you2idea@video:~$ watch 1aPk2VicN3M [51:11]
// transcript — 676 segments
0:02 launching something is like anyone can do it today which is [music]
0:05 just crazy to think because I used to think I would need to go and study an
0:08 engineering degree in order to do that [music] any go to market engineer anyone in go
0:13 to market can now produce an agent to replicate [music] anything in the go to
0:18 market flow but it's what's the quality how quickly can you iterate and learn
0:21 that as well before I go to sleep I always like or open a new [music] you
0:25 know cloud code session to say do this so next day when I wake up I can review
0:34 >> [music] >> So I guess one of the most interesting
0:37 things that's happened in the last few years is we often talk about code and
0:42 capital as two of the highest form of leverage that you can have as an
0:46 individual. And now just in the last few years people who weren't proficient
0:51 software developers now have access to code as a form of leverage. And so I
0:56 wanted to talk to you three as examples of people who have really taken
1:00 advantage of that and and shaped some pretty awesome things within your
1:05 companies and yourselves um that really demonstrate how how the world has
1:09 changed and what's possible. So maybe I'll start with you Jason just yeah
1:12 you've been a product designer for many years um self-taught AI engineer and now
1:18 you have taken it to the next level by building the tooling to help other
1:22 people whether designers or other nontechnical people in general build I
1:27 mean have access to coders leverage themselves I'd love to kind of start
1:33 with you know what was your work day like as a product designer five years
1:37 ago and and what is it like today? Yeah, it's uh definitely changed a lot. Uh so
1:43 when I started my career, I think Photoshop was the was the thing you go
1:47 and then you have like sketch, figma, you basically draw all the boxes from
1:52 from from scratch. Um so my last company randomness AI when I was there uh I
1:58 think one thing we trended is that uh we tried to get designers to so I was I was
2:02 leading product team there and uh one thing I was frustrated is because
2:07 outside work I vibe code a lot and uh I see all my engineer teams uh workflow
2:12 start fundamental change but designer PM is still doing what they do before then
2:17 why I ran experiment I tried to get my team to start using those tools uh and
2:21 initially We tried to use I think Lobo and and Vzero to just build prototypes.
2:27 So they kind of skip the PRD process. Um so that part like shot a lot. Um but
2:32 later I think things became really interesting when we I can't get my
2:36 design team into the actual real codebase. So uh I actually like kind of
2:42 lock my team in a day for room in a room uh and force them become front end
2:46 engineer almost uh getting them you know create a GitHub account uh cursor uh
2:52 teach them how to get a local server up running. It was a pain in the ass. But
2:57 after they they learned that I start saying okay the first PR ship from one
3:02 designer because uh they you know the the the old development process was like
3:07 you have marketing team give PM feedback and then PM get designer design
3:11 something and then engineer ships the actual design but engineer never ship
3:15 100% what you want they always last like 10% quality gaps. So I start seeing my
3:20 designers to just jump in the codebase um fixing all those things spend hours
3:24 where front engineer never want to do uh so that's kind of new I was seeing new
3:28 workflow that I didn't have five years ago uh because it's not possible um but
3:35 only uh and and I remember the one designer I think it was Wednesday and he
3:39 he was saying oh I realize I haven't opened Figma for past three days this
3:44 week yet and because he just uh you know for the initial ideation he would just
3:48 draw on on paper. Uh and then once you get it done, you go to the if the simple
3:53 things you can just go to cursor and then give a uh uh give kind of
3:58 screenshot to the to the cursor say hey this is kind of wireframe and help me
4:03 build the kind of feature and experience prototype it in real codebase with our
4:08 existing design system. So here we do something like that as well. Uh so I I
4:13 definitely think the the consumption or time we spend in in kind of traditional
4:17 design tools uh has has decreased a bit um compared with before um and uh that's
4:23 probably one big change into my designer workflow um and uh uh but for myself I
4:29 was uh also doing a lot of engineer work recently as well and uh uh I found like
4:36 uh when I talked to my best engineer team uh the way they work quite
4:40 interesting they were open, you know, uh, five or eight different kind of
4:45 terminal sessions at the same time. Uh, and and they were it really kind of
4:49 extract all the capitalism out of you when you have an employee that you just
4:53 feel like always working and I observe I myself doing that as well. I would just
4:57 open one session ask agent to hey build me this feature uh come up with a plan
5:03 and then I would say go execute while it is executing for the next five to 10
5:07 minutes. I like okay just just do more work let me open a new session let's do
5:11 the next feature so it's very addictive as well um and before I go to sleep I
5:16 always like or open a new you know cloud code session to say do this so next day
5:21 when I wake up I can review the work um so I definitely think like the software
5:25 engineer uh process also fundamentally changed and you don't really look at
5:30 like low-level implementation too much >> maybe like over to you s in terms of I
5:33 mean you're coming from a different perspective which is go to market But I
5:37 feel like you've had kind of a similar set of experiences. So I'd love to hear
5:41 kind of your version of what your workflow used to be and then where it is
5:44 now. >> So I think even it's like you look at it a year ago, but
5:49 you even look at it a few months ago to where like stuff is now with how fast AI
5:56 rapidly changes. I think 6 months ago um NN was like all the hype, right? And
6:05 um to execute all these tasks. But we look at the way MCPs move um Claude code
6:09 skills um and how much you're actually able to bring in to someone that's
6:13 non-technical and able to actually go and execute on really specific tasks.
