YT YouTube2Text

YouTube2Text → Transcripts → How to Build a Company OS in Claude Code | Jiaona Zhang | Product Growth

How to Build a Company OS in Claude Code | Jiaona Zhang | Product Growth — Transcript

by Aakash Gupta · 13,745 words · 1,997 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00You got these people who are these 1% AI
0:02users. They're highly AI pilled. And
0:04then you have the 90 to 99% of the rest
0:06of the organization who isn't sure what
0:08to use when.
0:09>> Meet Jay-Z. She's the chief product
0:11officer at Luro, the $100
0:14AI time platform. She teaches PM at
0:16Stanford. Before Jay-Z, she was the CPO
0:18at Linktree, and she's led product at
0:21Airbnb, Webflow, Dropbox, and WeWork.
0:24>> something pretty crazy in your
0:26interviews. Can you tell me how you
0:27interview people and really find these
0:30damn AI pale super IC PM?
0:32>> The fundamentals and the principles have
0:34never changed. In fact, they're even
0:37more important than ever before, but the
0:38tools and the way you operate, that's
0:41radically changed.
0:42>> How should people be thinking about in
0:44an AI native organization? This is the
0:46role of a PM.
0:47>> And so those are the four levels. Level
0:49one is you're talking to ChatGPT, you're
0:50talking to Claude. You're really using
0:52AI kind of in chat mode. Level two is
0:54where you start to automate a workflow.
0:55Level three is when you start building,
0:57you know, apps. And then level four I'd
0:59say is where you're actually building I
1:00call them shared apps.
1:02>> How does someone start from step one?
1:04What is the process somebody needs to go
1:06through in order to build up and create
1:08their own company operating system?
1:15Before we go any further, do me a favor
1:16and check that you are subscribed on
1:18YouTube and following on Apple and
1:19Spotify podcasts. And if you want to get
1:22access to amazing AI tools, check out my
1:25bundle.
1:26Where if you become an annual subscriber
1:28to my newsletter, you get a full year
1:30free of the paid plans of Maven, Arise,
1:33Relay App, Dovetail, Linear, Magic
1:35Patterns, Deep Sky, Reforge, Build,
1:36Descript, and Speechify. So be sure to
1:38check that out at bundle.akashg.com, and
1:41now into today's episode.
1:46Jay-Z, I've been teaching people a lot
1:48about how to use Claude code with
1:50personal operating systems, with team
1:52operating systems. You guys at Luro have
1:55taken it to a level I have not seen
1:57before. You guys have built out a
1:59company operating system.
2:01Can you show me what this is and what it
2:04does?
2:04>> Of course. All right, let me screen
2:06share here. Okay, let's start here.
2:08Let's go to GitHub, our favorite place.
2:10And so you'll see here that we have in
2:13GitHub a company-wide operating system
2:16where for every single function in a
2:18company, customer success, data science,
2:21design, engineering, finance, imitation,
2:24legal, marketing, we have um
2:27essentially all these folders that share
2:30how do you think about each phase of
2:33work that that function does. So in
2:35customer success, you do account
2:37management. And within account
2:38management, you're thinking about, you
2:40know, renewals, upsells. Um you do a
2:43customer enablement. And within that, we
2:45essentially work with our customers, we
2:47do office hours. We help them with roll
2:49out. We do training and onboarding.
2:52Each of these folders have a skill. And
2:55I think uh for those of you who are less
2:57familiar with GitHub, we'll actually hop
2:59over here to something that is very
3:01familiar, which is essentially your file
3:03structure, your folder structure. And so
3:05going to customer success, you can see
3:07that each of these folders have a series
3:11of um you know, folders that are the are
3:14the activities that they do. And then
3:16within each of them, they have skills.
3:18So how do you actually think about
3:20creating the right assets for the
3:22negotiation support or the right
3:23references? I'll go back one more. Um
3:25for renewals, right? Um what is the
3:28skill file there to really think about
3:30how do you walk through a renewal
3:32correctly with a customer?
3:34And now you're like, okay, cool, you
3:36have some folders in GitHub, you have,
3:38you know, some some stuff that you can
3:40download. How does this all come to
3:42drive
3:43like real change? And the way I'll talk
3:46about this is, you know, at the end of
3:47the day, we all live in some form of
3:51email or Slack. And so what I'll do
3:53really quickly is I'll open up my Slack.
3:55And again, this is not real data in the
3:57sense that we do have very sensitive
3:59data that I'm not going to be be
4:00sharing. So, this is a little bit more
4:02mock, but it shows you exactly how our
4:04how our team operates. So, for example,
4:07every single morning
4:08every person on a lot of these
4:10customer-facing teams, right? They're
4:12highly repeatable motions, the more we
4:15can sing from one voice and say the same
4:18thing the way we can create consistency
4:21in the awesomeness of the customer
4:24experience, that makes your company, you
4:27know, much more unified and
4:30it's a big part of the brand. And so,
4:31when you think about that and you think
4:33about a customer success person waking
4:35up in their day
4:36and really seeing, let me go here. This
4:38is a example for customer success.
4:40Here's your calendar. Here are all the
4:42meetings that you have, the check-ins
4:44that you have, you know, the onboarding
4:46sessions you have.
4:47This is something that a lot of people
4:49are building. This is a example of a
4:50chief of staff light concept, but what
4:53we're now doing is we're integrating all
4:55the skills. So, for example, when we do
4:57a handoff, when we do a session prep,
5:00all of these are actual skills. And what
5:02happens is then when anyone is using
5:05Claude, for example, I'll just go into
5:08I'll go I'll go really quickly into the
5:10organization settings and I go into your
5:12skills. You can start to see that you
5:14can upload all of these skills into your
5:17company context. And as a result, when
5:19you're going through your day, you can
5:21essentially say, "Great. I'm going
5:23through my day. I'm doing all these
5:24things.
5:25My
5:27I will use these skills so that I no
5:29longer have to spend all the time
5:31creating that one deck or spend all that
5:33time creating an email. It is actually
5:35something You know exactly what skill to
5:37use when. And I think that's the biggest
5:39thing that companies struggle with,
5:41which is you got these people who are
5:42these 1% AI users. They're tinkering
5:45with their workflows. They're highly AI
5:47piled, and then you have the, you know,
5:4990 to 99% of the rest of the
5:51organization who isn't sure what to use
5:54when. And so, as a result, you can
5:55actually integrate your skills, again,
5:58at a company level. So, across every
6:00single one of these functions, going
6:02back to files, each one of these
6:04functions and all the activity that they
6:05do in order to be able to understand
6:08what, um, skill should I should I be
6:10using when, and where should I be
6:11spending my time? Maybe the last thing
6:13I'll just show to kind of, uh, to really
6:15bring this to life is every single
6:17company, you can map every single
6:19function's work to what I call an
6:21ontology. So, in sales, you know, all of
6:24the work in sales maps to these
6:26categories that they're supposed to be
6:27doing. And within each category, there
6:30are series of tasks that happen. And
6:32this is actually what has informed the
6:34ontology that I just showed you. We've
6:35done the really hard work of mapping
6:37out, okay, for every single function,
6:39again, I'll scroll through this.
6:41Marketing, sales, customer success,
6:43implementation, design, engineering, so
6:44on and so forth. These are the things
6:46that we believe that each function
6:47should be doing. How do we actually
6:49create, um, a set of skills to for for
6:53you to, um,
6:54do the things that we want you to be
6:55doing more, and to also automate the
6:58things that we don't want you to be
6:59doing anymore. So, I'll go to product,
7:01which is, you know, a lot of the
7:02audience here today. In product, what's
7:04really interesting is that,
7:06you know, you should be spending your
7:08time like an engineer in many ways. And
7:10we talk about this later where, you
7:11know, the ontology or the work map of a
7:13product manager is starting to look look
7:15a lot more like an engineer. But there
7:17are a lot of things that used to be in
7:19the, um, day-to-day of a product
7:21manager, doing competitive market
7:22analysis, doing these all these like,
7:25writing for stakeholder management, or,
7:28um, really mundane, tedious
7:30organization, um, getting people on a
7:32phone, synthesizing, um, feedback, etc.
7:35All of these things, as we all know, are
7:37starting to get automated. But again,
7:39it's automated in a really lumpy way,
7:41where one PM might be doing a really,
7:43really well and other p.m. I might not
7:44be doing it as well. So, what we can do
7:46here is when you onboard everyone with a
7:48company OS, again going back to this
7:51GitHub,
7:52and going to, let's say, product, right?
7:55Um
7:56you can start to say, "Hey, these are
7:57all the playbooks, all the skills that I
8:00want to give every single person on my
8:02team." And then, when they come in for
8:04their daily briefing, what ends up
8:06happening is that they are able to see
8:09their day at a glance, and we
8:10essentially tell you where you can
8:12automate your day. So, you take the
8:13thing that is a that is essentially
8:15designed by the 1% of of every any given
8:18function, the person who is playing
8:19around the most, and are able to spread
8:21those learnings throughout the entire
8:23rest of the organization.
8:25>> Wow. I think this is so powerful because
8:27we all have been working in different
8:28teams where there's that one person
8:31who's got their skills locked, but if
8:34they're just compounding in a bucket,
8:35then nobody can really benefit. This
8:37company OS, this is bringing that power
8:40to everybody.
