How AI agents are changing GTM engineering

0:00 And you talk about like GTM is code, but

0:01 I always think like sales and marketing

0:03 really can't be reduced to code.

0:05 >> Can we abstract all that away? So it's

0:07 find a contact, add them to the CRM, and

0:10 send an email. It's just three lines of

0:11 code. That unblocks a lot of

0:13 non-technical users or less technical

0:15 users from accomplishing things that

0:17 used to take a team of like 10

0:18 engineers. What we've seen work well on

0:20 that end is human in the loop approvals

0:23 of basically just draft everything for

0:24 me. Then I go in provide feedback or

0:26 make the the final approval before

0:28 something gets done. I have a one of my

0:31 first 10 calls for initial product arrow

0:33 where it was the CRO of uh the company

0:36 Service Titan, but he was like, "Yeah, I

0:38 would not do what you're doing." And I

0:40 ignored that one sentence. He focused

0:42 too much on the product instead of the

0:44 person and the problem that person is

0:46 facing. So, I got to shout out our

0:48 sponsor for this episode, kingos.com.

0:51 It's a B2B data API by Outbound Sync. If

0:54 a single data vendor keeps getting stuff

0:55 like parent companies wrong, it keeps

0:57 showing international branches as like

0:59 different accounts, or if your reps just

1:01 don't trust the data, you should check

1:03 out King Minos. It pulls from multiple

1:05 top tier vendors and then uses an agent

1:07 to decide what information to surface to

1:09 you and your tools inside their API. You

1:11 can connect it to Claude, Cursor, or any

1:13 of your other AI tools using their MCP

1:15 or their CLI in seconds. Use my code KFC

1:19 20 to get,250 free credits and 100 more

1:22 credits every day. And you don't even

1:23 need to use your card to sign up. And if

1:25 you go to kingbos.com right now and do

1:27 this, they'll even double your first

1:28 purchase up to 10,000 credits. Hi, how's

1:31 it going, man?

1:32 >> Pretty good, Ben. Thanks for having me.

1:36 >> Awesome. Um, well, thanks so much for

1:37 coming on. Uh, so your X bio says that

1:40 you're a Paddington 2 evangelist. Is

1:43 that accurate?

1:44 >> It is arguably the best movie ever made.

1:47 >> Why Why is it so good? it uh unique Brit

1:52 like it's the British humor. Uh it's

1:54 it's got the same Pixar thing where it

1:56 is both funny for children and adults. I

1:59 have not shown it to anyone who has not

2:01 said it's one of the best movies ever

2:03 made, but also you might cry at the end.

2:06 So, for it to be both humorous for

2:08 adults and children and also get like 50

2:11 60% of people to cry at the end, uh, it

2:14 takes a special amount of skill that I

2:16 don't think any other movie has ever

2:18 accomplished. And, uh, Hugh Grant is the

2:22 villain, and he is an extremely

2:23 effective cross-dressing

2:26 nun villain.

2:29 >> Have you seen it? No.

2:30 >> Have you seen Oh my god. I actually

2:33 heard about it. It's a a joke. in that

2:36 movie with Nicholas Cage, the unbearable

2:38 weight of massive talent. It was just

2:40 like a side joke and it's like, "Oh

2:42 yeah, petting two two was the best movie

2:43 ever made." And then I looked up the

2:45 ratings and it's 99% on Rotten Tomatoes.

2:48 You're like, "Oh, well that's pretty

2:49 impressive for both audience and uh

2:52 critic score." So you're like that

2:53 that's pretty rare. Uh and then I went

2:55 and I watched it. I'm like, damn, they

2:57 were 100% right on that bit. The entire

2:59 bit is accurate. Yeah.

3:01 >> Okay. Well, I might have to check it

3:02 out. The last movie that almost made me

3:04 cry was probably Nolan's Odyssey. So,

3:09 >> way outside the bounds of GTM

3:11 podcasting, but

3:12 >> there's there's a lot of similarities

3:13 between the two movies. That's all I'm

3:15 going to say.

3:16 >> Oh, okay. I mean, I feel like every

3:19 movie, you know, follows kind of like

3:21 some variation of the hero's journey or

3:22 whatever. So,

3:23 >> yeah, that's that's what we're gonna

3:25 see. And Paddington's in prison for half

3:27 of it.

3:28 >> Oh, okay. Does he have to fight anybody

3:31 for a top punk? He does.

3:34 >> Okay.

3:35 >> You will. I can't ruin anymore.

3:38 >> Yeah.

3:38 >> Okay. So, um, a lot of people uh may or

3:42 may not know what you do. I'm going to

3:44 give you a really bad overview of what

3:46 Jai does, but he's got this little SAS

3:51 that basically uh makes uh

3:56 your coding agent do stuff for your for

3:59 your GTM stuff. Is that accurate? You

4:02 want to give a little bit more of an

4:03 accurate description?

4:04 >> No, that that's basically it. That's

4:06 that's what we're doing. Just the the

4:08 elves in the machine that make you good

4:09 at go

4:11 in the machine.

4:13 >> Yeah. No. Uh the the problem that we're

4:15 solving is essentially uh historically

4:18 it's been very hard to express go to

4:21 market tasks in code. So uh like the

4:24 most fundamental level if you can

4:26 express it in code, coding agents are

4:28 very good at doing it. uh a lot of what

4:31 go to market tools uh do is they

4:33 abstract like sending an email or

4:35 reading state about a customer but

4:37 there's a hundred different CRM they're

4:39 all kind of unique so your options are

4:41 build hundreds of integrations uh buy

4:44 data from all these enrichment data

4:46 providers and then piece it all together

4:47 in a system that you spin up yourself uh

4:50 or can we abstract all that away so it's

4:53 uh find a contact add them to the CRM

4:55 and send an email it's just three lines

4:56 of code and then once we've abstracted

4:58 it to code. Uh now any set of steps that

5:02 you want to do, a coding agent can test

5:05 all the possible ways of solving that

5:07 problem and find the optimal one for

5:09 you. So by turning it into code, you've

5:10 now made it way easier for coding agents

5:12 to solve the problems. Uh and that

5:17 unblocks a lot of non-technical users or

5:20 less technical users from accomplishing

5:22 things that used to take a team of like

5:24 10 engineers at a company like RAMP uh

5:26 to to execute on. So, um I know you guys

5:30 had done something before this called

5:31 Arrow, right?