6:18 So, you know, um I was speaking to our head of sales and he was telling us that
6:22 about a year and a half ago, they pre-grafted a pre-call research agent
6:25 that would be able to go and do really detailed research for an account
6:29 executive both on our internal tools and externally and then produce a really
6:33 like detailed brief of what we want. And that was going to be something like 160
6:37 different nodes in NN to get the quality that we wanted and everything else. And
6:42 then today we rebuilt that entire process and it's three nodes chucked in
6:47 with a number of different MCP servers, really detailed prompting. And it's not
6:51 only our ability to like deliver it really quickly, but it's the ability to
6:55 then iterate because um I don't know about you guys in the age where like you
6:59 can produce just about anything. It's actually the quality of what you produce
7:03 that actually becomes the most important because you can dump as much information
7:07 at sales reps and then they're going to use none of it. Um and so the ability to
7:12 be like we have something but we can really quickly change iterate and make
7:17 it useful to you um as a sales rep I think is changing so much. So with go to
7:22 market it's any go to market engineer anyone in go to market can now produce
7:27 an agent to replicate anything in the go to market flow but it's what's the
7:31 quality how quickly can you iterate and learn that as well >> how is that set up internally is there
7:36 kind of one person who is building these and getting the feedback and changing
7:40 them or do you actually have like your your sales people trained to the point
7:43 where they're able to kind of build an n themselves or I mean where are you up
7:48 to? Yeah, I think for us we're actually just moving over that phase right now.
7:52 So probably probably for the last 6 months it was our go to market
7:55 operations team that was really in the tools building in like workflows and all
7:59 of these different things. But I think increasingly we're moving towards a
8:03 world where we want to like democratize the use of AI and especially agent
8:07 building to our sales teams themselves to actually meet them where they're
8:11 working. And I think it's actually really easy to do today. So, we're
8:15 moving more and more to a world where we build frameworks and then each person
8:18 can actually personalize the agents to themselves. So, I'll be like, "Hey guys,
8:22 here's a pre-core research agent that our ops team has built, but you're
8:26 actually you might have questions that every sales rep has a unique opener or
8:30 every sales rep has a unique like point of view on what they want to do. So,
8:33 it's how do you actually then customize that to your own workflow?" Once the
8:37 structure is built out, it's super easy and as long as you enable your team to
8:41 do it, um you democratize that use back. And I think yeah MCP with the addition
8:46 of kind of all the tools being built now has been such a game changer in that.
8:51 >> Are you using N to view those things by themselves? We're using N8N to build um
8:57 more complex workflows but also natively now in Claude projects. Um you can
9:01 actually get a lot of the utility previously that you had to get out of N8
9:07 or something else in product where N8 is useful is and where I found mostly these
9:10 things to be useful is kind of there's two forks in the road. One where you
9:15 want a really desired specific output to plug into somewhere and you need it to
9:19 run at a specific schedule, right? So let's just say I need it to run at a
9:23 certain time of the day or I need to run it when a certain event happens. So
9:25 there's some sort of trigger and you just need that triggering mechanism that
9:29 any workflow builder is really good to deliver. But where you want for example
9:35 the ability to like live run an agent. So I'll give an example. Um it's better
9:39 to actually leave it in clawed as an environment and and users can actually
9:44 get that. So for example, one of the most like I was a sales rep back in the
9:47 day and the thing I hated the most was end of month admin, right? you get to
9:50 the end of the month and you got to clean your pipeline. Um, close states,
9:54 it's got to be updated, there's a bunch of deals that haven't happened before,
9:57 blah blah blah blah blah, right? And so now instead of doing that, um, with a
10:01 really good HubSpot MCP setup that has the right permissions, each user can
10:05 literally go in and say, "Hey, show me my pipeline, show me my active pipeline
10:10 where close dates are overdue." then also look through my emails and figure
10:15 out where I need to be moving the what are the close dates, what were the next
10:18 steps that were stipulated via my emails. So now what was previously for
10:22 me the most dreaded hour of my Friday at the end of the month to like update
10:26 close dates is like two messages to claude in a predefined project and then
10:29 it moves those close dates out for me as well. There's no manual work on my end
10:35 and so my interface with a CRM with any sort of object completely changes.
10:40 You mentioned quality before >> and when you said that what I
10:43 immediately thought was like how accurate is that and I guess how much is
10:47 it improved and how much do you expect it improve to improve in the next kind
10:50 of 3 six months because the worst version of that I guess is it gets 70%
10:54 of them but it doesn't get the other 30% and then you end up just with inaccurate
10:58 data >> 100%. That was so that's been my experience with it as well and I think
11:04 it's a big learning curve that any go to market engineer anyone working in this
11:08 space goes through which is so about 4 months ago we rolled out what I would
11:12 call is like the one of our like precore brand research agent to our sales team
11:16 and self admittedly now it was not a success by any means whatsoever and like
11:22 why because it again got 70% of the way there right but when the reps started
11:26 poking holes in the the like remaining 30 it either pulled data that was too
11:30 old, not within the confined restraints that we put on the agent, other bugs
11:34 that would creep in, not finding the right stakeholders, little things like
11:39 that, their trust in that like agent completely diminished as a sales rep,
11:43 right? Like people might endemically already be a bit more jarred by AI. And
11:47 so that was like definitely a learning curve for us of being like just because
11:51 you can spit out an agent quickly doesn't mean shipping it to the entire
11:55 team is like what you should do straight away. and it's about how do you actually
11:59 get it to a place where it can execute that task to the highest degree um not
12:03 just like ship quicker. So that's why we went and it took us like 3 minutes three
12:08 months to rebuild that process. Um and now I would say again in that 3 months a
12:14 the ability of the LLM has improved greatly with every new model release. LM
12:19 are much more accurate with every new feature upgrade. the way you can point
12:23 LM two very specific directions has improved but also our ability to like
12:28 learn from our teams what is important and continue to refine prompting down
12:34 process down um has had like a massive impact on the end output
12:38 >> is there a kind of specific team structure that you I mean I guess you've
12:42 probably iterated a bit around like it's sitting within ops and it's sitting
12:44 within teams specifically is it centralized is it decentralized like
12:49 what's been working best for you >> it's a great US. I'm I'll give my point
12:52 of view, but I'll also say I think it's an open discussion that's going to
12:56 change a lot over the next like few years. Um, and again, like I always say
13:03 like we built an agent 3 months ago that does like um case study creation for us
13:08 internally now. And if I was to rebuild it today, it would be completely
13:11 different just because of how fast that entire process has moved. Um, now going
13:17 back to your point, I think having a centralized go to market engineering or
13:22 ops function is really beneficial from the point of view of looking ahead of
13:25 like what's coming and being able to implement and trial and iterate because
13:29 what people strug and like this is exactly it. It's like what we've said
13:33 internally is if you want a sales team to go and build it themselves or CS
13:37 teams to go and be building agents. What what today you're asking people to do is
13:41 do their full-time job and learn a new language at the same time but at work,
13:44 right? And if everyone's stretched already and you're telling everyone take
13:48 multiple hours of your day and go figure it out, it's never going to work. And
13:53 the reason also is it's like people can go and try something out. It's like oo I
13:56 just Gemini launch, right? Like let me go try anti-gravity or like let me go
13:59 look at Nana Banana Pro. It's like I created a few images, great, hysterical,
14:04 but now how does this actually fit into my workflow like where do I actually
14:08 take this was great I experimented and that's where you need a centralized
14:13 function whose role it is to figure out this is how you integrate the new
14:16 technology that's just been released into customer CS um sales marketing
14:23 across the board um and helps drive that adoption across the board.