8:42Now, you guys are an AI-native company.
8:45You guys are an AI company yourselves,
8:47and so you guys would have certain
8:49advantages
8:50in building this. How does someone start
8:53from step one? What is the process
8:55somebody needs to go through in order to
8:57build up and create their own company
8:59operating system?
9:00>> I like to think about it as three
9:02different steps. And so, let me screen
9:03share again, and I will um share how do
9:06I think about essentially um getting
9:08your steps in, going from most simple to
9:11most advanced. So, the first way to
9:13think about this is, how do you just
9:14start small? What is one workflow that
9:17you or your team does that is incredibly
9:20tedious that you shouldn't be doing
9:22again? So, typically, for many, many
9:24functions, it is, you know, I write this
9:26email, and I want this email to have a
9:28template that is automatically, you
9:30know, um kicked off for me when um XYZ
9:33things happen, or there's a sequence of
9:35things that happen. I don't want to
9:36input um my data into our CRM anymore. I
9:39want that to be automated. So, there's
9:41some degree of thinking about what is
9:43super mundane, takes a lot of time out
9:45of your day-to-day, and if that were to
9:48be automated away, you'd be thrilled
9:49about. And I'll give you one very
9:52product-oriented example, which is there
9:55are so many companies out there, so many
9:57PMs out there that spend a lot of their
9:59days responding to questions,
10:01escalations. So, the sales team comes
10:04into a channel
10:05>> I'm notoriously bad at my inboxes. I
10:07guess there's a version of that where I
10:09seem cool and unavailable, but the
10:11reality is I miss sponsor emails, guest
10:12pitches, and stuff that my team actually
10:14needs me [music] for. So, I got an AI
10:16assistant, the sponsor of today's
10:18episode, Arise.
10:20Arise connects to my email, calendar,
10:22and [music] Slack. Then I just chat with
10:23it over Slack, and it helps me with
10:25everything. It builds workflows to
10:27respond to emails, resolve [music]
10:29customer issues, prep me for meetings.
10:31It actually comes to my meetings,
10:33>> [music]
10:33>> updates its own notes, and remembers
10:35context from past conversations. So,
10:38every time I talk to it, it already
10:40knows what I'm working on.
10:41>> [music]
10:41>> I used to pay for Granola and Lindy
10:43separately. Arise replaced both. One
10:45tool does more, and it lives right in
10:47Slack where I already work. [music]
10:48Check it out at arise.ai/akash.
10:51That's a r i s o dot a i slash a a k a s
10:55h.
10:55>> Here's the dirty secret about
10:56prototyping. You spend two weeks
10:58building a prototype, you validate your
11:00assumptions, engineering loves the
11:01direction. Then what happens? You throw
11:03the whole thing away. Bolt changes this
11:05completely. When you prototype in Bolt,
11:07you're not building throwaway mock-up.
11:09You're building real front-end code that
11:11integrates with your existing design
11:13system. So, when you hand it to
11:14engineering, they don't throw it away,
11:16they ship on top of what you've built. I
11:18use Bolt every single day. I host my
11:20land PM job cohort on it, and honestly,
11:23I'm up till 2:00 a.m. some days just
11:25vibing in the tool, having fun, and
11:27building. That's when you know a product
11:28is good. When you're using it past
11:30midnight not because you need to, but
11:31because you want to. Check out bold at
11:33bold.new/akash.
11:35That's b o l d . n e w / a a k a s h.
11:40Link in the show notes. Today's podcast
11:42is brought to you by Pendo, the leading
11:43software experience management platform.
11:45McKinsey found that 78% of companies are
11:48using gen AI, but just as many have
11:50reported no bottom line improvements.
11:52So, how do you know if your AI agents
11:54are actually working? Are they giving
11:56users the wrong answers, creating more
11:58work instead of less, improving
11:59retention, or hurting it? When your
12:01software data and AI data are
12:02disconnected, you can't answer these
12:03questions. But, when you bring all your
12:05usage data together in one place, you
12:08can see what users do before, during,
12:10and after they use AI. Showing you an
12:12agent's work, how they help you grow,
12:14and when to prioritize on your road map.
12:16Pendo agent analytics is the only
12:18solution built to do this for product
12:20teams. Start measuring your AI's
12:21performance with agent analytics at
12:23pendo.io/akash.
12:25That's p e n d o . i o / a a k a s h.
12:30>> Okay, so let's move into Slack and see
12:32what this might look like. You know, a
12:33lot of companies, if you just go into
12:35any ask product channel or any channel,
12:37you see so many um success folks, um
12:41support folks, sales folks, um other
12:43teams hitting up that channel asking
12:46people, "Hey, I have a question. I have
12:47a feature request." And so, the very
12:49small workflow that we did, and I'll go
12:50all the way down, is we created a Slack
12:53automation that essentially said, "Look,
12:56when a feature request comes in, we
12:58typically spend a bunch of time going
12:59back and forth asking about how many
13:01times was this asked about, send me the
13:04gong recording where I could watch what
13:06the customer's actually saying, um what
13:08is the impact of this for your customer,
13:10which requires some degree of judgment
13:11from the person managing the account, um
13:13what is actually going on here, give me
13:15some more details." All of those things
13:17usually require back and forth. So,
13:19again, if I go back to this this system
13:21of how do you think about a place to
13:23start? What is something that you do
13:24over and over again that you could
13:26really easily automate. And that
13:28automation for us was as simple as,
13:31"Hey, let's just automate what we ask
13:33someone to fill in. And then what often
13:36happens is you then have to triage it.
13:37You say, 'Hey, you know, is it for this
13:40team or that team? Is it for this PM or
13:42that PM? And what's the SLA to getting
13:44back to the requester on what we're
13:47doing about this feature request?'" And
13:50so all of that you can build into
13:51something as simple as Slack. So again,
13:53a lot of people have Slack, Teams,
13:55whatever it is you're using to chat with
13:56your teams. You can do something very
13:58simple where you essentially say,
14:00"Okay, great. I come in here. I'm going
14:02to automatically ask for all of this
14:05information. So, you know, what is it?
14:07Who is it coming from? What's going on
14:09here?" It automatically assigns it to
14:12the person that makes the most sense to
14:15go look at this. And then it
14:16automatically creates some kind of
14:17ticket so that we can track it. And so
14:20all of that, again, this is 101, I would
14:21say, right? It's just like a very small
14:23step in in creating your operating
14:25system. So, I start there.
14:26The next step is this idea of how do you
14:29start to really automate based on a
14:32bunch of things that your team is doing.
14:35And so the example here I have is, you
14:37know, again, a team that usually has a
14:39lot of people, a lot of humans.
14:41At Laurel, we have a large, you know,
14:44GTM team. And within GTM, go-to-market,
14:47we have really awesome success folks
14:50who are essentially, you know, what I
14:51call like time consultants. They're
14:53getting kind of forward deployed into
14:55these organizations, helping them use
14:57Laurel as a as a product. And so what
14:59we've done is we've essentially created
15:00a playbook. And again, this is very,
15:02very long. I think anyone who's ever
15:04created a playbook before, this is 50
15:06pages. It covers everything from
15:08implementation to onboarding to user
15:10onboarding. And and depending on who you
15:12are, is it the admin, is it the actual
15:14timekeeper, etc. You know, different
15:16onboarding. These things, by the way,
15:18are very fast now with Claude. You can
15:20actually create this from a lot of
15:21sources and have it be written really
15:23quickly. But what the struggle most
15:25companies has is now that I've created a
15:27playbook,
15:29how do I actually get people to do the
15:30playbook? And how much of the playbook
15:32is actually done by the human versus
15:34actually done by, you know, agents or
15:37workflow automations, right? And so this
15:39is where, again, going back to this
15:40concept of of the playbook um model,
15:43this is where you can say, "Okay, well,
15:45I've created a playbook. I've went
15:47through and I've audited the things
15:49that, again, it requires a human to do.
15:51It requires a human to get on the phone
15:52with someone. It requires a human to go
15:54fly on site. Um but here are the things
15:57that we think we can automate." This is
15:58either um something we can productize or
16:01this is something that we can create an
16:02agent to do. And so that is, I would
16:03say, the the next step that you graduate
16:05to, where you essentially create a
16:06playbook and then off of the playbook
16:08you decide on a set of skills. And And
16:11that's, by the way, where we um
16:13we started to to get the first version
16:16of the OS I showed you earlier. When we
16:18went into customer success and we said,
16:19"What are all the things that someone
16:21might be doing?" These large buckets. It
16:23was largely off of playbooks. The
16:26playbooks for implementation, the
16:27playbooks for um activating a customer,
16:30the playbooks for really um
16:32talking to them the right way to make
16:34sure that they're set up for success.
16:36And so that is really the the second way
16:37to think about it. And um maybe I'll
16:39share one thing here, which is there are
16:41a lot of um agent builders out there
16:43today in the world. So you could use,
16:44you know, Claude itself. They've
16:46launched, obviously, a lot of um agents.
16:48You can use um uh
16:50a lot of things from Open AI as well.
16:52You can use a Glean. You can use a Dust.
16:54We at Laurel use Dust. And so I'll take
16:58a moment to see if this loads.
16:59>> So if somebody hasn't heard of Dust,
17:01yeah, this is an agent building tool?