5:33 >> Yeah.

5:33 >> Yeah. And and I one of the things I

5:35 really wanted to talk to you about was

5:36 like finding product market fit. Um

5:39 because obviously you guys are a

5:40 venturebacked startup. I don't know how

5:41 much money you raised. It was like two

5:43 or three million or something like

5:44 something like that.

5:45 >> Um

5:47 >> uh

5:48 >> and I know from like my own experience

5:50 like working at a early stage, no you

5:52 guys are farther along than I was at my

5:55 the startup that I was at. um uh where

5:58 we were like pre-product market fit. Uh

6:00 we were basically pre-revenue for most

6:02 of the time that I was there.

6:04 >> Mhm.

6:04 >> Like how are you thinking about finding

6:07 product market fit? And I guess like if

6:09 uh another founder or maybe somebody

6:11 who's thinking about starting a SAS

6:12 company is out there, like what would

6:14 you tell them about how to find it? I

6:17 know if you knew how to find it, you'd

6:19 write a book on it and just go consult

6:22 uh for people to help them do that. So I

6:24 don't expect you to have all the

6:25 answers, but yeah,

6:26 >> you're operating in a very interesting

6:28 space for for talking about this.

6:30 >> No, absolutely. That that's a great

6:31 question. I think the if you did want to

6:34 read a book about potentially getting

6:36 there, I think the the guy that has

6:38 captured what I felt most accurately and

6:41 we we followed some of what he said and

6:44 it probably saved us six months. Uh his

6:45 name is Rob Snyder and it's called like

6:48 uh the power of pull. Uh and basically

6:51 like

6:51 >> Yeah. So the the mistakes that most

6:53 founders make and I I talked to a lot of

6:55 founders who are in the exact same spot

6:56 as as we were. Uh there's there's a

6:59 couple things that commonly happen. One,

7:02 you have a lot of conviction on what you

7:04 think is the right way to do something.

7:06 So you will kind of force people's

7:08 opinions into what you think is right.

7:11 Uh I have a call uh one of my first 10

7:15 calls for initial product arrow where it

7:18 was the CRO of uh the company Service

7:21 Titan and we have like a friendly

7:23 relationship and he was like yeah I

7:26 would not do what you're doing basically

7:27 like in in the nicest words he's like

7:29 this doesn't really seem like a good

7:31 idea and I ignored that one sentence.

7:34 He's like don't don't sell to CRO don't

7:36 solve this problem we don't really care

7:38 about analytics. I I'll talk about what

7:39 era was in a second, but uh he just said

7:42 it directly and I ignored that sentence

7:44 and I listened to all the positive

7:45 feedback he gave about what we said and

7:48 I was like, damn, he really tried to

7:49 tell us so early on that we were kind of

7:52 going the wrong direction. Um and I

7:54 think it's like your internal biases of

7:56 what you want to be true. Uh you assume

7:59 that what your experience is is what

8:01 everybody else's experience is. And so

8:02 that that's the hard thing to skip out

8:04 of. You have high conviction, but you're

8:06 almost always wrong. uh very few people

8:08 are high conviction and right at that

8:10 stage and if they are they're like the

8:14 um folks who just spun out a tool from

8:16 like Facebook or Uber and then just

8:19 productized it like those are the people

8:21 have conviction because it's it's a

8:22 clear problem. Uh the other thing that

8:24 happens is you focus too much on the

8:26 product instead of the person and the

8:29 problem that person is facing and

8:31 problems almost too g general as well.

8:34 It was a project so we could have this

8:36 like nice to have product which was

8:38 Arrow. Uh Arrow was essentially very

8:41 similar to what we're doing now but

8:42 there were two additional pieces to it.

8:45 >> One we essentially required a data

8:47 warehouse. We assumed that data

8:49 warehouses were going to be the brains

8:51 of an enterprise org and that's where we

8:54 needed to have everything live and

8:56 that's what we needed to pull and source

8:57 from. There's a couple other companies

8:59 that have this this thesis. uh that is

9:01 true but it's a very small number of

9:03 companies that is true for that don't

9:05 have their own uh team to solve it or

9:08 actually know that that is how they want

9:10 to solve it or have that opinion. So you

9:12 have this it's basically a bad business

9:14 decision to to focus on that market.

9:16 >> Um okay so so one thing uh I I've always

9:19 super appreciated people who like give

9:21 you really candid feedback. I actually

9:22 got um super candid feedback on my sales

9:25 process the other day from somebody um

9:27 >> and they were like, "Well, uh you know,

9:29 you really need like a deck and I like

9:32 >> professionalize your sales process if

9:34 you're not just going to be getting a

9:35 bunch of like inbound referrals from

9:37 like your buddies." And I'm like,

9:39 >> "Okay, that's actually really helpful.

9:41 Um that's not the kind of stuff I think

9:43 about." Um

9:44 >> no, absolutely.

