where it's been crowbarred into a pre- subscription model is probably not something that you should be using for anything important because the margin on those plans just doesn't leave enough space for an intelligent reasoning model to work.

s185
00:16:13.920 --> 00:16:20.240
So, find something that is at least tier two here and tier two means it's got to be a powerful model.

s186
00:16:20.240 --> 00:16:24.520
You need it to have attributes like sub agents, plan mode, full MCP support.

s187
00:16:24.520 --> 00:16:28.280
It should be able to use file editing and there are a lot of these out here.

s188
00:16:28.280 --> 00:16:35.080
So, do not let your team just use the chat GPT web interface and expect that it will be a great result.

s189
00:16:36.440 --> 00:16:38.080
And that ends the talk.

s190
00:16:38.080 --> 00:16:40.400
So, I hope you stick around for the rest of the track.

s191
00:16:40.400 --> 00:16:47.600
I know Jeff from XAI is coming up later and he's got a really bang of a talk about how they basically created Upside for all the rest of the world's data

s192
00:16:47.600 --> 00:16:50.800
and I shall be outside for a few minutes if you have any questions.

s193
00:16:50.800 --> 00:16:52.960
Thank you.