17:03>> This is an agent building tool. And what
17:05we find is often a lot of the um things
17:08that someone does can be turned into a
17:11series of repeatable steps that gets
17:13automatically triggered. And so a great
17:15example, and I'll just go scroll down
17:17here really quickly.
17:19All of these are agents that we have
17:21built. So, going back to the
17:23the playbook concept, if you say, "Hey,
17:26I have a playbook of all the things that
17:28you need to be doing here." And again,
17:3055 pages worth, I don't think anyone's
17:32going to read anything here. What we can
17:34start to do is go into an agent builder
17:36and say, "I'm going to create an agent
17:37for each of these steps." If I have to
17:40draft emails a lot as a customer success
17:42manager, if I have to actually scrape
17:44LinkedIn a lot as a salesperson, if I
17:47have to look at the market as a
17:48salesperson, or think about prospecting
17:50questions, each of these can be you can
17:52build an agent for each of the
17:55um
17:55parts of the workflow here. And then
17:57going back to really thinking about how
17:59does everyone engage with your operating
18:01system thoughtfully? No one's going to
18:04remember that they're going to call the
18:05specific agent that's going to do the
18:07email, and the specific agent that's
18:09going to do the RFP. The the big
18:11learning that we've had is how do you
18:12create a wrap like a like a mega agent,
18:15something like the like a
18:17a go-to-market agent that can be called
18:19by the sales team at any point, by the
18:22success team at any point, and then that
18:24agent is able to route the ask, the the
18:28need, or the help to whatever one of
18:30these sub-agents that is actually
18:33useful.
18:34And then going back to like it really
18:36the delivery piece is so important. Even
18:38the friction of coming to something like
18:41a different interface, coming to a desk
18:43and asking it questions, is is really
18:45low. In- instead, actually going into
18:48your um
18:49your Slacks, your emails, and delivering
18:51people um just-in-time
18:54playbooks and automations is really the
18:56way to go to get to the point where
18:58you're you're actually getting people to
18:59use the agents and the workflows that
19:01you've built.
19:01>> So, help me understand this part. Why
19:04use Dust instead of just all Claude or
19:06Claude code?
19:07>> Yeah, that's a great uh question. We
19:09started using Dust back in fall of last
19:12year. And so I think there was just a
19:13maturity of the tools. Um back then, it
19:17was just much easier to use something
19:19that specialize in agent building like a
19:21Glean or a Dust. I do think today
19:23there's
19:24um that gap is shrinking quite rapidly.
19:27And so as a result, I don't think you
19:28need to go out there and buy a
19:30specialized tool that does these. And in
19:32fact, you can just build them in Claude.
19:34Um and and this is actually a little bit
19:36where we're going, which is um if I go
19:37back to the operating system that I was
19:39showing you earlier. And all of these no
19:42longer have to go through a Dust or a
19:44Claude. Um
19:46instead, what we're able to do, make
19:49this much larger, is we can we can take
19:51all of these skill files and go into
19:53Claude itself and put them in as skill
19:56files. And so as a result, you could um
19:58now you can literally just say, "Hey,
20:00I'm inside whatever it is I'm doing, and
20:03I can just call the, you know, that
20:05skill
20:06/morning briefing product." And as a
20:08result, it gives me my briefing right
20:10there as opposed to me having to go and
20:12call an agent builder.
20:13>> Mhm. And then should people be setting
20:16up like Claude automations on top of
20:18these to be running these as your daily
20:20morning like running on a schedule or
20:21something like that?
20:23>> Yeah, that's a great question. It's so
20:24funny. Um I'll go I'll share a little
20:26bit of my personal experience. So I set
20:28up a bunch of these scheduled things.
20:29And even if I just go to scheduled, um
20:32I'll go right here. You can see that I
20:34have a lot of these scheduled tasks. And
20:37you only see a couple of these pinned.
20:39And what I found was that um
20:42I it was almost overkill. It was like I
20:45sat there. I was like, "Oh, I might
20:46automate this." And so I built it. I was
20:48like, "Oh, I might automate that." And
20:49so I built it. I was like, "That might
20:50be interesting information." I built it.
20:51And actually I think we're in a world
20:53where we are we have information
20:55overload. And so this is why we took the
20:57time as a company to be like, "We can't
20:59just assume that people, first of all,
21:01that they're going to do this for
21:02themselves. And second of all, that
21:03they're not going to be overwhelmed by
21:05the number of like automations and you
21:07know
21:08schedule things that happen. And as a
21:10result, that's how we consolidated it
21:12all into what I was showing you earlier,
21:14which is this this idea of actually
21:16having all in one place, because the
21:18chances that you're going to come back
21:20and say, "Okay."
21:22And again, this is this is this is also
21:24to to make sure that the information or
21:26like the adoption of AI is actually
21:28consistent across the org. And that's
21:29the main thing. I think that you see a
21:31lot of let's say PMs be super AI native,
21:33a lot of engineers be super AI native.
21:35You don't see the same across all the
21:37functions and potentially sometimes they
21:39go to market functions. And so, as a
21:41result, we really think hard about how
21:43do we deliver that to you in the form of
21:46something that you can look at on a
21:48daily basis and really be integrated
21:50with
21:51your workflow. And the last thing I'll
21:53share at Laurel is we think a lot about
21:56how do we surface it even more just in
21:57time. And what we're able to do in terms
22:00of our product is we're able to detect
22:02what it is you're working on when.
22:04>> Okay, so I think I get it, right? The
22:06thing that you are encoding that's most
22:08important is not the scheduled tasks or
22:11this particular interface in Dust, it is
22:13the actual skills and you are enabling
22:16the least AI proficient people at your
22:18company to operate at a similar level to
22:21those AI native people. What is the
22:24right company culture? How do you really
22:26get people to take advantage of a
22:27company OS like this?
22:29>> Yeah, absolutely. I think it really
22:31starts with culture. I just have a few
22:33photos from our offsite
22:36about 3 months ago. And
22:40it's really important to for it to start
22:42from the top, from leadership to say,
22:44"This is so important to us. It is not
22:45just an engineering thing. It is a
22:47cross-company thing." And what we did at
22:50this offsite is we did a company-wide
22:53hackathon. And I do know of a lot of
22:54companies that do this on a regular
22:56basis. How do we do a company-wide
22:58hackathon
22:59every quarter, every 6 weeks, right? Or
23:03how do we even get the just the
23:04go-to-market teams to do a company do a
23:06hackathon and show what it is that
23:08they're building. So, that the
23:09expectation that, you know, everyone is
23:11a builder is is true everywhere in the
23:13company, not just in engineering. So,
23:16with this, um what we did is we did two
23:18things. One, we did um training. And so,
23:21what we did is we actually did a lot of
23:23training around like how do you actually
23:24ship to production, even if you're not
23:26technical. Uh so, we created this
23:29um enablement guide for how to ship
23:31features with Devon.
23:32And so, you know, Devon essentially is
23:34like an agentic engineer. Um you can
23:36give it tasks. It It It started off, I
23:39would say, a year ago, 2 years ago, when
23:41we first started using this as almost
23:43like intern-level engineer. And today, I
23:45think it it's actually, you know, a
23:46decent software engineer. It's not a
23:48staff-level software engineer, but it
23:49does a lot of things. And as a result,
23:52you know, um my team is able to ship.
23:54And I'll just give go through a couple
23:56examples.
23:57Here is um a feature, an end-to-end
24:00feature, which includes front end
24:02changes and back end changes, where um
24:04you know, we enable people to
24:06to delete temporary initiatives. So,
24:08when you're keeping your time, sometimes
24:10you don't know um what matter or what
24:12project you're working on yet, but you
24:14know that you're doing some amount of
24:15work that should be grouped together and
24:16submitted at the end of the day. And so,
24:18that's where temporary initiatives is
24:19really powerful. Now, that again is a
24:21front end and back end feature. It is
24:23not just a front end um like almost like
24:26cosmetic change. It's actually pretty
24:28deeply rooted in how does it interact
24:31with PMSs and other systems and when
24:34does it release versus not? There
24:35There's a lot of complexity in something
24:36like um temporary initiatives. And so,
24:39this, by the way, you know, if you look
24:40at the person actually um
24:43knocking down those tickets and
24:44committing these PRs, this is actually a
24:46PM on my team. And I'll just go to their
24:48LinkedIn briefly. Um Nick, who's
24:51awesome, has been at Laurel for some
24:52time. If I go back to his educational
24:55history, right? Like um we we didn't
24:57grow up. Many of us didn't grow up as
24:59engineers. And yet Nick, um,
25:03I would say he probably identifies
25:04self-identifies more on the design side
25:06than on the engineering side, is able to
25:08take this feature end-to-end, which I
25:11think is just so cool. Um, similar to
25:14similarly, um,
25:15within,
25:17you know, many parts of our product,
25:18I'll just go through another example
25:19here.
25:21This is, um, the empty state for when
25:23someone comes in. So, really think about
25:25new user onboarding. What is it that
25:27they see? How do How do we make that
25:29experience super delightful? All of this
25:31is done by, um, by Jessica, who is,
25:34again, a PM on my team, not an engineer,
25:37and also not a PM who necessarily
25:39started their career in, um, in
25:42engineering or studied computer science.