9:45 >> Um but I'm curious like what what does a

9:46 like an ideal like team look like for

9:49 you guys? because we have clients who

9:50 use deep line and they're not like super

9:52 technical but I just wonder about this a

9:54 lot from like um I don't know from all

9:56 sorts of angles like um you have to kind

10:00 of understand the basic mechanics of

10:02 deep line to like buy it like would you

10:05 agree with that or or do you think like

10:06 a total like Neanderthal could just be

10:08 like oh I'm going to go deep line

10:11 >> yeah I think the Neanderthalss uh can

10:14 can in fact use deep line

10:17 >> good news for all you Neanderthalss out

10:18 there Yeah. Yeah. The uh the distinction

10:22 I would say is they have to be

10:23 comfortable in a coding agent and kind

10:25 of like letting letting Jesus take the

10:27 wheel of to sorts of like being able to

10:30 say like I'm just going to describe

10:32 something I have trust or confidence

10:34 that it's going to do something and that

10:36 will be in the right direction and then

10:38 I'll iterate and and get to the final

10:39 state. um if that like mental model has

10:43 been captured which is true for both

10:44 like technical and nontechnical folks I

10:47 don't think that's necessarily the

10:48 breakdown of it's basically like how AI

10:50 pilled are they or how how cloudp are

10:52 they uh and if we can get to a really

10:54 good uh we have tons of users who are

10:57 power users who don't understand every

10:58 detail of what's happening but they know

11:00 how to define a test or an output an

11:03 outcome uh to test against

11:05 >> and so I think we definitely skew a

11:07 little bit more towards more technical

11:09 users just because of this like code

11:11 abstraction and being able to commit

11:13 everything to a GitHub repo. That's an

11:15 important feature for more technical

11:17 teams. But the average RevOps person

11:19 probably doesn't care about that yet.

11:21 They will eventually and that that's

11:22 kind of our bet is more things are going

11:24 to head towards this code first approach

11:27 because uh intuitively it just makes

11:29 sense for agents to be able to

11:31 communicate with each other about go to

11:33 market. They need a common language.

11:34 that code is going to eventually be the

11:36 the right uh form factor for for that

11:39 type of information. Uh code plus

11:41 context in in another format. Um that

11:46 doesn't require someone. So you can go

11:48 in, you can just toss in a prompt.

11:50 Deepline will get you a solution and

11:52 it'll get you potentially the best

11:54 solution uh depending on what you're

11:58 what you're optimizing for. If you're

11:59 optimizing for coverage or uh coverage

12:02 at any cost or cost at a specific target

12:06 of coverage, you have to give it that

12:07 guidance. So, it's how comfortable are

12:09 you with a coding agent uh or an agent

12:11 uh versus how comfortable are you with

12:13 code uh would be the distinction I I

12:16 would make. Um

12:18 does that answer your question?

12:19 >> Yeah. No, it does. it. Um, it also one

12:22 of the other things I wanted to ask you

12:23 about is like

12:24 >> I think you're

12:26 this is an interesting market because

12:27 your total addressable market is

12:28 actually like getting bigger every day

12:31 which is almost like what you want right

12:32 like um when I was working at SIFT um we

12:35 were almost like too early for some of

12:37 the things that we were doing and I

12:38 won't go into all the details but it was

12:40 an AI tool for uh managed service

12:41 providers which is just like IT firms

12:43 and stuff

12:44 >> and they were still like very AI

12:46 suspicious and and I feel like if we

12:49 started it now, not even that long. Not

12:51 even that long later, right? Like 18

12:53 months later, the company started in

12:55 2023,

12:56 >> whatever. Like if it had started in like

12:58 2025,

12:59 >> like I feel like we would have like

13:00 killed it. Um we had other issues other

13:03 than that, but um

13:05 >> yeah,

13:06 >> you know, I wonder like uh

13:08 like what what do you think the total

13:10 addressable market on this is like

13:12 today? Um because I am obviously very

13:16 comfortable in like cloud code and stuff

13:18 but um

13:19 >> I don't know like what what do you you

13:21 tell your investors?