14:28 >> Nice. Renee, um, you obviously took, I guess it was a year ago today, I think
14:32 even, that you, um, you quit your job, went on like to become a solo builder
14:37 and to try out all these new tools and see what leverage you could get. Would
14:40 love to hear the story of your experience and then kind of how you're
14:43 using the tools today at Chronicle as well. Yeah. And I think my my story like
14:49 I've been non-technical um like in every job that I've had um like I was a
14:53 product manager and then I did strategy and ops and I was a chief of staff for a
14:59 little bit and I took it like six months off purely because a the AI tools were
15:04 exploding and I just thought I'm never going to be able to do this in a job. So
15:07 I'm going to take some time off and actually just experiment with with all
15:12 of this and build. Um, and I think part of it comes from always being
15:15 nontechnical and not being able to actually conceptualize something and
15:19 actually bring it to market. So I think it was just like a ripe opportunity to
15:25 do so. Um, it's also how I ended up deciding to move into marketing. Um,
15:29 because when I started playing with chatbt and my first workflow to build a
15:34 product was very rudimentary. I would copy and paste chat code into my cursor
15:41 workspace and um then kind of push it to Apple and and that's how I got my first
15:46 iOS app out there. Um but through that process um like launching a product and
15:51 I still consider myself a vibe coder. I can't launch production grade product
15:55 but launching something is like anyone can do it today which is just crazy to
16:00 think because I used to think I would need to go and study an engineering
16:03 degree in order to do that. So I think that's such a huge unlock for people who
16:07 just have random ideas where they should definitely go ahead and try and ship
16:12 something. But I think through that I actually realized okay people can build
16:16 and people can launch things now it's actually the distribution side which is
16:21 very hard to crack and I don't think AI can really solve that for you. Um, and
16:27 so in kind of making a career pivot, that was where I decided to spend my
16:31 next role in was more on the go to market side and learning growth,
16:34 learning distribution. How do you grab people's attention? How do you build
16:39 content and and try to scale that and make that go viral in order to then
16:43 promote the products that you can now build with AI? Um, and so yeah, that
16:48 that's kind of how I've landed in my current role at Chronicle doing growth.
16:54 Um, and these days how I use AI is very much content production and like scaling
16:58 that workflow and trying all these different channels to promote our
17:03 product. Um, and then in my spare time I'm able to do like product all the way
17:08 to distribution for my own little apps that I just have fun with playing around
17:12 with. So yeah, it's really like the past year has just train changed so
17:15 drastically when it comes to tools available and what people can now do
17:21 themselves without having to like you know be in a company to do or you know
17:25 have a team to do. You can very much be a solo builder now. I feel like
17:29 marketing has changed so much in the last couple of years and
17:37 whether it's launch videos or like geo or just how you create content like I
17:41 would love to hear if there are any specific workflows that you think um are
17:47 really effective for you today that wouldn't have been possible two years
17:49 ago. >> Yeah, I think it's it's one of two things. It's content production and I'm
17:56 a oneperson marketing team at Chronicle and so I have to wear many hats. I have
18:01 to do SEO. I need to do our socials. Um I like own our brand and and what gets
18:06 put out there. So many different workflows which are possible now with
18:11 just the one person and and we don't plan to um grow the team anytime soon
18:16 either and I'm very much able to do what we need to do still. Um so yeah on the
18:21 content production you know for SEO we push out a lot of long form content on
18:26 the website. Um I am able to spin up landing pages um if we want like an
18:31 enterprise page or a partnerships page like I'm able to do that um immediately
18:37 like within the hour just using some AI tools both for the writing part and
18:41 copyrightiting but also for the um actual website and the design. Um, and
18:46 then when it comes to brand assets, I use like NO Banana Pro and Image Gen and
18:53 and a lot of design tools to create the assets that are on brand. And I think
18:57 before you would have to rely on a brand designer and a whole design team to, you
19:01 know, give them the brief, wait for it, iterate on it, then be able to launch
19:05 and that might be a process of minimum a few days. Whereas now I can do that
19:10 again within the hour. So it's kind of like reigning in the control but also
19:14 being able to move at a much quicker speed and execution wise where you can
19:18 afford to just have a oneperson team these days. >> How does another banana fit in the
19:24 workflow? Do you guys do like Instagram post or uh what's the out?