25:45And so, I think this is just such a
25:46great example of people being able to
25:48ship even when they're not technical.
25:50And maybe the last thing I'll show you,
25:52cuz I think this is
25:53even cooler, is, uh, this little picture
25:56here, um, which is this is someone on
25:58our customer success team. Ashley's
26:00amazing. She deeply understands our
26:02customers and their needs. And by
26:04working with the PMs on the team to
26:06really create this enablement guide for
26:08Devon, they worked on this together.
26:10So, that, again, if if you are even less
26:12technical than a PM, right? If you're on
26:14the success team, how might you use this
26:17guide to be able to really ship the way,
26:20um,
26:21you know, in a safe way, in in a
26:24reliable way. And then all of these
26:25pieces we then broke down to say, "Well,
26:27should we start building skill files,
26:29you know, agents to help you?" So, that
26:31when you're trying to do this thing that
26:33typically is a playbook. And again, this
26:34is not 55 pages, but it's still eight
26:36pages.
26:37You are able to get the help and the
26:39support you need. And so, that is
26:41really, again, the the crux of it all is
26:43understanding what is the work that
26:45you're doing, how do you start to
26:46document that down, and then really
26:49clearly define these are the parts that
26:50remain human-centric versus these are
26:53the parts that should be automated away.
26:57I'll pause there, but I think it's also
26:58really cool to look at this ontology,
27:00which is essentially, you know, for
27:01every single function in the company,
27:03what are all the buckets of work that
27:05they're doing? And And what we do is
27:07actually we we actually spend cycles
27:09saying, "You know what? We believe that,
27:11like I said earlier, a product person
27:13should be operating like an engineer."
27:15So, all of the the things that we expect
27:17an engineer to do, we expect them to be
27:20doing feature work. We expect them to be
27:22testing. You know, we're we expect them
27:24to actually like crank through the
27:25backlog. The exact same things show up
27:28in what we want PMs to do. Um it is it
27:31is not a um
27:33an error where, you know, here in the
27:35ontology,
27:36we really have things like we want you
27:38to be, you know, doing feature work with
27:39agents. We want you to be um, you know,
27:42actually QAing your your product and
27:44fixing the bugs, not just like QAing in
27:46in the ways that people were doing
27:47before.
27:48And what we don't want you to be doing
27:50is things that were really tedious, like
27:52synthesizing competitive market, you
27:54know, intelligence, um actually writing
27:56these like detailed briefs, doing
27:57research planning, doing reach out for
28:00the the research, synthesizing the
28:02research. Like all of that, um you know,
28:04competitive analysis is a great example.
28:06It should You should be spending time
28:08building the agent to pull the
28:10competitive data, and you should just be
28:11moderating it, but you shouldn't
28:13actually be doing the deep work every
28:15single day. And like set up the system
28:16instead. And so, when we actually create
28:19this ontology, we're able to say, "Well,
28:21we want these numbers to go up. We want
28:24everything in green, the time spent
28:26doing that to go up. I want to see Nick
28:29doing this. I want to see Jess, you
28:31know, shipping this this this feature
28:33end-to-end. But what I want you to stop
28:36doing is I want you to stop doing these
28:38things that are really tedious, or the
28:40very least be calling an agent every
28:42single time that you want to do that."
28:44And And then
28:45again, going back to how do we make that
28:48true? By building the skill files, by
28:51building the agent agentic workflows
28:53where necessary, and making sure that
28:55we're surfacing for surfacing them where
28:57people work. And that's ultimately the
28:59the key pieces of the system.
29:01>> Wow, there is so much gold buried
29:05in the various parts of your answer
29:07there. The first part I want to double
29:09click on first is PMs shipping to
29:11production. Okay, people have heard
29:13about that. But PMs not just shipping,
29:16okay, here's this little growth
29:18experiment where we change the text in a
29:20button, which is a front-end only
29:22change, but a front-end plus back-end
29:25core feature, this temporary initiatives
29:27feature for instance that we looked at.
29:29That's crazy. So, talk to me a little
29:31bit about
29:33what is the scope of what PMs do ship to
29:36production, and how should people be
29:38thinking about in an AI-native
29:39organization,
29:41this is the role of a PM today.
29:44>> We talk a lot about this in terms of
29:47what is engineering anyways, what is
29:48product anyways, what is design anyways,
29:50and we really landed on this concept of
29:53um
29:54we want there always to be a captain of
29:57any given initiative, and the captain is
30:00the person where that skill set is the
30:03most important. And so, there are lots
30:06of features, um let's say we need to
30:09overhaul
30:10a system in order to make it much easier
30:13for, let's say, PMs to ship agents to
30:15work in that codebase. Usually, the
30:17captain is an engineering captain
30:18because that's an architectural change.
30:20If we have a feature where um like the
30:24interaction is really king, you know,
30:26we're doing this really cool stuff on
30:27mobile to make it so easy and delightful
30:29to kind of like look at how you spend
30:31your time in a given day and get
30:33insights from that.
30:34>> I hope you're enjoying today's episode.
30:36Are you interested in becoming an AI
30:38product manager making hundreds of
30:40thousands of dollars more joining OpenAI
30:42and Anthropic, then you might want to do
30:44a course that I've taken myself, the
30:46AIPM certificate ran by OpenAI product
30:49leader Mickdad Jaffer. If you use my
30:51code and my link, you get a special
30:54discount on this course. It is a course
30:56that I highly recommend. We have done a
30:58lot of collaborations together on things
31:00like AI product strategy. So, check out
31:02our newsletter articles if you want to
31:04see the quality of the type of thinking
31:06you'll get. One of my frequent
31:07collaborators, Pavel Hearn, is the build
31:10labs leader, so you're going to live
31:11build an AI product with Pavel's
31:13feedback if you take this AIPM
31:15certificate. So, be sure to check that
31:17out. Be sure to use my code and my link
31:19in order to get a special discount. And
31:20now back into today's episode. I used to
31:22think I had a retention problem. Turns
31:24out I had a messaging problem. I was
31:26sending the same onboarding emails to
31:28every new user, whether they activated
31:30on day one or never logged in again. I
31:32had no [music] idea who was slipping or
31:33why. Customer.io changed that. Every
31:35message I send is now based [music] on
31:37what users actually do in the product.
31:40Someone hits a key activation moment,
31:41they get nudged to the next one. Someone
31:43goes quiet, [music] they get a different
31:44path entirely. Their AI agent makes it
31:46fast. I describe the campaign I [music]
31:48want and it builds the full journey for
31:49me. Triggers, timing, copy, even
31:52branching logic. And when I want to know
31:54how something is performing, [music] I
31:55just ask the agent directly and it tells
31:57me what to do next. They also have an
31:59MCP server, [music] which means AI tools
32:01like Claude can see directly what's
32:03happening in your Customer.io workspace.
32:05[music] Your segments, your customer
32:06data, your attribution, all of it. So,
32:08instead of explaining your business
32:10[music] context every time you need
32:11help, Claude already knows it. Notion
32:13used Customer.io to personalize their
32:15onboarding and hit nearly 50% open rate,
32:19improved conversion by 6 to 7% with
32:21localized campaigns, and pushed open
32:23rates up another 20% through AB testing.
32:26[music] The idea is simple. Customer.io
32:28helps you deliver more impact from every
32:30message you send. If you're a PM or
32:32founder and your onboarding is still one
32:33size fits all, try Customer.io at
32:35Customer.io. That is more so than
32:38anything, it's it's a data problem. So,
32:40we have, you know, data science really
32:41plugged in there. But, really um
32:44the interaction is the most important
32:46thing to really sweat and make sure it's
32:48delightful. And as a result, a designer
32:50is a captain of that work stream. And
32:52then something like what I just showed
32:53you, something like temporary
32:55initiatives, something like the empty
32:57states,
32:58really having deep customer
33:00understanding, but also business context
33:03is really important. How do I know what
33:05people want to do with temporary
33:06initiatives? How do I know what the user
33:08wants to do, but also how do I know what
33:10the firm really wants to get out of it
33:13and or not? And so, what we spend a lot
33:15of time now thinking about is what is
33:17the what is the most critical piece to
33:20nail for the outcome that we're looking
33:23for and therefore the feature that we're
33:25building? And as a result, how do we
33:27appoint a captain that is skilled in
33:30that particular area? And so, that
33:32that's generally how we think about um
33:34the model evolving. And so, going back
33:36to a feature that might touch the front
33:38end and the back end, if we believe that
33:41the back end is in a good enough spot,
33:43and by the way, you can ask GitHub uh
33:45sorry, Devin um or even, you know, you
33:47like anything that's connected to your
33:48GitHub account to to look at the code
33:50and say, you know, in what state is
33:52this, right? And and it actually gives
33:53you a pretty good answer. Hey, you know,
33:55this this is this is what I would be
33:57careful of. Um uh and then you can
33:59actually pull in engineering um to to on
34:02the parts where you're like, this is
34:03probably the most contentious or this is
34:05where it gets the most risky. And again,
34:07you don't do this by yourself because
34:08you happen to be the most technical
34:10person. We're not. Um you do this
34:12through the help of asking, you know,
34:14Claude code to look at your code base,
34:16cursor, what I mean, whatever tool of
34:18choice you choose to use, you can ask it
34:21to really give you answers the same way
34:23that a marketer would say, "Look, I'm
34:25giving you some copy. Now, battle test
34:27this and go back and forth." It's the
34:28same concept. And and then going back to
34:31if you are clear on again, what is the
34:33hardest thing to to get right in a
34:35particular feature? Um for example,
34:38empty state. The empty state that we're
34:40working on here, it's the hardest part
34:42to get right is definitely not the
34:43engineering. The hardest part to get
34:44right is not even the design, it's the
34:45content. And again, the content has to
34:47do with the user and the business and
34:49the firm, and that is a very classic PM
34:52thing. And so it makes sense for the PM
34:54to be the captain of that. And so that's
34:56really the model we think about.