13:22 >> Yeah, I think right now we're seeing

13:24 it's about 20% of the venturebacked uh

13:29 market series BD generally teams that

13:32 are building uh or built their go to

13:35 market motions in the last like year or

13:37 so. uh that market is kind of the the

13:40 prime for this right now because they

13:42 are generally hiring technical folks

13:43 when they're building out their their

13:45 stack to start. Uh or you have to have

13:47 enough engineers within the company that

13:49 one engineer working on go to market is

13:51 not unusual. Uh and that tends to be

13:54 companies like modal cognition, listen

13:56 labs, aoka, these like kind of

13:58 hyperrowth, series A to series B to D is

14:02 is where we're getting a lot of

14:03 traction. And then uh before that they

14:06 tend to not have they might have like

14:08 one AI pled person uh and that's enough

14:10 to get a ton of value out of Deepline,

14:12 but the people are going to get like a

14:13 100x value or are going to be in that

14:16 market. And it generally is about 10 to

14:18 20% of go to market engineers and

14:21 applied AI go to market that like uh

14:24 more technical side of go to market

14:26 engineering and yeah 20% seems like the

14:30 pretty consistent metric that we get to

14:32 uh when we go across kind of all

14:33 ventureback startups. Uh I think it's

14:35 going to keep growing especially on the

14:37 later stage side. I think the the ROI of

14:40 one engineer with very

14:44 um opinionated approaches to go to

14:47 market uh infrastructure is now

14:49 completely outsized. Like the the ROI on

14:52 that person is insane. So you're going

14:54 to get more demand on that side. I think

14:56 series A if you're pre-product market

14:58 fit uh before that there's like systems

15:01 don't necessarily add as much value as

15:03 just like getting out there and doing

15:05 stuff. I think that side uh we won't see

15:08 as much growth on this demographic. I

15:10 think people generally will get more

15:12 technical, but it's not going to be this

15:14 like very more technical archetype. This

15:16 podcast is brought to you by

15:17 peradvisor.com. Pero Advisors is the

15:19 premier GTM engineering and revenue

15:21 operations partner for ambitious private

15:23 equity and venture capital backed

15:24 companies. If your emails are landing in

15:26 spam, if your team can't agree on what a

15:28 good account is, if your reps are

15:30 spending hours calling people who don't

15:31 answer the phone, then hit me up. We'll

15:33 see what we can do for you. We find

15:34 every buyer that you could ever sell

15:36 into. We make sure that there's at least

15:37 some dashboard out there that shows you

15:39 what line of business you should be

15:40 doubling down on. And then we build the

15:42 systems to reach out to these people and

15:44 the agents that automate all the

15:45 drudgery. And the end result is that we

15:47 run outbound with you that turns into

15:49 real pipeline. Learn more at

15:50 apparadvisors.com

15:52 and thanks for watching the pod. So, um,

15:54 one thing and, uh, I'm sure you guys

15:56 always get like comparisons to Clay or

15:58 whatever, so I won't belabor the point

16:00 there, but like, um,

16:02 >> so they have like the whole like GTM

16:03 engineering thing, and I've obviously

16:05 like leaned into that for for various

16:07 reasons, like

16:08 >> branding, marketing. I think it's a good

16:10 I think it's a good uh, you know, it's a

16:13 good way to simplify a complicated

16:15 concept. And you talk about like GTM as

16:17 code, but I always think like you know

16:19 like sales and marketing really can't be

16:22 like reduced to code. Like it's a

16:25 >> it's a you know you're trying to form

16:27 like an emotional sort of like

16:29 connection between two people so that

16:30 they can like come to the truth or

16:32 whatever and see like hey is this a good

16:34 beneficial transaction for like both of

16:36 us. And obviously getting contact

16:39 information and stuff like that is very

16:41 important like I this is stuff I do all

16:43 day right. Um,

16:44 >> yeah,

16:45 >> but this may be too esoteric of a

16:47 question, but like how do you think that

16:49 deep line is helping people like get to

16:52 that like faster? Like

16:54 >> I mean you can even describe a

16:56 interesting data workflow or something.

16:58 But like my experience here is that um

17:03 all this like AI stuff, cloud code,

17:05 cursor, whatever, like all the stuff

17:07 that I do, like you can get 90% of the

17:08 way there. I'm I'm back in I'm back

17:11 doing like full cycle sales at Apparo

17:13 now and I'm like, "Okay, dude. I've

17:14 spent so much time doing all of the

17:17 rigomearroll like data stuff and now I

17:19 actually have to like

17:20 >> talk to people again like very jarring."

17:24 Um

17:24 >> yeah.

17:25 >> Uh how do you see it like being in

17:28 service of that? because that's like

17:29 whenever I'm thinking about like GTM

17:30 tools I'm like okay

17:33 >> I 90% of the stuff I can do I probably

17:35 don't even need whatever like whiz pays

17:37 for like right and so maybe um if uh

17:41 >> if you're like a full cycle seller like

17:44 how would Deepline help you and and do

17:45 you have like full cycle sellers using

17:47 deepline like to a large degree or is

17:48 that still like a small percentage of

17:50 your like customers?

17:51 >> Yeah, I would say it's relatively small.

17:53 Um, and it kind of goes back down to

17:55 that like threshold of series B like

17:59 what what problem is the GTM team or GTM

18:02 engineering team solve uh for a company?

18:05 And after a certain threshold when you

18:06 have more sellers uh when every minute

18:10 that like 10 sellers is spending on a

18:12 problem uh now compounds faster than

18:15 than it does at an earlier stage that

18:17 like post series B 50 sellers 100

18:20 sellers uh that's where the systems get

18:22 way more important and I think that's

18:24 that's where the ROI of like the GTMS

18:26 code the more infrastructure based

18:28 approach of hey like a pre-all research

18:31 there's some nuance like by different

18:33 groups of the the

18:34 But you don't need to be the best seller

18:36 in the world to do a really good pre-all

18:38 research automation that saves your team

18:40 time. Uh and if they can save 20 minutes

18:42 per call and you're just frontloading

18:44 the information, uh 20 minutes times 100

18:46 sellers, uh that that is impactful. So

18:50 uh before that, I think the GTM muscle

18:52 and knowing how to build relationships

18:54 and actually being focused on all right,

18:57 what is the the use case pain point uh

19:00 what is the angle that we're going to

19:02 enter with? I think earlier stage that's

19:04 way more crit uh critical uh because you

19:07 don't like systems are to build

19:10 repeatable processes uh before the

19:13 process is fully repeatable. You need

19:14 more muscle around how do I find the

19:16 repeatable process and so it's it's kind

19:19 of like a before and after is how I view

19:21 it. Um before we do have folks who just

19:23 run the entire cycle on deepline because

19:25 we have all your CRM integrations I run

19:28 our entire sales motion on deepline. all

19:30 our follow-ups, everything is drafted,

19:33 kind of gives me the approval process.

19:35 Uh, I think what we've seen work well on

19:37 that end is human in the loop approvals

19:39 of basically just draft everything for

19:42 me and get me like 80 90% of the way

19:44 there and then I go in provide feedback

19:46 or or make the the final approval before

19:49 something gets done. So whether that's

19:50 updating the CRM or sending a message

19:53 with like notes from a call, uh, all of

19:56 those are are just do the work, let me

19:58 review. Um, but I would say that's

20:01 probably like 10 to 20% of our customers

20:03 uh versus like 80% are one central

20:06 person building for a larger uh set of

20:08 sellers.

20:09 >> Okay, this is something else I've been

20:11 thinking about like with like Muse and

20:13 Gro Muse is like having its like moment

20:15 instinct. Um, and I'm like very Grockbot

20:18 pill and all that kind of stuff. Um

20:22 what is the benefit uh or sorry not not

20:26 I wouldn't say benefit but like do do

20:28 you foresee this is actually like two

20:30 questions really like do you foresee

20:32 like a future where deepline is just

20:35 another like API that people are calling

20:37 from like a centralized orchestration

20:40 agent or do you see it as something more

20:43 where where you're going to utilize like

20:45 not the UI is like the wrong word

20:46 because I guess I guess you guys do kind

20:47 of have a UI even in the even in the

20:49 terminal or whatever.