19:27 >> Yeah, it it's like social posts announcing new features where you know
19:31 you would have the product mockup. You might have um some motion behind it. Um,
19:37 it's it's really just like brand imagery and and we have a very specific kind of
19:41 brand pallet where it needs to be on brand and so we can I can just generate
19:46 that today rather than going into Figma and like having a designer design it.
19:48 So, >> Gotcha. How do you get like give you on brand? Do you just simply chunk this
19:54 part? >> Yeah, reference images are always the best. ref you give it a reference you
20:00 just say create it but change around these little things and I think the the
20:04 prompt engineering I don't have necessarily the language to do it but
20:08 you if you know what you're going for then you can kind of prompt your way to
20:11 get that output >> interesting for sure yeah I feel like
20:15 nana or reason model really hit the production gray bar recently
20:19 >> um yeah interesting >> you were talking a little bit about
20:22 quality earlier and how kind of designers can now essentially become
20:27 front-end engineers and Um, that's a great thing because they love pixel
20:32 perfect and like now they can actually ensure that things are pixel perfect.
20:36 Um, I'm keen to get your take on like how design evolves with the quality bar
20:41 and um, maybe kind of taking that next step which is like it used to be
20:45 engineers that ship the code then it was like designers who make it pixel perfect
20:49 and then like what is the next version of like other people who could
20:52 potentially ship pixel perfect design? >> Yeah, definitely. Uh, that's a good
20:56 question. So uh I I think I have one one observation I saw is that previously you just have
21:02 this pipeline for delivering any software you have this uh like spec uh
21:09 someone race spec who is nontechnical u PM uh founder and engineer delivering
21:15 that um but the problem is that uh often engineers even though they use AI tool
21:20 they often care more about like is this how this system is architected but those
21:24 people who are nontechnical and are generally care more about final
21:29 customer output. So generally I actually I found those people are more empowered
21:33 and can properly produce better things than engineer doing that. Um and for
21:39 like design specific uh is that so we've been building like a design agent for I
21:43 would say past two months and uh initially we were try to like uh uh
21:48 getting the agent to uh to deliver better output by giving like some more
21:53 you know design practice uh as well as like the the the the company's own
21:57 product context brand all those things we allow people to input in uh I would
22:03 say one month ago the product quality bar it generated probably do not as a as
22:08 a quality bar that a really good designer were one uh because it just
22:13 generally not that great as you know front end implementation but uh like
22:17 literally just two weeks ago when nano banana sorry Gemini 3 pro show up that
22:23 that whole uh output quality just trend dramatically we start we found you just
22:27 need to give a enough context it can actually hit the real production
22:31 designer quality and that that's like the part I feel really exciting as the
22:36 name like every few weeks you just new model that change your paradigm about
22:40 about about the uh and the key thing about the uh hitting the quality bar for
22:45 that certain scenario output. Um so I definitely think uh yeah so with gymnast
22:50 it really start hitting that bar. I see. Uh which means that um with with enough
22:55 context engineering plus the models uh we can actually produce you know uh a a
23:01 a what a good designer can produce uh traditionally uh and probably that
23:05 quality is already better than you know 80% of the design freelancer on Upwork
23:10 that they were produc uh which is interesting how that changed the the
23:15 dynamic which means if I'm am if I am a a founder uh who uh who don't have
23:21 designer um uh sorry if I'm founder I'm building product one option is that I
23:26 don't use a designer um and uh or I go to a design agency to to to generate
23:32 something uh which normally uh you know if you hire designer you will know that
23:35 they probably also don't have enough context about your product uh and and it
23:39 has the sometimes they have a lot of opinions and will push back your request
23:43 but if you have AI you just like tell it whatever ridiculous requests you have
23:49 but you know from the customer you talk to. It's probably true. It'll produce
23:52 something. It will try its best to produce something and then you just like
23:55 within a minute, right? And then you give feedback and it produce more. So
24:00 that iteration and feedback loop just much better than you can get from a
24:04 traditional service in the actual uh company setup. What I see is that the
24:10 role start really blurring um because each individual can do so much more than
24:14 before. If you have growth team, why why do you why do you still have the traditional
24:19 setup of one PM, one designer, two engineer? Uh well, you can have a like,
24:24 you know, growth people like like Rene just have idea ship push to production
24:31 code. uh and uh and and if you if you go through those kind of traditional you
24:34 know stop to scale up journey you'll notice that at early stage the the
24:38 decision making quality and speed is almost always better because one person
24:42 just hold all the contacts and they can just make it the right decision. uh but
24:46 as you as you grow in the traditional way you just adding you know break one
24:50 row into multiple and then you start creating silos of communication and it
24:54 just generally leads to you know slower decision making and and sometimes worse
24:59 uh uh decision quality. So yeah, I think that that's that's that kind of
25:02 something really interesting since the role is really boring. It's like um you
25:06 you can at least when I was at random AI, we start experimenting for the
25:10 growth team. We just have put one designer on the growth team. He's
25:15 responsible to like look at all the user feedback uh and and and figure our idea
25:20 and just you know ship a PRS and we do have a one engineer to review that. uh
25:25 uh that that that setup wasn't possible before and and and I think yeah I I I
25:30 haven't I kind of left after that so I don't know whether it work out but it's
25:33 very exciting right with with those type of setup uh so so my my gut feeling is
25:38 that potentially actually pretty keen to hear like with this setup probably we
25:42 don't need very big team instead of you maybe have more team more initiatives
25:46 but just one or two person start shipping a lot of stuff and holding all
25:49 the context and autonomy >> yeah I was going to ask what do you see
25:53 as the future of this like product team that's trying to like ship a feature.