34:58Captains, you know, using um um you
35:01know, LLMs essentially like ask how hard
35:04something can be. Obviously, we still,
35:07you know, have code review, um and we
35:09make sure that um engineers are code
35:11reviewing the things that are risky. Um
35:13and so all of those pieces together
35:15makes it so that we can all ship,
35:17including, you know, customer success,
35:19which is again really wild. And and like
35:21go-to-market sales.
35:22>> And I think we can all immediately see
35:24how that allows engineers to work on the
35:28highest leverage back-end tasks, PMs to
35:31work on higher leverage features if CSM
35:32and go-to-market are enabled. What is
35:35the
35:36right set of checks and balances you
35:38need to put in place in your
35:39organization? How do you You mentioned
35:42code reviews. Where Where do those come
35:44in? How do you
35:45CSMs or uh
35:48go-to-market, for instance, make sure
35:50that what they're building isn't in
35:51conflict with something the product team
35:52is building over here, in contact
35:54conflict with somebody else's metrics?
35:56Usually, that's where the PM came in and
35:58did a lot of the glue work. How do you
35:59handle that in this new way of working?
36:01>> Yeah, that's a great question. Um so we
36:05again, I believe in the power of humans.
36:07So something as simple as, you know,
36:08creating a channel like ask Devin
36:11reviewers, and being able to go through
36:13here, um and making sure that there's
36:14visibility around all of the Devin um
36:18all the ways we're using Devin to ship,
36:19and then tagging in the right person,
36:21tagging in, you know, um a front-end
36:23engineer to really look at something,
36:24tagging a designer to look at something
36:26else, really going through and um making
36:29it visible. I think the first advice I'd
36:31give is transparency is everything.
36:34Um the second piece of advice is you do
36:37need to set some ground rules, right? So
36:39again, going back to our enablement
36:40guide, we've set some ground rules here
36:43as part of even the way Devon works. We
36:46often we
36:47we actually used to
36:50do this quick check where whenever
36:52someone, let's say someone on support
36:55had an idea.
36:56They essentially could go into this
36:58channel and post their idea and get a
37:00really quick check on
37:03is this something that
37:05makes sense? And again, I'll just let me
37:07zoom in here. Like I'm proposing a
37:08change to this experience.
37:11Um getting some some feedback, right?
37:14And and being able to say, "Hey, play
37:15around with the first version of it."
37:18And and getting you know, people to
37:19chime in and say, "Hey, this makes
37:21sense. This doesn't make sense. I'm I'm
37:23on the engineering team and let me give
37:24you some feedback. I'm on the success
37:25team. Let me give you some feedback."
37:27What you're really doing is you're
37:28taking what used to be a product review
37:30that used to take time to schedule and
37:32time to get all the stakeholders in the
37:34same room and you're just compressing
37:36it.
37:36>> Double click on product reviews for me.
37:38You guys have a really interesting
37:39process for when you do and don't do
37:41product reviews. What is
37:43the right balance so that you enable
37:45people to move fast but you're building
37:47the right level of collaboration on
37:49bigger features?
37:50>> Yeah, the same way we have this
37:52captain's model, I think about a
37:54framework what we call two tracks. So
37:56there's one track which is much smaller.
37:57If you have something that
38:00even some of the features I just showed
38:01you, like they're they're small enough
38:02where again, a PM, um somebody, a
38:06product captain or a product builder,
38:08right, can take it end-to-end. Those
38:10don't go through the same degree of
38:13rigorous review but they do go through
38:15things like that ask Devon channel. They
38:17go through things, you know, like a like
38:19someone looking at the PR, making sure
38:21things are good. Um, you, by the way,
38:23you are responsible for end-to-end
38:24testing of your features. I think that's
38:25actually really positive. The number of
38:27times where in a waterfall model, you
38:29would PM throws over to the designer and
38:31the designer throws over to the engineer
38:32and then engineer throws back to the
38:33designer, do design, QA, and the
38:35designer's like, this is not what
38:36[laughter] I designed. Um, that is just
38:38I think it's just such a It's like it's
38:40almost a meme cuz it happens so often.
38:43And so, I think that's actually really
38:43empowering to say, I am the end-to-end
38:47product builder and I take something
38:49from beginning to end and I own and I'm
38:51responsible for the quality and impact
38:54of this thing. And so, first of all, I
38:55just think that's a much more empowered
38:56way to work. Um, so, but but then going
38:59back to the two tracks, you have things
39:01that, you know, can really take the
39:02product life cycle and compress it down
39:04to a day, an hour, you know, like and
39:07that's how you get the velocity. But,
39:09there are some things where you're like,
39:10look, I think that the way that this
39:12product is going to behave, what I'm
39:14suggesting is a change, the feature that
39:16I want to do, it requires more much more
39:19alignment. So, a great example is, um,
39:22within Laurel, if you're going to change
39:23the complete way that activities are
39:25displayed,
39:27that's a that's a pretty radical change.
39:29Um, and how might someone a user go zoom
39:31in and out of their day?
39:33That is not a small thing. It require It
39:35touches, um, it's the whole like user
39:38interaction. And as a result, we say,
39:40look, we do want to do a product review
39:41for that. Want to make sure that, um, we
39:43talk about, well, how do we think about
39:45the entire product as a system so that
39:47we're not adding some random thing over
39:49there and a random thing over there. Um,
39:51but a lot of
39:53I think the first step is to actually
39:54even say, what is in what bucket? So
39:56that the things that could be running
39:58really fast are, but also, I really
40:00don't believe in this. I think a lot of
40:02quote-unquote AI-native companies are
40:03just like, roadmaps are gone, planning
40:05is a gone, everything is gone. Um, and
40:07what I say is, well, if everyone's
40:09running in different directions, even if
40:10you're running incredibly fast, you're
40:12not really going to get anywhere. And I
40:14see a lot of, um, great like local
40:17maximizations, but sometimes it's really
40:19hard to get to the global max, you know,
40:21a whole new set function change in your
40:23product in your market positioning
40:25without real rigorous thought around
40:28what is our strategy, what is our plan,
40:30why are we differentiated? And those are
40:32the things that require much more of
40:34what I call a true product review
40:36process, where to me it's more like
40:38product strategy review, and then
40:40there's architectural review, right?
40:42Making sure that the system actually
40:44will support all the changes that you
40:45want and that you can get to a next
40:47level of running fast.
40:49>> Mhm.
40:50So, did temporary initiatives go through
40:51a product review?
40:52>> It did not.
40:53>> Wow. Okay. So, what would be the like
40:56the right aperture? What have been some
40:58of your recent product strategy reviews?
41:00>> Yeah, so um
41:02today Laurel is beloved in a lot of
41:05firms that think about billable hours.
41:08And we're starting to find that there
41:09are a lot of firms, even if they don't
41:11have billable hours, um they really
41:12think they still need to think about the
41:13concept of time. Um I would say it even
41:16applies to tech. I think about the
41:17concept of time all the time. What are
41:19my PMs doing? Going back to this
41:21ontology um and and this work map of
41:24every single function. I mean, all of us
41:25should be thinking about the concept of
41:27time. What should sales people be doing
41:29today versus what should not be human
41:32anymore.
41:33And this is I want to be very clear,
41:34this is not a therefore we do not hire
41:36humans. It is a put the humans on the
41:38most important things. And I'll give you
41:41some great examples in here.
41:43Relationship building. You just will
41:45never replace a real check-in, a real
41:48moment of you know, true hospitality and
41:51delight, an actual on-site, taking a
41:53champion out to dinner. That cannot be
41:55replaced by agents.
41:57But what will make it so much easier to
42:00operate and and no one actually wants to
42:02do these things. What if the scheduling
42:04for the on-site and making sure that all
42:06of the back and forth and logistics is
42:07taken care of. You know, again, in
42:09marketing we do a lot of events. What if
42:10all the logistics of event planning were
42:12gone? Even this idea of unreasonable
42:14hospitality, and I I think this is such
42:16a a great example. It is such a core
42:19value of us of ours here at Laurel,
42:22where we really want to delight our
42:25customers all the time. We want to
42:26delight each other, want to delight our
42:27customers, and so we really have
42:29codified unreasonable hospitality as as
42:31almost like a like a cultural principle
42:33that we have, a company value.
42:35A lot of companies do this, by the way.
42:36They're like, this is a cultural company
42:38value, and then it's in a doc somewhere.
42:41People read it and then they forget
42:43about it. And what we do instead is we
42:44say, "Well, what does it actually mean?"