20:50 Um,

20:51 >> but like how do you see Deepline a year

20:53 from now? Uh, assuming that um the AI

20:57 doomers aren't right and uh

20:58 >> Yeah.

20:59 >> Yeah. We survive. We survive

21:01 >> toasters. Yeah.

21:02 >> Yeah. Yeah. I think what I'm already

21:04 seeing and feeling I feel like there

21:06 there's a couple companies I look at. Uh

21:08 so owner.com is one where they're

21:10 generally like 6 to 12 months ahead of

21:12 everybody else. And then I also look at

21:13 like my personal usage behavior and like

21:15 where I'm I'm getting value. And what

21:17 we've seen pretty consistently is I use

21:20 chat, uh, codeex, cloud code, grockbot,

21:23 instinct, whatever. I'll try all of

21:25 them. Uh, and I never have to set up an

21:27 integration because I just add the

21:28 deepline CLI and then I'm done. And that

21:30 has every possible go to market

21:32 infrastructure tool uh, that I need. I

21:34 never have to log in. It's all just like

21:36 that one off. Uh, and it's a shared

21:38 backend. So let's say I I do have a

21:42 particular workflow like um enrich this

21:45 user and add to the CRM and I have a

21:48 very particular way of uh doing that.

21:50 What information I think is useful or

21:52 important? Are they hiring GTM

21:54 engineers? Uh how many RevOps people do

21:57 they have on the team? Did they just

21:58 fund raise? Those types of signals are

22:00 important for my enrichment but might

22:02 not be for anybody else. So I have a

22:03 play. I can say add this company to the

22:06 CRM in any of the interfaces. uh and it

22:09 will do the same thing every single time

22:10 because that play is is predefined. So

22:13 if we have a a customer where they want

22:16 the same kind of set of steps to be

22:18 taken for every rep uh or every

22:20 individual end user, it's a very

22:23 distributed set of agents on one central

22:25 set of deterministic and

22:27 non-deterministic workflows. Uh and all

22:29 the integration in O is handled

22:31 centrally. So

22:33 >> for like yeah for our less technical

22:35 listeners um

22:37 >> I'm wondering so I I

22:39 >> and maybe for me too because I I have an

22:41 idea about how this works but so you

22:42 have the deepline CLI which is like

22:44 installed on your computer which is just

22:46 a way to think about is just like a

22:47 little application that interfaces with

22:49 deep line on your on your PC or whatever

22:51 right

22:52 >> and then um for the other coding agents

22:55 like how would they know to go call the

22:58 deep line like as a tool or or whatever.

23:01 the deep line.

23:03 >> So yeah, so we we have an API, an MCP

23:06 CLI. We have like infinite number of

23:08 ways you can access uh the platform. Uh

23:11 how do they know you when you install

23:13 it? We kind of just say check deepline

23:15 first. Uh either in your like in your

23:18 config, agents MD, something like that.

23:20 Uh and then we have skills that teach

23:22 any of those agents how to use the the

23:24 tool correctly uh and how to package up

23:27 asks. Uh the way I would kind of uh view

23:30 it in it's just an MCP. It's the same

23:33 thing as like an MCP that you use for

23:34 something else, but instead of just

23:36 notion or just HubSpot, it's all 106

23:40 integrations that we have. Uh and it

23:42 will if you set up the integrations in

23:45 in our platform and say, "All right, I

23:47 want to add HubSpot or Adio." Uh as well

23:49 as our LinkedIn ads audiences and Google

23:52 Ads audiences, all of those are just

23:54 managed integrations. So it's basically

23:56 just one MCP that fans out to hundreds

23:59 uh that you no longer have to think

24:01 about did I connect this one uh who

24:03 should have access to which ones. All of

24:05 that is centrally managed.

24:07 >> Okay. But you have to tell your codecs

24:09 or whatever that hey when I run when

24:13 >> well this is what I've done. I'm just

24:14 wondering if this is the the best

24:15 practices

24:16 >> like uh

24:18 >> so like cloud code and cursor they have

24:20 like all of the

24:22 >> API keys and everything that I use and

24:24 then um

24:25 >> I have like a very large like context

24:27 file with explanations when to call

24:29 these tools and this is pre-loaded into

24:31 every session or whatever so that it

24:33 kind of has like some directions but um

24:35 but you're saying that you have to tell

24:37 Codeex for example hey when I'm doing

24:39 like GTM stuff like I want you to go

24:42 call deepline because it has all these

24:44 plays, right? Is that kind of like what

24:45 you're saying?

24:46 >> Uh yeah. Yeah. Effectively, if you if

24:49 you just say uh run this, I want to pull

24:52 data from people data labs, uh it will

24:55 most likely find deepline because we

24:57 have that context on on the CLI of uh

25:01 which tools. If you want to be 100% sure

25:03 that it's going to call deepline, you

25:04 can say use deepline or uh use one of

25:06 our skills and then it will uh find the

25:09 best plays. It'll find the right info

25:10 from there. Uh but basically the we have

25:13 a play and the goal is to get it to

25:15 trigger that play. It can be off a

25:16 keyword. It can be use deepline 2. Uh or

25:20 deepline GTM there. There's a couple

25:23 different ways you could trigger it. Uh

25:25 the goal is just to get it yeah get the

25:27 agent to try using deepline or check if

25:30 deepline solves this problem. Uh, and so

25:32 you I have run into issues where like oh

25:35 it'll try the HubSpot MCP first and I'm

25:37 like oh wait always use deepline for

25:39 HubSpot because it's already there and

25:41 it's a CLI it's much faster. There's no

25:43 like uh limits on how much data I can

25:46 upload or download from it. So I I have

25:48 to give it that context going forward.

25:50 Uh because some of the agents like

25:51 Rockbot specifically is one where it

25:54 will default to the MCP first uh a

25:56 native MCP. So, we need to tell it

25:58 instead of always defaulting to the

26:00 native MCP, go to deepline first.