25:56 What does that look like in the future? Is it a designer and an engineer and
26:02 like do the PM and front-end developer roles kind of get made redundant? Is
26:07 that like where we're moving towards? Let's take a designer as example.
26:11 [snorts] Traditionally designer team has a huge backlog. Some of them just like
26:18 low-level basic things. uh you know uh we we have part feature we need to add
26:22 you know growth banner here or we need to add button here all those requests
26:26 still need designer traditionally but those things whether it's still need
26:29 needed to go to designer can PM or engineer they just connect to a system
26:33 they already know their you know design system uh your your product contest just
26:37 spin up something that is good enough and to be honest I also feel like AI
26:42 with right contest might produce better outcome from designer uh than
26:45 traditional designer for those low-level things as So that's kind of like potentially one
26:51 way is more like look at each row uh what their current backlog what are the
26:56 parts that can potentially uh handled by by by by another role with the AI help
27:02 and so then that role just focus on more like difficult uh high quality problems
27:07 move up the value chain like that's and so like I feel like so much of what all
27:11 of our AI tools are going to like do is like forcing us up the value chain and
27:15 you can like do more do you think it like Not only so it like saves time for
27:20 designers. What's the like aha moment? >> Yeah, good question. Uh so Google has
27:25 this uh I think it was an introduce this concept of gen generative UI.
27:29 >> Uh what that means is you know currently you still everything still goes through
27:32 design work but if you have one agent that's so great at come up with design
27:37 on spot within the context and hopefully very fast. Do you do you need to go
27:41 through a design process? So the the concept is like what if we have this
27:45 chat interface when you you know currently when you chat to chat to BT
27:48 ask what's the weather it will just give you a bunch of text but you can like
27:52 have have agent to generate UI dynamically uh so that uh so that's a
27:59 new concept they introduced um it's not heavily adopted at moment yet because
28:04 it's just slower uh and like trade-off for user right do you trade off for the
28:08 uh for the longer waiting time to get a a slightly better data visualization
28:13 in many cases is it's not RORI is not there u but as a change I think there
28:19 there could be scenario that very that the ROI is is actually there um so if
28:24 you can think about I'm a user I go to your website uh it can based on my
28:29 information dynamically craft the experience for me um that would be quite
28:34 interesting uh kind of gets away from like what everyone hates by which is
28:37 like AI slob to like the ultimate form which is AI personalization like your
28:41 actual entire experience is the way you prompt, the way you like experience
28:45 something could be kind of cool. >> I'm really curious like something that I
28:49 really struggle with um when building is like I have an idea for a product that I
28:54 want to build then I get extremely stuck in the product design part because I
28:58 also have a particular vision in mind but I'm not a product designer and I get
29:02 I can't think like a product designer. I don't know like where to even begin. So
29:08 for say product minded founders who you know have this vision how do you suggest
29:13 them using these new product design AI tools is it does like where is the value
29:18 is it the prompt engineering that you need to know or is it knowing the right
29:22 references like how what how would you suggest them get to their product vision
29:26 by using a tool like yours? >> Yeah good question. Um so
29:32 uh I think a lot of people currently using the coding tool is like you know
29:38 go me to the app. It's like imagine that you have engineer in front of you as a
29:41 PM you never do that just go be me to the app you're giving proper respect and
29:45 that's where you you engineer human engineer start lining. So I feel like
29:49 it's probably same case for for for the uh for AI as well. uh so generally the
29:56 quality star uh the quality of the of of the of the of AI is really depend on
30:00 input you give to it uh so which means the more kind of a plan and context
30:05 alignment you can do with AI is better uh and uh so normally let's say I'm
30:09 building landing page normally the the flow I I personally use a lot recently
30:14 is that I would go to you know the best model uh let's say GS3 pro uh to give a
30:20 lot of product context to it uh and then tell this all the product, this the user
30:24 help me think through what are the content structure on the on on the
30:29 website. Then uh obviously back and forth a few times and until I got a
30:33 version that I feel pretty good uh then I will take like section by section uh
30:39 to to go to actually I use nano banana as well. I will take this this section.
30:43 Let's say I have this hero section. I know I want to talk about this this and
30:48 this and then again attach with my brand assets and generate
30:52 a a really good UI mo because I found image generation model is actually more
30:57 creative like it's max creativity compared with coding agent. Uh so I I do
31:02 that and then we got a moch there. uh and then feed to a coding agent who are
31:06 more maybe trained on top of the like you know design and front end design to
31:10 to output it. >> What about visual design like you know
31:14 something that an agency you would pay an agency for whether it's like branding
31:18 your colors you want something that stands out whereas I find with a AI
31:23 tools it's still very much I mean it's probably better with Gemini now but you
31:26 you get your classic purple buttons and like you know the typical AI generated
31:32 stuff. So any tips on you know standout visual design using tools like this?
31:38 >> Yeah, I think uh uh actually with Gemini 3 it changed a lot. It it was uh I would
31:42 say one month ago no matter how much you prompt it still go back to the to the
31:47 old one but uh with with new model what do we sell it became really creative uh
31:52 and uh the the part you really want to do is you you want to uh almost like
31:57 point their creativity towards a direction you want. All right. And and
32:01 that's where you start bringing the context of your product, your user, your
32:05 your brand style to making sure to produce something something really good.
32:07 Um >> I need to play around with >> Gemini for past three weeks. So
32:14 >> nice. I want to hard pivot a little bit to the idea of kind of build versus buy.