42:46We actually want to make sure that no
42:48matter who you are, even if you're the
42:51most thoughtful person in the world, or
42:52you're not the most thoughtful person in
42:53the world, even if you're 4 years into
42:55your time at Laurel, or you're 4 days
42:57into your time at Laurel, you understand
42:59that unreasonable hospitality is a
43:01requirement of how we operate. And
43:03especially if you're on the customer
43:05success team, we expect that you do this
43:08with our customers. How do we
43:09systematize that? And and and that's a
43:12real real question. You know, again,
43:14there are people on our team who
43:16just from who they are as humans,
43:18they're the kind of people who's like,
43:19"Someone told me that they're going to
43:20Mexico." And so, and it's the first
43:22time, by the way, that they're traveling
43:23outside of the country. And so, I bought
43:25them an engraved passport holder.
43:28That is, by the way, a lot of people on
43:30the Laurel team, but if I were to scale
43:32that to like a lot a lot of people and
43:34make sure that everyone's doing it every
43:36point in time, even when they're really
43:37busy with other stuff, it's pretty
43:39unlikely that's going to happen.
43:41Instead, we say, "Well, we actually want
43:43to make sure that unreasonable
43:44hospitality is a check that we put in."
43:46And so, again, going back to the OS I
43:47was showing you, "Hey, if you have a
43:49check-in with someone,
43:51and you know, you haven't actually done
43:53anything like this in a while, you
43:55haven't had an in-person touchpoint, how
43:57might you surprise and delight them?"
43:59And here's some ideas that we've already
44:00pulled for you. We pulled from your Gong
44:03transcripts that these are things that
44:04they love. And we pulled um, from the
44:07fact that they love these things,
44:08instead of making you do all the work of
44:10figuring out, is it a passport holder?
44:12How do I even get a passport holder
44:13engraved? We are going to we're going to
44:15systematize that. And so, that's this
44:17real idea of like deeply understanding
44:19your your your company's work, your
44:22team's work. What are the things that
44:25makes you special? Where do you put the
44:27humans on the things that make you
44:29special? And then, where do you even in
44:30those moments, like unreasonable
44:32hospitality, make it so it's easier to
44:34do that job and to deliver that
44:36particular feeling.
44:37>> So, you This isn't your first rodeo.
44:40You've been in product for a really long
44:44time. If we rewind back to some of those
44:46experiences, those formative
44:47experiences, let's say like Airbnb in
44:502015
44:51or Dropbox in 2013 or WeWork in 2019,
44:55you've been in these large organizations
44:57that most people listening to this
44:59podcast have been in where
45:01the PM traditionally never had access to
45:04get
45:05let alone like the amount we're showing
45:07here where they have a dev and agent
45:08that is shipping front end and back end
45:10features.
45:12And you'd be surprised, even at
45:13companies like Adobe, PMs are still
45:15living in that world that you and I were
45:17in back then. They still don't have
45:19access. They're looking at what we've
45:21just showed them and they're saying,
45:23"Gosh, this is too far away from my
45:25reality.
45:26Is it true that like this just won't
45:28work in certain type of companies or
45:31will they eventually get there?
45:33It is just a matter of time."
45:34>> I'll start with the end, which is I do
45:36think it is a matter of time that every
45:37company is going to have to get there.
45:40You can't keep doing the same thing if
45:42everyone else, including all your
45:43competitors, are moving at 10 times the
45:46speed. So, I do think that there will be
45:48pressure to ultimately get there for
45:50everyone. Now, what you want to be for
45:53the company and for the individual is
45:55you want to be, um, as far, uh, you
45:58know, as as advanced as it's
46:00right, in that curve as opposed to just
46:02waiting for it to happen to you. And
46:04this is where I go back to, you know,
46:06the first step is just start small.
46:08Start with one workflow that you are
46:11doing. Um and or I really I really push
46:14on this. I think this is really
46:16uh a great way to to get your feet wet
46:18and and start to think about this. Go
46:19find another team in your company
46:21somewhere. And even if you're like, "I
46:23I'm not quite ready to ship to
46:24production for whatever reason." And
46:26usually the reasons are not that you
46:28can't like you're not physically able
46:30to. It's usually something about the the
46:33system or the process that it's not
46:35quite there yet. But but let's just say
46:37that you you don't feel like you can in
46:38the next month. Go somewhere else where
46:42there's always somewhere in the org that
46:43is hungry for product thinking and
46:46hungry for a tool to make their life
46:48better.
46:49And I would start with let's just go
46:51build a tool for somebody in a different
46:54org to make their life better. And
46:56simultaneously pick up one part of your
46:58workflow that is taking you a lot of
47:00time and there's really no reason that
47:02you should be doing that. Again, great
47:04examples. I'm getting into a customer
47:05call.
47:06I would love to be pre I would love to
47:08be prepped for that in a way that, you
47:10know, an agent is serving me that
47:12information as opposed to me having to
47:14pull from multiple different sources,
47:15right? That's a very simple example. I
47:17write the same um
47:19same email over and over again. It
47:21should be auto-populated. Again, these
47:23are just small small little automations.
47:25Or you can call them templates. Whatever
47:27it is that you that um makes sense to
47:29you. Start there.
47:31And then I would say if you're ready to
47:33take on something bigger, this idea of
47:35like, "What does a function do? Or what
47:37is an end-to-end
47:39um operational journey look like?" And
47:43there I would start to say map out your
47:45ontology or take your playbook and
47:48really, you know, write that down. And
47:50again, what I what I find really fun is
47:53like these playbooks, I think if someone
47:55was tasked to write a playbook back in
47:56the day, they'd be like, okay, I'll do
47:58it. It'll take me a couple weeks. These
48:00playbooks can be written in an hour.
48:03Actually, the first draft can be written
48:04in sub a minute, but to make it actually
48:07right and and you know, really
48:09reflective of your business, yeah, it
48:10will take a little bit more time, but
48:12we're talking hours here, maybe days
48:14max. We're not talking weeks. And so, I
48:16think when you get a feel for how much
48:19you can enable yourself and you can
48:21enable others, you're going to create a
48:23culture, again, even just going back to
48:24culture change. You're going to create a
48:26culture where um
48:29it's celebrated and it's fun. And and
48:31again, if you're leadership, what I'd
48:32really encourage you to do is make that
48:34the culture. Celebrate those wins. Take
48:37the people who are your 1% and take
48:40their workflows and figure out how to
48:42scale that workflow to every single
48:44person on the team. When you create that
48:46um expectation and you celebrate those
48:49wins, you'll get more and more of that
48:50behavior.
48:51>> For you guys, did it happen as a
48:53transformation? Were you guys always
48:55this way? Did it start with the CEO and
48:57the founder? How did it come about so
48:59that now you guys do feel the confidence
49:01that you have this enablement playbook
49:03of Devin where anyone can ship to
49:04production?
49:05>> Yeah.
49:06Um I think there were a lot of pieces,
49:08but I'll highlight the pieces that I
49:09think are most relevant that someone
49:11listening to this could take and and
49:12replicate. Um the first piece I've
49:15already shared, which is the idea of
49:16just doing a hackathon. And in the
49:18hackathon um making everyone participate
49:21cuz it changes this idea of you have to
49:23be technical to build something. But and
49:25and and again, I think most people have
49:27done that. Um so, I would expect that,
49:30you know, 90% of the people listening
49:31have participated in some kind of
49:33hackathon. If you have not, that's the
49:35first step. Um the second step is to
49:38really think about all the different
49:41ways you can again, automate the
49:43workflow. I think that a structural
49:45thing that I would really recommend is
49:48actually making this idea of playing
49:52with AI tooling, creating workflows,
49:55automating, you know, large swaths of
49:57somebody's day in a way that makes them
49:59much more productive, make that the
50:02actual charter and mandate of a full of
50:04a person full-time. And what I really
50:06find is a lot of times when you say it's
50:08everyone's responsibility, it's no one's
50:10responsibility. And so what we have at
50:12Laurel is we actually have an AI
50:14operations team. And to me AI ops is a
50:17new biz ops. Before biz ops, they were
50:19doing I'm really meaningful things, but
50:22often it was it was very
50:24high-level, all the different hats, a
50:26lot of like market level stuff. Now if
50:29you repurpose this idea of having biz
50:30ops, which is really again a Swiss Army
50:32knife in many ways, to finding people
50:36who are insanely curious, tinkering with
50:38the latest technology, and relentless
50:41about finding efficiencies, that's the
50:43DNA I really look for. And so we've
50:45actually built out an AI operations
50:47team. We started with Sasha,
50:50who has built out a lot of the things
50:52that I've I've shown today.
50:54And what he did is basically was like,
50:56I'm going to demonstrate value in having
50:58AI operations. And very soon when you
51:00when you have one person who's doing an
51:02excellent job, every single other
51:04function is like, I want my Sasha.
51:06I want my own AI Sasha. And that is how
51:08you then get the buy-in to say, okay,
51:10well maybe we have an AI person AI
51:13operations person just doing
51:14go-to-market, and a separate AI
51:16operations person just doing product,
51:18and a separate AI operations person just
51:20doing finance. Because all of these
51:22functions, by the way, finance, rev ops,
51:25product ops, research ops, you name it,
51:28all of them are changing so
51:29dramatically. And so being able to
51:31retool your the way that your company
51:34works with someone who's really
51:36dedicated to pushing that forward really
51:38really accelerates the the journey.