26:02 >> Okay. Interesting. Um, are you good to

26:04 go for five more minutes?

26:05 >> Yeah. Yeah, I'm good.

26:06 >> Okay, cool. Perfect. Um,

26:09 okay. Next question. Um, if one thing

26:13 that I've struggled with like at

26:14 companies that I've worked at

26:16 >> is just trying to get people to like

26:18 adopt like new tools. like um

26:22 gosh, this is going to sound bad, but

26:23 like when I was at when I was at Aerops

26:26 um I don't want to sound salty about

26:27 AirOps like

26:29 >> but uh you know um

26:32 trying to figure out the best way to say

26:34 this, but there was there was there was

26:36 a an instance in which some new people

26:40 had been onboarded and they assumed that

26:42 I'm like GTM AI guy or whatever. So

26:45 they'd ask me questions which I was

26:46 always very humbled by that they would

26:47 come to me for like ask me questions or

26:49 something but you know I was like hey

26:51 you know you should like set up the

26:52 HubSpot uh uh API I can get you a

26:55 readonly key and you can install into

26:57 cloud code and they're like

26:59 >> how do you do that? And uh like Jesus

27:02 Christ like

27:03 >> what are we doing here guys? Um,

27:05 >> yeah.

27:05 >> Um,

27:07 >> do you have any like sort of advice for

27:09 maybe like a RevOps person or maybe that

27:11 one

27:12 >> AI pill sales guy in the org to like

27:15 help them like get these tools adopted

27:19 because I'm on the same page like the

27:22 leverage that you get from this kind of

27:23 stuff whether it's deepline or any of

27:24 this other stuff like um

27:26 >> is so outsized versus the one day of

27:29 banging your head against the wall that

27:30 it really takes to to learn this stuff.

27:33 And I I do think that like using AI

27:35 tools requires sort of like a different

27:37 mental model in middle managerial

27:40 skills. Uh interestingly enough, um

27:42 there's some study which anybody can go

27:44 look up probably that the people who are

27:46 like effective with AI are the ones who

27:48 have like management skills. Um so uh

27:52 that's an interesting stat. But like how

27:55 do you think about maybe somebody who

27:56 wants to sell I'm just going to give you

27:58 a really soft here like if you want to

28:00 sell deepline internally but uh you

28:02 don't know you're not sure like if you

28:04 can get these adoption from your team

28:06 like maybe what's like some way to to

28:09 encourage that for for people. Yeah I

28:12 don't know open-ended question.

28:13 >> Yeah I I think there's a bit of

28:15 psychology here of like you and I are

28:18 we're I'd probably describe us as like

28:20 tinkerers. like we are interested in

28:21 trying things and just seeing how it

28:23 works and and testing new things out.

28:25 80% of the uh employees at a company are

28:27 not they do not care about the new tool.

28:30 They would rather not onboard a new tool

28:31 if they can avoid it. What I've seen

28:33 pretty consistently work is two two

28:35 kinds of approaches. One is uh just

28:40 focus on doing things where they work

28:43 today. Like just avoid the adoption

28:44 problem entirely and say uh the central

28:47 team is going to populate the

28:49 information you need in Slack or

28:51 Salesforce or uh what like text you

28:54 something like whatever is the least

28:58 different from what they're doing today.

28:59 Just take that on to the the central

29:01 team and that is going to have the

29:03 highest adoption that's going to have

29:04 the highest ROI over time. it is hard to

29:06 figure out exactly what the best option

29:08 is all the time. Uh I am usually the

29:10 joke I make is I'm usually wrong on like

29:12 what I think the best solution is going

29:14 to be. I'm like going into a company

29:16 this it's going to be Slack at this

29:17 company. Slack is going to be the main

29:19 way we can communicate with reps and

29:21 then it's definitely Salesforce or

29:23 Salesoft or something else. Uh so it's

29:25 often hard just to like pick the best

29:26 option, but centralizing it and getting

29:28 it to a rep without having to actually

29:31 have them do anything is is always going

29:32 to win here. Um, and you can't like

29:35 force someone to be curious about these

29:37 types of things. And so it's it's like

29:39 what just gets the job done and and

29:42 centralizing is going to win that most

29:43 of the the time. Uh, you what you'll

29:46 find is like 20 to 30% of uh reps are

29:50 actually tinkerers and they are actually

29:51 interested in doing this stuff. So 70%

29:54 will get the value, 30% are like, oh, I

29:56 want to add this or change this and then

29:58 just give them the tools to adopt. like

30:01 you you can't really uh force curiosity.

30:04 It's it's some people are going to have

30:05 it, some people aren't. Uh but that

30:06 doesn't mean they're not great sales

30:08 reps. Uh the other pattern, so that

30:11 that's kind of one pattern I've seen

30:12 work. The other one that I've seen

30:15 worked well and this worked really well

30:17 at companies that I thought would take

30:19 really long to adopt AI and it was just

30:23 start showing them examples uh but make

30:26 them all optin examples of hey like uh

30:30 here's a automated email that we're

30:32 sending to like two reps and here's what

30:34 you have to do to set it up but uh we're

30:36 not going to do that for you. You just

30:37 tell us if you want it and we'll help

30:39 you set it up. And uh there was this one

30:41 like uh more legacy company that sells

30:45 into an extremely legacy industry like

30:47 as legacy as you can think uh like World

30:49 War II legacy uh

30:51 >> some buggy manufacturing.

30:52 >> Yeah. Yeah. Exactly. So let's go with

30:54 manufacturing as the the target. Um and

30:56 the sales reps are like older. They

31:00 don't want to change anything. Uh but

31:02 when we started showing them like

31:04 pre-drafted

31:05 research emails with custom landing

31:07 pages, they're like, "Oh my god, this is

31:09 this is fantastic." They went and all

31:11 signed up for their own cloud accounts

31:13 and like started doing things and you

31:16 just kind of have to like peique the

31:17 interest and examples are consistently

31:20 the the best way I've seen to do that.