32:20 Um, one of the things I've observed is that when you um, in all the companies
32:26 I'm working with today, there's a legitimate answer, a legitimate question
32:29 of whether they should really bother buying software tools anymore or whether
32:33 they can just build it themselves because a lot of the kind of low-level
32:36 software tools that you would normally do, um, you know, you would normally buy
32:41 a software tool for 20 bucks a month, 50 bucks a month, even a few hundred bucks
32:45 a month, you can probably build it yourself pretty quickly now. Um, how are
32:49 you thinking about that internally? Like where would you choose to build? Where
32:52 would you choose to buy? And and I guess like projecting three months, six months
32:56 ahead. Like we've just spent a bunch of time talking about how I mean this is
32:59 kind of an advert for Gemini at the moment, but like [laughter]
33:02 we've talked a bunch about how Gemini 3's unlocked new capabilities. Like
33:06 that's I mean probably by the time this comes out that will that will be an old
33:10 model and there'll be kind of two or three models ahead. So I guess how are
33:15 you thinking forward on that as well? Yeah, it's a I spend most of my time
33:21 daily thinking about this. Um, just a personal fascination. I think more and
33:27 more low value, low lift things will have to be build yourself because often
33:32 times like you get a product that fits like 90% of what you want, but there's
33:35 always that 10% that you're like, I wish it did this and I wish it did that and
33:40 then you push the CS team and you hope that it ends up on the product road map
33:44 and then it never really ends up there. In today's world, the argument genuinely
33:48 becomes that um when you total up the cost of software, is it easier to just
33:53 hire one person to do that and how much of a percent of a time would it take?
33:57 Like we have softwares for lead routing, we have softwares for this and that and
34:01 specifically very niche softwares that do um middleware between our different
34:06 tools and can we just build that today? um we've talked about how far coding
34:11 models have come um in maintaining those coding models has become and so is that
34:16 like potential I very much think so now the other side of the argument is
34:20 you always have this option of like you know building internally but who's going
34:25 to maintain it long term um and that obviously becomes a big one and then
34:31 like do you want to shift focus away from what is your core product and your
34:37 core USB um and it will still take resources and you have to start shifting
34:41 the focus of the business um away from what you deliver to customers.
34:44 >> It almost feels like there's this software budget and that software budget
34:47 is just going to get shifted into like an internal ops team where there are
34:52 just a few people internally this kind of center of excellence the way you were
34:55 describing it and it's their job to kind of build and maintain these tools that
34:59 you would have typically spent like 20 grand a year on. Yeah, I I think I think
35:05 for me it's definitely build internal tooling and workflows like you have at
35:10 tracksuit. Um, we haven't really started doing too much of that just because
35:13 we're we're a tiny team and I do think like we'll have an internal ops team at
35:17 some point where they will be responsible for internal tooling whereas
35:22 I can't see us building say a headless CMS for our website and blog because
35:27 that is a lot to build and so I think it depends on the scope of the tool but
35:31 when it's extra personalized I think it makes a lot more sense to build
35:34 internally because you ultimately know and can be in control of what that
35:38 workflow is. Um so that's kind of how I would think about it. Um for chronicle
35:43 >> actually I car like in in in the track it looks like you guys did a lot of
35:47 experimentation. Is it set up you guys just have one person like AI ops who is
35:51 responsible for maintaining the software or >> Yeah. So right now it's actually really
35:55 curious and we're going through looking at how we do this next year. At the
35:59 moment our AI I've got a market ops team is maybe five people. Um and whoever
36:05 builds maintains. So if I build an agent today, I maintain that agent. But we are
36:10 thinking long term about how do you even in within that model start to define out
36:14 roles, right? Like there are roles that are better at prototyping and shipping
36:19 version one of an agent or of an internal tool that needs to maintain
36:22 something and there are other people that can then take that tool and make
36:27 sure it's not only driven end to end but get real value like again it goes back
36:30 to personality types but get real value from seeing something completely being
36:34 finished and executed. and how do we develop those people as well? So, you
36:38 might be able to ship a large degree of agents really niche across the business
36:41 and then have other teams maintaining and driving that long term. And that's a
36:44 shift that we're actually like currently working through. >> That's a big team, five people.
36:52 >> Yeah, five people. Um, but our go to market team over like it serves a whole
36:55 go to market team. So, that might be I think nearly 90 people across sales,
37:01 marketing, um, customer partnerships like all across the board. And so when
37:03 you actually like work that out in terms of ratio, it's very small by the type of
37:08 impact you can drive. And I think as orgs start to go forward over the next
37:13 like few years and as you guys are like prime examples of is you don't want to
37:17 scale through headcount. You want to scale through efficiency, right? Like no
37:20 one's out here wanting to just stack headcount for every problem. It's about
37:24 how do you bring efficiency into the business? How do you move I think um
37:28 Jackie had sent me this stat. It's like sellers spend 30% of their time actually
37:33 talking to like clients. Why should that not be 60 70%. And then your like
37:39 productivity like your AR per rep everything um will be and already I
37:44 think are core measures of success for any business. >> This is something I've been thinking
37:49 about a bit because it's hard as a board member sometimes to get a sense of how
37:55 much these metrics should be changing. How are you thinking internally in terms
37:59 of like sales quotota is a really good example of one where you think well with
38:03 tools today like maybe you typically would have asked a sales rep to deliver
38:08 a million dollars a year and like should that now be 1.5 or two or um you know
38:14 you don't want to move the goalposts on people but also you you should with this
38:20 extra leverage and you know salesperson's time with customer going
38:24 from 30% to 70%. Should that change their outcomes as well? I mean, how are
38:28 you thinking about setting the bar on performance internally as well these
38:36 >> It's it's hard to directly change the end metrics. uh because I when I was
38:41 working at random AI um so we were launching that AI sales sales agent and
38:46 when we sell to the to the customer obviously their expectation okay can you
38:50 just deliver more you know deals but that that's hard because there so many
38:54 like points to like you can deliver something that promise automating things
38:58 but it's hard to design deliver something that promise delivering the
39:03 final output um so but I don't I found what what worked kind of well is like
39:07 it's kind of when you write the probability it's like when you do growth
39:10 experimentation you also don't probably just optimize for you know the the user
39:14 turn you kind of start defining the leading indicators for for the product
39:18 so I guess potentially it's very similar things for each row you start defining
39:23 those leading indicator like for sales agent what we found is that like the the
39:28 the number of activities uh that that that can deliver and and we just
39:32 optimize for that uh and and with hypothesis that more activity you do you
39:36 should lead to more uh final hopefully revenue but but it's hard to
39:41 just measure put it direct harder you know revenue goal but maybe you saw some
39:45 portfolio company did I don't know >> well I'll say I like partially agree but
39:50 I'll give you a contrary opinion as well right um which is yes I think a massive
39:54 part of like a sales agent is how do you make it more efficient do the admin blah
39:58 blah blah right but that's also partially the boring stuff I think like
40:02 on the flip side is how do you actually drive more revenue and I don't know if
40:05 this is a direct leader quota yet cuz that has overall implications from like marketing and a
40:11 bunch of other sources. But how do you actually use like a sales AI agent to
40:16 deliver more quota? It's actually how do you lift metrics throughout the
40:19 business. So one example that we've specifically done is looked at how can
40:24 we deliver custom assets to prospects before they've ever met with us. Right?