51:40>> Mhm. That's really interesting. And you
51:42guys were founded before the AI
51:44revolution. So
51:46I guess like for other companies that
51:48were founded before then, I think you
51:49were 2018, like where
51:53who is like the right driver? It feels
51:55to me like it probably has to start like
51:56literally with the CEO, right?
51:58>> Yeah, I was going to say I want to give
51:59a ton of credit to Ryan. So, yes, um
52:02Laurel um was founded actually I would
52:04say Time by Ping, this is what Laurel
52:06was called uh previously, was founded in
52:072018. Um but Ryan actually had the the
52:11foresight and and the the courage really
52:14to say, you know what? When I think
52:15about what time looks like in a world of
52:19AI and LLMs, it's very different. And
52:21when I think about um at the time our
52:23core product um timekeeping, right? Like
52:26what does timekeeping look like in a
52:27world that um where you have to kind of
52:30enter it manually or just do it through
52:32call it what I call integrations um
52:34versus a world where you can actually
52:36start to really see everything that's
52:38happening on your computer and
52:39synthesize that and run that through an
52:41LLM. Like he basically had the foresight
52:43and again courage to say, I'm going to
52:46re-architect my entire product, my
52:49entire company to be AI native. And so
52:52it's it's really interesting. Like I
52:54really believe that um and I experience
52:56this day to day. Um I wouldn't you know,
52:58like I I I was like I I I want to be
53:00building at the cutting edge. Um
53:03Laurel is AI native.
53:05Although it was founded like more than 3
53:08years ago. Um and so that it does start
53:10with the CEO. Um but even if it if uh
53:14people don't have that degree of change
53:17um
53:17and and conviction, I think you can
53:19still do it at every single level where,
53:21you know, if you are not the CEO, but
53:23you're uh an executive, you can say,
53:25well, this is how I expect my function
53:27to really operate. Here in my function,
53:30I am a marketing leader. I fully expect
53:32that this is what we are doing and let
53:35me go color code everything in here that
53:38should be AI enabled, right? Like when
53:39you think about copywriting today, you
53:41should not be writing copy by hand. You
53:43should be editing when you're doing
53:44videos. Like if you're not using um a
53:47lot of the AI tooling out there, you're
53:49spending a lot of money on studio, on
53:51video in a way that you don't need to
53:53anymore. So, being able to go line by
53:55line in terms of your again your your
53:58work map. What is it that all my humans
54:01do and how do I really think about where
54:03do I need to keep that person versus
54:05where can I actually really AI charge
54:07supercharge them?
54:08>> Amazing. So, I think that's the key
54:11point for a lot of people that I talk to
54:13at least is that they have they don't
54:15have any access to the stuff we're
54:17showing and probably it needs to start
54:19like all the way at the CEO level and
54:22then it can work its way down where like
54:24you need really amazing CPO like
54:26yourself who is also AI filled in order
54:28to make this happen and that's kind of
54:30the next layer I want to talk about is
54:33as an AI filled CPO, what is your take
54:36on the types of product teams we're
54:38going to see in the future? What types
54:40of product managers are you hiring and
54:43what is the shape of their role today?
54:45>> I think for many people, um I'm sure
54:47this is dialogue that's happening
54:48everywhere. This idea of are you a
54:50product manager or you're just a product
54:52builder? And how many people are product
54:55builders? Meaning is it just the product
54:58person themselves by functional title or
55:01is it also the designer? Is it also the
55:03engineer? Um I'm a big believer of the
55:05fact that I think everyone should be a
55:07product builder. It goes back to my um
55:10uh how we operate the team today with
55:12captains and taking features end to end.
55:14Um what I do look for specifically in in
55:16product uh builders who are product
55:19managers by training, um I look for a
55:22couple of things. Um I found that uh if
55:25you're incredibly senior in the sense
55:27that you have the judgment, you've gone
55:29through the hellfire, you've shipped
55:31things that haven't worked and I think
55:33for all of us that have shipped things,
55:35most of the time it doesn't work in the
55:36first go around.
55:38If you have if you kind of have that
55:40battle-tested judgment, I'm finding that
55:42the combination of that experience plus
55:45this intense curiosity, this desire to
55:47be hands-on, I think you see
55:49a little bit of a bifurcation. There are
55:50a lot of people who are very experienced
55:52and almost scared that their job is
55:54changing and um they're feeling more
55:58fear than I would say excitement. And I
56:00would say that there's another group of
56:01people who are very experienced and
56:03they've been they've never been more
56:05excited. Like I I've never been more
56:07excited by the way to
56:08not be doing all these things I used to
56:10do in the past that took me forever that
56:12there was no part of me that wanted to
56:13be doing that. Instead, I love, you
56:16know, actually like shaping a product,
56:18really getting hands-on. Um and then so
56:22so being able to find those people who
56:24are excited, who are curious, but yet
56:26have the the judgment and the reps is
56:28really really important. And so again,
56:31um this is not necessarily by design,
56:32but what I found really interesting was
56:35you know, there are a number of people
56:36who previously were, you know, CPOs, VP
56:39of products, head of products. They've
56:40come in and they they're the ones
56:42building end-to-end. They're the ones
56:43shipping end-to-end. Um and again,
56:45they've never been more excited. They've
56:47never been more excited to not have a
56:51team to have to manage because they
56:52realize that a lot of that is just
56:54overhead. A lot of that is just coord-
56:56coordinate coordination cost. They've
56:58realized that a lot of it is just
57:00coordination cost and instead they can
57:02just be um enabled to get right in there
57:05and drive the change they want to see.
57:07>> That's crazy. So you have embraced the
57:09super senior ICPM. I think you said
57:12something pretty crazy actually when we
57:14were talking before which was
57:16the more senior you get, the longer
57:18you've been in product, the smaller your
57:20orgs have become. Is that the trend of
57:22the future, smaller and smaller product
57:24orgs?
57:25>> I think so. Yeah, I mean I've had
57:26hundreds of people and today I have five
57:29PMs and four designers and um there
57:33isn't a real reason to grow that um
57:37because again, like when you add more
57:38people, you add more coordination costs.
57:40You actually um
57:42it have a harder time making making
57:44people feel like they are absolutely
57:46responsible for taking something
57:47end-to-end. And so, I do think of that
57:50as the future. I think that um the best
57:52teams are going to be lean, but not so
57:56lean that they're starved. And so, it's
57:57really important to find that line.
57:59>> So, you said you do something pretty
58:01crazy in your interviews. Can you tell
58:03me how you interview people um and
58:05really find these gem AI pilled super
58:08senior ICPMs?
58:10>> I think a lot of people are talking
58:12about. Of course, you know, you do a
58:13session where people have to build with
58:14AI. Um I I think that's all fine and um
58:18uh I think it makes a lot of sense to do
58:20that. Um it it takes cycles, by the way,
58:22to even have a standardized interview
58:24loop. Um some companies it makes sense
58:26because they're large enough where
58:27they're hiring, you know, enough PMs.
58:29But again, I I do think many people are
58:31saying, "Hey, let's actually get a
58:32little bit more um particular about who
58:35we hire and make sure that they're
58:36really seasoned. And we'd rather pay a
58:38few really seasoned people, you know, a
58:40lot more than having just an army of
58:42people." And so, um what I've been
58:44doing, and I I do this, by the way, for
58:46every function, not just product or
58:47designer, so you know, so and so forth,
58:50um is I I do ask people to screen share.
58:52And what I found is it is so easy to
58:55say, "Hey, we are, you know, I'm AI
58:58pilled, we're AI pilled, we do a bunch
58:59of stuff with AI." But as soon as you
59:01get into um like if you really peek
59:03under the hood, you're like, "Actually,
59:05I think you're what I call like level
59:07one." And maybe I'll just take a moment
59:08and talk about the levels for me. Level
59:10one is you're talking to, you know,
59:13you're talking to ChatGPT, you're
59:14talking to Claude. You're really using
59:16um AI kind of in a chat mode. Almost
59:19like like search mode, right? Like I ask
59:21a question, you give me an answer. Level
59:23two is where you start to automate a
59:24workflow, right? And this is what I was
59:26showing earlier around just the first
59:29step is like start small. An OS does not
59:32start necessarily as an OS, but it
59:33starts with a first
59:35um automation.
59:36Right? A first little piece of workflow
59:39that everyone's going to start doing.
59:40And so that's level two. Um level three
59:43is when you start building, you know,
59:44apps. Right? You say, "Hey, you know,
59:46it's really important that I um I'm I'm
59:48doing this thing. It's really tedious.
59:50I'm going to build an app to make it
59:51like less tedious." And then level four
59:53I'd say is where you're actually
59:55building I call it shared apps and or um
59:57if you really think about the product
59:58life cycle, you're you're really
59:59shipping to your customers. And so those
1:00:01are the the maybe the four levels um
1:00:04that you can assess yourself on. You can
1:00:06assess a given company on like which of
1:00:08those four levels is the majority of the
1:00:10organization um
1:00:12operating at. And so um what I find is
1:00:15when you actually ask someone to screen
1:00:17share and show them how and show you how
1:00:19they AI, you're very you can very
1:00:22quickly get a sense of
1:00:24are you at level one? Are you basically
1:00:26just talking to chat GPT? Um or have you
1:00:28actually created um like a a some way to
1:00:31really like scale yourself, some kind of
1:00:33workflow, some kind of agent? Or are you
1:00:35starting to build like apps? And or, you
1:00:38know, what are you shipping? Like truly
1:00:39truly shipping? And so really getting to
1:00:41see that um live on screen is really
1:00:45really interesting because otherwise
1:00:46it's really easy just be like, "This is
1:00:48what I do." And it's you know, pulled
1:00:50from LinkedIn or pulled from the latest
1:00:51thing you saw on the internet. Um but
1:00:53actually peeling it back and be like,
1:00:54"What is on your screen?" is is really
1:00:57fascinating.