31:22 And that's why we love our like

31:23 lightning talk stuff which you were one

31:24 of the first you were the first speaker

31:25 at the first one for the record of

31:28 people really like the imagination is

31:30 really what we're trying to to get

31:33 across of broaden your horizons on

31:35 what's possible show all the things that

31:37 are possible and then people will tend

31:38 to say like oh I could use X for my case

31:41 or this might actually solve my problem

31:43 over here and then there's a little bit

31:45 more interest uh generated. So like if

31:48 you want adoption, if that's your number

31:50 one priority, uh remove humans from the

31:53 equation and just do it for them. But if

31:55 you want to actually peique curiosity

31:57 and interest, uh demo days, examples, uh

32:00 just kind of show like an AI internal

32:03 demo day every week at a company is

32:05 going to have more ROI there than uh

32:08 giving everybody access to all the tools

32:10 uh because they they just don't know

32:11 where to start. And I I think we got

32:13 this from our original platform, Arrow,

32:15 where we had a hosted managed agent,

32:17 kind of like cloud code in the cloud,

32:19 that sat on top of your data warehouse.

32:21 And the number one piece of feedback I

32:22 always got is uh we'd had a couple

32:25 recipes that you could press buttons

32:26 for, but every person when they saw an

32:28 open chat box would be like, "Okay, now

32:30 what? What do I do?" And I'm like,

32:32 "Yeah, just type in your problem." And

32:33 they're like, "Well, I have a hundred

32:34 problems. I don't know which one to type

32:35 in." And people just go blank when they

32:37 see a blank box. And I think there's

32:39 some psychology around this of there's

32:41 just too many options and then you just

32:43 panic and crumble and don't know what to

32:45 do. So if you just give someone claude

32:47 and say go do anything, they're not

32:49 going to know what to do. But if you

32:50 give them five examples that they saw in

32:52 a demo last week, they're like oh yeah,

32:54 I need this one. This one would solve my

32:56 problem. Uh and it just gives them

32:58 something more tactical to grab on to.

32:59 >> So that's super interesting. Um we're

33:01 we're almost out of time so I'll

33:03 >> have this one last uh discussion point

33:06 but so I had uh started about think I

33:09 had thought about doing my own SAS um

33:12 which is a very maybe cliche I don't

33:14 know um at some point and um it was

33:16 going to be like a Salesforce uh tool

33:19 for um just all kinds of stuff. Uh it

33:22 was the concept and you're anybody who's

33:24 listening to this is welcome to steal

33:25 this. Um, I'm not going to start the

33:26 SAS. Um, I decided it would be eaten by

33:30 the frontier models eventually, so I

33:31 didn't think it was a good idea. But,

33:33 um, the the concept was essentially like

33:35 a context repository for like your CRM

33:38 because like everybody has all these

33:40 like crazy, you know, variables and

33:42 custom fields and custom workflows and

33:43 it was going to go ingest all that

33:45 >> and then it was going to interview you

33:47 though like in the beginning. Um like

33:48 there would obviously it would work out

33:50 of the terminal and stuff but the first

33:51 thing it would do is be like hey you

33:53 have like this Salesforce flow like

33:54 what's I don't understand it like what's

33:56 the purpose and I would want it tagging

33:59 metadata into fields so that when APIs

34:02 or or whatever uh were calling these

34:04 fields like it would have some kind of

34:06 context on it. Um I'm just curious like

34:09 I'm just throwing this out there like uh

34:11 >> do you think that um a guided like

34:14 interview process would almost be like a

34:15 better onboarding experience because

34:17 that the open chat box like

34:19 >> like canvas uh thing it's it's a problem

34:22 that like every sort of PLG tool has and

34:24 I guess I would categorize like deepline

34:27 >> as like uh PLG and so so anyway my

34:30 question is like

34:32 >> what maybe for other like PLG founders

34:34 out there since I'm now categorizing you

34:36 as PLG Maybe erroneously, I don't know.

34:39 >> Yeah. No, no, that's fair.

34:40 >> What do you What sort of like advice do

34:42 you have for them to I don't know for a

34:45 better product experience and like the

34:46 age of AI and the blank slate era?

34:49 >> Yeah, I mean we are testing that uh

34:52 pattern right now. Like that is that is

34:54 something that we're looking into

34:55 because the the way I've thought about

34:58 it is fastest time to value like what is

35:01 that aha moment? And so that that

35:04 creates this kind of bifurcated

35:06 experience of okay well we want power

35:08 users to be able to get to something

35:09 useful as quickly as possible and we

35:11 don't want to like get in their way but

35:12 if you don't know what's going on we

35:14 also still need to get those folks

35:15 across the finish line and we can

35:17 prioritize experiences and say like all

35:19 right well we're going to make it really

35:20 easy to move fast but also like give you

35:22 this exit path. Um

35:25 there there's kind of this concept of

35:27 the golden path of like the e what are

35:29 you expected to do at every single step

35:31 and make that very clear and visually

35:32 clear and when you have an open box it's

35:34 not clear. So we've tried recipes, we've

35:37 tried tried these prompts and doesn't

35:39 really land. Uh and then the other piece

35:41 of feedback we get a lot is can you help

35:43 me set up my context folder? Uh and the

35:46 thing about deepline is we don't

35:47 actually need a context folder. like we

35:49 can get most of what we need in runtime

35:51 from uh your website or maybe your local

35:54 files. If you've already run prompts in

35:56 that folder, we're probably going to get

35:57 all the information we need. Uh but

36:00 that's some everybody like wants that

36:02 step of make sure I'm doing this

36:04 correctly. And so there's there's this

36:05 like confident and trust problem that

36:07 we're we're also seeing and the guided

36:10 experience solves kind of both of those

36:12 of where do I start and is my entire

36:15 like workspace correct for this? Do I

36:17 have all the pieces I need to be

36:18 successful here? And the short answer is

36:20 you probably are without realizing it.