40:28 So like an example for us, we know our product actually is like an aha moment.
40:31 When people get a demo, they're like, "Oh my god, this is amazing. We need
40:35 this in our company." And so it was all about how do you bring that forward to a
40:39 prospect. And with AI now we're able to deliver customized like personalized
40:43 dashboard to each prospect from like an SDR which was which would have never
40:47 been able to be done before through like a dummy environment with dummy data. But
40:50 today you're able to like bring that to a customer and so th that's where like
40:54 you can start to get incremental and saying no no we are going to drive
40:58 greater incremental pipeline and greater incremental revenue specifically driven
41:04 from an unlock through an AI ops team that we weren't able to do before.
41:08 >> Jason you're and and I guess to you to a degree as well like you guys are at the
41:12 very beginning really of the journeys. I mean like Chronicle is only a few years
41:16 old. Um you're really just getting started in the last few months. But how
41:22 will you build the team differently today versus how you would have I guess
41:26 at the other startups you worked at at in the past and and what I guess if you
41:30 forecast 12 months forward like you what do you see the composition of the team
41:33 being? >> Yeah. Uh uh so yeah so maybe I share like so two
41:42 star I worked before civic culture and and also relevant CI and I or join them
41:48 like very during the growth stage so I al so traditional way obviously just
41:52 grow rapidly with with with number of people um but the uh but yeah we we talk
41:58 about all the problem that you can also have come with those head concrete the
42:01 way I'm thinking about building the the team is that we have just a ideally a
42:06 very small amount of people um hopefully 10 people team uh and you you get you
42:11 care more about like the revenue per hat uh that's kind of the metrics I think
42:15 will be more interesting and that come down to uh each individual is like
42:21 really strong um AI native which means they like needs to be strong and also AI
42:25 native which means they automate a lot of work they it should work in in uh
42:29 that different from traditional traditional ways um and uh yeah so
42:35 that's kind of like the the way uh we currently think about uh uh the team
42:42 structure. Uh for me as a little staff, we we I feel like we just want mostly
42:48 engineer uh to to to uh to deliver the product and and just uh you know if it
42:54 depend I think it really depend on a B2B or or or or bottom up growth. Uh so if
43:00 the bottom of growth which is my where I am I just won't hire good engineers. uh
43:05 and and that's it. Um but if the B2B obviously you need to do the like sales
43:10 side that would be a that part I don't know whether you can um just change the
43:17 par um uh um yeah paradigm there totally. Um yeah for for PRG I can see a
43:21 lot of change. >> Yeah that makes sense and you're already
43:25 seeing it with so many companies like curs is a billion in revenue. I don't
43:28 know it's got tens of employees I think. So um yeah it's pretty spectacular.
43:31 Yeah. >> But what about you, Ry? For me, um I've been thinking about like what does
43:37 Chronicles growth org look like moving forward? Like how will it change over the next
43:44 few years? And um I look after a lot of our content like product marketing, SEO,
43:49 um socials, and I think that that will get to a point where it'll be automated
43:53 to an extent where there will be workflows where um posts get scheduled
43:58 by itself and we just like generate an asset that's on brand. And I I I see
44:04 myself automating that and then moving into more of the B2 B2B side and doing
44:09 more sales and spending actual time with prospects and customers. And I so I kind
44:13 of see my role shifting that way over the next year. Um and when it comes to
44:18 who will I actually hire and who will I look for, I would go for more
44:22 specialists. Say for example, performance marketing is something that
44:25 I have no idea how to do and I would hire specifically for those um because
44:31 you can now be such a generalist with creating all these workflows that you
44:35 can create your own little associates essentially um in the workflow. So I
44:43 would go more specialist um and then move on to other parts of go to market
44:47 and tackle them kind of one piece at a time um rather than and I don't see the
44:51 team getting bigger than a couple of people um over the next year or two
44:56 actually. So I don't think that the volume of people will grow that much at
45:00 least in the growth side of things for Chronicle. Just to round it out, I mean,
45:05 if you were um giving advice to anyone today who who may be a couple of years
45:10 or months behind you in this journey, you know, where would you get started? I
45:16 mean, like down to, you know, tomorrow, log on to this, read this particular
45:22 document online. Like, how would you get started tomorrow? Maybe I'll start with
45:25 you, Scott. >> Number one, I'd go watch Jason's YouTube
45:29 channel. Um, it's it's a plug, but I actually um love the content. But no,
45:36 um, I'd say pick a personal problem um, and start building it. In terms of you
45:41 can go to chat today, take a long walk, turn it on voice mode, tell it the
45:44 problem you want to solve. It'll give you the instructions and the tools and
45:49 everything. And you might flop three times doing it. I know I did whenever I
45:53 was building anything. Um, but go do that and and you'll break a lot of
45:56 things along the way, but you'll find the limitations of what is and isn't
46:00 possible. And in doing that itself, you'll you'll learn a tremendous amount.