1:00:58>> Wow. People don't believe me when I keep
1:01:00saying this is the new interview. This
1:01:02is what I'm hearing. You've heard it
1:01:04from a CPO herself. So,
1:01:07a lot of people are feeling pretty bad
1:01:09about this whole transition. Like there
1:01:12there's a lot of FUD going around in the
1:01:14PM field. If you check out Reddit or
1:01:16something, people are feeling a lot very
1:01:18nervous about this change. They're
1:01:20saying, "Hey, we're compressing out the
1:01:21juniors."
1:01:23You had a really interesting take on
1:01:24this, which is that
1:01:26the best PMs are actually getting more
1:01:28roles and the rest are feeling fear and
1:01:30destruction. Can you unpack that for us?
1:01:32>> I think it's because one PM can do so
1:01:34much more than ever before, but there
1:01:36aren't that many of them who are that
1:01:39skilled, that have that judgment, who
1:01:41are AI pilled, um who fearlessly are
1:01:43going through all of these pieces and by
1:01:46the way know that one of the most
1:01:47important things forever and will never
1:01:49change about the PM role is that they
1:01:50have to stay close to their customers.
1:01:52Right? So like the the Venn diagram of
1:01:54all of those traits is not large in
1:01:58terms of the actual number of people
1:01:59that fall into that and that's what
1:02:01every company's going to want.
1:02:03And and around the edges it's like why
1:02:05do I why would I go hire someone who is
1:02:07not all of those things? I'm going to
1:02:08have to supplement them in some way and
1:02:11it's going to create overhead.
1:02:12And then when when in in many ways I can
1:02:14take that piece that is not excellent
1:02:16and I can build like, you know, again a
1:02:18workflow and agent around that. So I I
1:02:20think it's really finding who I call the
1:02:22orchestrators, right? The people who are
1:02:25big picture in terms of their thinking,
1:02:27but you know, down to the detail in
1:02:29terms of terms of their execution. Those
1:02:31are the people who are worth their
1:02:32weight in gold and I think that a lot of
1:02:34people who need to be complemented by a
1:02:37designer or complemented by an engineer
1:02:39or like you know, complemented by many
1:02:41many many other people, it just doesn't
1:02:42make sense anymore because why go hire
1:02:45all these people when again one person
1:02:47can be the end-end builder.
1:02:48>> It's not me saying it's guys, it's her.
1:02:51I've been preaching this for months and
1:02:52months.
1:02:54This is the future of product
1:02:56management. We just gave you
1:02:58the entire playbook. She just screen
1:03:00shared literally everything, the company
1:03:02OS, how their
1:03:04PMs are knocking down linear tickets. If
1:03:07you want a really amazing job, apply to
1:03:09Laurel. Julie is not just doing this
1:03:12though, right? You actually have so much
1:03:15cool stuff going on. You teach product
1:03:17management at Stanford.
1:03:20I think you had at some point have been
1:03:21involved with Reforge. Can you catch us
1:03:23up outside of Laurel? What's the world
1:03:25of JZ? What's going on?
1:03:28>> Um I do teach every year at Stanford. Um
1:03:30I do it for the love of um really just
1:03:34getting to meet the next generation of
1:03:36of builders. Um I also get the really
1:03:39awesome benefit of meeting people like
1:03:42the Sasha's of the world who, you know,
1:03:44once took my class, then TA'd for me,
1:03:46and now is at Laurel. Um and uh you
1:03:49know, teaching for me has been a
1:03:50combination of of passion and honestly
1:03:52uh of pipeline. Um so I teach at
1:03:54Stanford, um I teach at Yale, and I
1:03:56teach at Reforge. Um and it's it's just
1:03:59how I think that when you teach
1:04:00something, you have to know it like the
1:04:02back of your hand in order to actually
1:04:05share that with someone else. Um so I I
1:04:07again I just find this really funny. A
1:04:09lot of times I'll teach and then I'll be
1:04:10like, "Ah, good reminder, JZ. Like uh
1:04:13were you doing that today in your
1:04:15day-to-day? Uh were you customer-centric
1:04:17enough? Were you problem-space first and
1:04:19not solution first enough?" And so I
1:04:21just find it both um so gratifying
1:04:23personally, but also um such a great
1:04:26reminder of of what product really is.
1:04:28And I'll I'll say one last thing, which
1:04:29is
1:04:30what's funny is that
1:04:32um so I I I teach AI leadership through
1:04:35Reforge, and that curriculum changes
1:04:37literally by the month. Um you know, we
1:04:40teach it every 6 months, and the amount
1:04:41of change between the 6 months is
1:04:42massive. But when you actually teach the
1:04:45fundamentals, when you teach the what I
1:04:47call like PM 101, those core principles
1:04:50have not changed. You should still
1:04:52always never jump to the solution. And
1:04:54now that you can build faster than ever
1:04:56before, it doesn't mean you just build
1:04:57everything. Like what actually is
1:04:59important is to know why and for whom
1:05:02you're building for, and what is it that
1:05:03you're trying to solve for, and what
1:05:04success looks like, and therefore you
1:05:06actually know you've hit your target.
1:05:08And so what's really ironic is that
1:05:11through teaching all these different
1:05:12levels of of product people over the
1:05:14years, I find that the the fundamentals
1:05:17and the principles have never changed.
1:05:20In fact, they're even more important
1:05:22than ever before, but the tools and the
1:05:24way you operate and the way you
1:05:26can blast through the bureaucracy and
1:05:28feel empowered, that's radically
1:05:30changed. And so as a as a leader, the
1:05:32way you empower your team is very
1:05:34different. Do you have the right
1:05:37culture? Do you have the right team? Do
1:05:40you have the right space for people to
1:05:42even build? Do you have the right
1:05:44operating system? Do you have the right
1:05:46knowledge of what people are doing
1:05:48day-to-day? Do you have all of those
1:05:50pieces? That is changing dramatically,
1:05:52but in your actual, you know, one on one
1:05:55on one basics around what it is that a
1:05:57product person is supposed to be doing,
1:06:00um
1:06:01the speed has changed dramatically, but
1:06:03what you're supposed to be doing at the
1:06:04heart of it, that has not changed.
1:06:06>> What a way to end it. All right, guys,
1:06:08we have hit a crazy milestone when we
1:06:10crossed 40,000 YouTube subscribers. We
1:06:12have also crossed 565,000
1:06:15average views per listen per episode.
1:06:17When I started this podcast 2 years ago,
1:06:19I wouldn't have believed it. I want all
1:06:21565,000
1:06:22of you to flood Laurel's PM
1:06:25applications. For my money, this is like
1:06:27the coolest PM job you could possibly
1:06:30have. And I would say if you are in a PM
1:06:32job where everything we were just
1:06:34talking about feels really foreign and
1:06:36like 10 steps away from what you are,
1:06:39find a job like this with an AI PM CPO
1:06:43like JZ. You are going to learn so much
1:06:46more than if you get to this 4 years
1:06:48from now and then you learn it. Apply to
1:06:50Laurel, get her the best AI PMs in the
1:06:52world. Check out her class at Stanford
1:06:55if you are in the Bay Area so you can
1:06:57really learn AI PM. And if you are a
1:06:59leader, check out her course at Reforge.
1:07:01This is just me saying this. You can see
1:07:03how much value I got out of this
1:07:04episode. You can see I'll be writing
1:07:06about a company OS soon in my
1:07:08newsletter.
1:07:09Jazy has absolutely killed it. Thank you
1:07:11so much, Jazy.
1:07:12>> Thanks for having me.
1:07:13>> I hope you enjoyed that episode. If you
1:07:15could take a moment to double-check that
1:07:16you have followed on Apple and Spotify
1:07:18podcasts, subscribed on YouTube, left a
1:07:20rating or review on Apple or Spotify,
1:07:22and commented on YouTube, all these
1:07:25things will help the algorithm
1:07:26distribute the show to more and more
1:07:28people. As we distribute the show to
1:07:30more people, we can grow the show,
1:07:32improve the quality of the content and
1:07:33the production to get you better
1:07:35insights to stay ahead in your career.
1:07:37Finally, do check out my bundle at
1:07:40bundle.akashsharma.com
1:07:41to get access to nine AI products for an
1:07:44entire year for free. This includes
1:07:47Dovetail, Mobbin, Linear, Reforge,
1:07:49Build, Descript, and many other amazing
1:07:52tools that will help you as an AI
1:07:55product manager or builder succeed. I'll
1:07:57see you in the next episode.

About this transcript

This page contains the full transcript of How to Build a Company OS in Claude Code | Jiaona Zhang | Product Growth by Aakash Gupta, generated from the public captions YouTube serves with the video. The transcript has 13,745 words across 1,997 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.