36:22 Uh but people want that like second

36:24 level of of validation that yes, we have

36:26 confirmed that you are good. Um and then

36:28 the first one uh we'll we'll see how it

36:31 pans out. We're we're launching it soon.

36:33 So I think I'm I'm bullish on it for a

36:36 particular type of user and bearish on

36:38 it for uh like you I think would be

36:41 annoyed in the process. I think

36:44 >> you would be annoyed. So, how do we like

36:46 let you exit it or skip it? So, it

36:48 becomes like yes, it's going to add

36:49 value, but like how do we make sure it's

36:51 the right person and the right value uh

36:54 that that we're delivering? And I I do

36:56 think just PLG has changed remarkably.

36:58 And we were I'm I'm going to go out on a

37:01 limb and say we're probably the first

37:02 cloud code native company where you

37:04 could not use us if you were not using

37:06 cloud code uh in like March or whatever.

37:08 Um and the the benefits there are

37:11 onboarding is really fast. Time to value

37:13 is incredible. You don't have to

37:14 actually understand anything about the

37:16 platform to get value. Uh the hard part

37:18 is we have very little UX surface area

37:20 to teach someone about what we do. Uh

37:22 you have what you give back to cloud

37:24 code and or codeex or whatever and that

37:28 is limited uh and it's noisy. So it's

37:30 going to show a bunch of code. You might

37:32 give some feedback in there, but it's

37:34 just like you don't have the surface

37:36 area that you had 10 years ago in

37:37 productled growth. Uh you don't have the

37:39 UI that you can guide people to. We have

37:42 power users of our platform who

37:44 literally never log in to the UI. They

37:46 just do everything through the terminal.

37:47 And I was like, "Oh yeah, we launched

37:49 this product." And he's like, "Yeah, I

37:51 don't look at the UI. You have to tell

37:52 me in the terminal or I'm never going to

37:53 see it." So it's education is now way

37:56 harder. Uh you don't own as much surface

37:58 area. And both of those are constraints

37:59 that we have to optimize around. Uh, but

38:01 you now have this magical guided like

38:04 Clippy that will help you get started on

38:06 any problem. So, what's the right

38:09 experience there? I don't think I've

38:10 seen anyone nail it. Uh, I I personally

38:13 don't like the really guided onboardings

38:14 also. So, I think we're we're biased in

38:16 some ways, but I do think it's something

38:18 along those lines, but maybe a lighter

38:20 weight version of that.

38:22 >> Super interesting. We could we could

38:24 chat about yeah in the age of AI for

38:26 forever because it it really has like

38:28 it's broken down all of the

38:30 >> traditional sort of uh I don't know

38:33 pathways for adoption that product

38:37 designers would would rely on to like

38:39 get a conversion or or you know upsell

38:41 people later on. And um yeah it's super

38:46 super annoying like because it's super

38:47 annoying for me when I get the product

38:49 got on boarding experience or whatever.

38:51 Yeah. um for for obvious reasons, but

38:54 it's like but if you're a product

38:56 designer or a CMO who wants somebody to

38:59 like get this stuff into your brain like

39:01 what do you just like oh no they don't

39:03 have to do the onboarding experience

39:05 like you know it's a really difficult

39:07 decision like I think people are having

39:08 to like wrestle with from like a product

39:09 design standpoint and um

39:12 >> yeah we'll we'll see what happens uh but

39:14 yeah I'll let you know how it goes.

39:16 Okay, cool, man. Well, thank you so much

39:18 for you're you're my first guest on what

39:21 are the chances.

39:22 >> Yeah, I don't know. This wasn't by

39:24 design. You just in um did this

39:27 I I was like, dude, I should have

39:28 blocked off my calendar for the podcast.

39:30 No podcast before 11 a.m., but

39:34 >> sufficiently awake for this one. So,

39:36 >> perfect. Appreciate it.

39:38 >> So,

39:38 >> I'll do later one next time. Yeah,

39:40 after.

39:40 >> No, we'll definitely have to do a part

39:42 two. Um got a season two planned. So, I

39:45 booked all of season one. Now, I'm

39:47 booking season two, which is gonna

39:48 happen later on in the year.

39:50 >> Yeah, that looks like a week. Very fast.

39:53 >> Also, I'll probably have a sponsorship,

39:57 but if you want to sponsor season two,

39:59 let me know.

40:00 >> Yeah, we'll talk.

40:01 >> Um, anything you want to leave the

40:03 audience with? Uh, any lightning talks,

40:05 anything coming up? Obviously, your your

40:07 stuff's going to be in the show notes,

40:08 but anything you want to pitch before

40:10 you go?

40:11 >> Yeah. Um, nothing top of mind. We just

40:14 had our lightning talks last week. Uh

40:15 we'll have another one in November. Uh

40:18 or late late October, early November. Uh

40:20 most likely New York one more time and

40:23 then SF for the next one. We try to do

40:25 these pretty often. If you have talks

40:26 that you think are interesting or things

40:28 you want to show off, uh this is these

40:30 are community events. So, uh always love

40:32 submissions. Uh, and if you want to try

40:35 out Deepline, uh, we have, uh, a promo

40:38 code I can share for this podcast

40:40 specifically, uh, which will probably be

40:43 Apero, uh, or Ben at Fine, whatever you

40:47 want to call it. We'll edit it later.

40:49 Put that in in post. Okay.

40:50 >> Uh, yeah, we can we can do some credits

40:53 for uh, your your listeners here. And,

40:56 uh, yeah, if you guys have any feedback

40:57 on the product, always excited to hear

40:59 from uh, folks.

41:01 >> Awesome, man. Well, uh, thank you so

41:03 much everybody. Go try out Deepline and,

41:06 uh, talk soon.

41:07 >> Appreciate it.

41:08 >> Yep. Bye.