46:05 >> What about you, Renee? >> I same thing, like find something that
46:11 would excite you to solve. For me, my first app was a um wine AI where I would
46:17 go to a wine bar and I wouldn't know what the wine menu said, like all these
46:20 different wines and I would take a picture and it would tell me based on my
46:24 palette what you should order. Um and it was just fun for me. So, pick something
46:30 small, just ship it, like go to chat or you can use cursor these days and tell
46:34 it this is what I want to build and it'll give you every single step that
46:38 you need. Um, so just follow what they say and like try to get it to a shift
46:43 point. Um, and then my second thing would be once you have a sense of
46:48 building, I learn how to market yourself or market the product and get
46:53 comfortable with posting on LinkedIn or Twitter or whatever your channel of
46:57 choice is, but start to put yourself out there and get over the hump of coming
47:01 off as cringe. That's something I had to get over. And um, I think it will open a
47:05 lot more doors and opportunities. you'll meet new people and um I I think it
47:10 comes hand in hand like building and also creating content and marketing.
47:16 >> Yeah. Um I I just thought about a uh thing like so yesterday I was interview
47:20 uh one of the other engineer and my founding engineer asked me a question in
47:23 the end as an engineer coding agent is getting so good what's the value you're adding
47:29 as a human and I thought that was brutal man and uh uh but that's a good question
47:33 like assume the future based on all we talk about is that uh the people who
47:38 will be really good is like uh people who your uniqueness come from certain
47:43 aspect you might be using AI tools but you just have better understanding of
47:46 customer, you just have better on the taste. There's something unique about
47:49 yourself. So you should like uh just build things based on your unique uh
47:55 kind of quality. Uh but uh also uh when I look at think about all the all the
47:58 all the all the like talents on the market we you want to making sure that
48:02 business don't need to make a hard choice. Do we hire someone really good
48:07 experienced or someone is AI native? You want to be the person in the middle. So
48:10 uh which come down to like you just try a lot of yeah tools because so new that
48:15 no one knows best practice to automate whatever role you're doing u and but you
48:19 just need to try those things so you get a lot of like good taste about what's
48:23 going to work what's what's going to hit the quality bar uh so I I would say do
48:28 that and uh uh and fundamentally if you work in part or you should uh you should
48:32 probably learn a bit more like fundamental technical about like large
48:36 language model how agent works and they also come down to just build your own
48:40 personal product using cursor then and they just learn a lot after just one
48:43 project. Um that's probably >> if I had to ask for one recommendation
48:49 from each of you either a podcast or a blog substack um anything else like
48:58 where would you go? Uh I mean this channel called AI Jason. So
49:02 [laughter] >> people can go uh >> uh and uh uh I mean the uh like if if
49:09 they're like engineers they they're like a YouTube channel called AI engineer
49:13 which is really good. They always share like the latest kind of technical thing
49:19 and uh uh for like design I I personally uh follow this one person called Mento.
49:23 I think he share a lot of like good good design tips in props.
49:32 M E N G T O. Yeah. Uh and uh and and uh uh and I think part of person you have
49:37 the uh Lenny. Lenny. >> Yeah, Lenny. He's pretty good. He invite
49:42 a lot of good people. >> I was going to say Lenny. Um also
49:46 because if you do the not a plug for Lenny, I don't get anything back. But if
49:49 you do the yearly subscription, you get like a access to pretty much every AI
49:54 tool for free on the pro version for a year, which I think again goes back to
49:57 the experimentation point of view. Subscribe and then you also get access
50:00 to actually go like they push you to go and build. Um, which I think is like
50:04 actually the best way to go do it. >> Nice. >> This is specifically a a tech
50:08 recommendation. >> No, it could be anything. Um, I've actually actively been trying not to
50:16 consume startup slash AI content. Um, because I find it it's it stresses me
50:20 out like just being like working in it and and all of that. So, I've been
50:23 actually trying to like go completely the opposite direction. I have a book
50:27 recommendation. I'm reading a book called A Swim in the Pond. Um it's a
50:33 book which analyzes a number of Russian short stories like Toltoy and Czechov
50:37 and it's an English professor who breaks down those essays and it kind of talks
50:41 through the techniques in that book and it's just been really interesting kind
50:44 of work that I would never read before because I would find so boring. But
50:49 actually understanding the the pros and the literature that's been actually
50:53 really refreshing. So, I sit down every morning with my coffee and do a bit of
50:58 that before I jump into work and tech. So, that would be mine.
51:03 >> That sounds so wholesome. Love that. Well, thank you all for your time.
$

The era of the non-technical engineer

[AI agents and automation][marketing and growth hacking][product development and MVP][productivity and workflows][solo founder and bootstrapping]
// description

What happens when non-technical people get access to code as leverage? For decades, code and capital were the ultimate forms of leverage. If you couldn't write code yourself, you needed to hire engineers. But now you have designers shipping code. Marketers building products and one-person growth teams doing what used to take 90 people. Welcome to the AI-native era. Join Jackie Vullinghs, Partner at @airtreeventures, as she sits down with three practitioners who've built their entire workflow

now: 0:00
// tags
[AI agents and automation][marketing and growth hacking][product development and MVP][productivity and workflows][solo founder and bootstrapping]