The AI Distribution Shift, Navigating PMF Collapse & Building AI-Native EPD Systems w/ Brian Balfour #241

We're doing a special in-episode feature with our friends and sponsor, Span. People have been working for years on quantifying engineering productivity, and the reality is there's no one single metric that you can boil it down to. It's a far too complicated problem space to try to reduce it down to something simple. So, the way we talk about it internally is using multiple different signals for the instigation of inquiry.

Stay tuned for later in the episode, Stephen Paletto, field CTO at SPAN, deconstructs the tactics behind how top companies are navigating the AI transformation, navigating new bottlenecks and accelerating work. The best products mold their product to fit with the distribution channel, not vice versa. That's always the mistake. We have no control over the channels and where that we can tap into, where our customers are living. Those channels define their own rules.

We can't change those rules, but the thing that we can change is our product and how we mold to them. And I'm Patrick Gallagher, and we're your hosts. Our show shares the most critical perspectives, habits, and examples of great software engineering leaders to help evolve leadership in the tech industry.

In today's episode, we're joined by Brian Balfour, founder and CEO of Reforge, to deconstruct the two probably highest stakes questions facing technology leaders right now. What do I build and how do I build it? So our conversation with Brian today is about the strategy. Thank you. as platforms and how to emerge as a winner in the new competitive AI landscape.

Brian also shares with us why leaders need to avoid the local maxima trap in how your teams work together. We talk about why you need to siphon off skunk work teams to take high-risk AI native bets. Plus... how to rethink what to build, avoid commoditization, compress your product discovery from weeks to hours, scale feature variations and prototypes, and evolve products to solve harder and harder classes of problems and shift your specialist roles from inboxes to system.

and so many more insights around what a unified innovation engine looks like now in this era. So if you ever had any questions about how what's possible to build is changing or how do the team compositions and how our teams work together to build those things.

is changing. This is an incredible conversation. Enjoy this episode with Brian Balfour. So I was thinking about how to frame our conversation. So I know there's sort of two questions that are going to be driving what we talk about. And it's this big question that folks in our community want to answer, which is...

what do I build right now? And the other being, how do I build it? So how do I evolve engineering, product, and design to do that? But I guess maybe when I was thinking about how to frame this, I was going to start by sharing how much I value and respect your work over the last 10 years, how you've critically shaped my thinking and how I believe you're...

research and reforges research over the last number of years has elevated the industry. But I realized there's probably a more fun way to start with something that I probably believe that you value more than anything else, which would be like user research and insights.

This week was interesting. We hosted a webinar. We had about 100 engineering leaders join us and we ran a couple polls. And I think the insights really set up for an interesting conversation for how we answered these two questions. The first question was, where are you primarily deploying and using AI? And five to one, the answer was coding.

Second one was, what is your biggest challenge? And four to one, it was quality. And then there was a big debate about what do you mean by quality? So like it also spiraled. But I thought that was interesting because to me, it sort of signaled there's...

a huge focus on adoption which makes sense that's been a lot of what people have been focusing on and it also sort of signaled to me that it's primarily optimized within engineering like if you're primarily focusing on coding then maybe you're optimizing in engineering and then now people are starting to confront like the downstream bottlenecks of that

That's like the top of mind thing. Outside of this, though, there was a couple other conversations. I'm talking to a founder of a DevEx company, and their number one most requested feature was like, I want you to aggregate all of the AI tools being used in coding right now.

And they've made a conscious choice not to build that. They're like, no, that's like not giving you the most amount of signal. Like that doesn't give you the holistic look into what's going on. And it may only be a short term.

insight until people sort of build the behaviors and habits and using these different tools. And on the other side of this, we have the CTO of Atlassian that I'm talking to, and he's owning the OKR for adoption for everybody outside of engineering. That's like his core mandate.

And then I'm talking to the CTO at Brex and he was like, I'm becoming an IC for two months and my job is to figure out how do we disrupt Brex and think about it from an AI first, AI native principle and take a whole team and dedicate it to that.

So sort of a long setup to ask you the question of what's your reaction to some of these signals? And I guess as I kind of go through and maybe share a little bit where the community is at, like what's your take or reaction to where maybe folks are at right now? Yeah, I mean, people look at like when they've looked at AI and applied AI and like internal adoption, there has just been an extreme over focus on essentially.

adopting ai to produce more lines of code to write code faster the problem with that is that that's not the actual output that product teams epd is actually solving for I say that because it historically has always felt like that because that was like the constraint, right? And it makes sense why people started there as like one of their key focus items because it was the historical constraint.

However, the output of EPD is about producing product that our customers adopt, right? So there's two parts of that. It's actually producing product. And then there's the adoption of that product. And that's ultimately success.

To get that to work, that output is produced by this system of... typically product design ing and if you want to include adoption in that a little bit of like pmm and so when you look at it like that and when of course when you look at a system like you mentioned is like if you hyper focus on one constraint and you resolve that constraint um you actually just move the bottleneck to another part

the system, you actually don't improve the system. That's kind of like the common saying in systems thinking is that your system only goes as fast as the slowest part of the system. And that's what we're seeing. And all sorts of new narratives have popped up around this, like PM is the new bottleneck was one of those narratives that... a little bit viral a couple months ago on LinkedIn. And part of this also is that AI is a new technology. So it makes sense that...

the more technology driven people are kind of the first to adopt it. And so I think this is part of the trend that you see with like chief product technology officers, like this thing merging into one and so that they can drive. But the best people are driving adoption across engineering.

product and PM kind of holistically in an even way. And that's what's going to actually produce more of the output that they're ultimately seeking. The other symptom of this that I'm seeing is like, maybe this is a little controversial, but I'm seeing a lot of teams celebrate right now. They're like, oh my gosh, I got, you know, my non-technical designer or PM set up on Cloud Code. And now they're like pushing this change to production. And that to me is like...

they are doing it 100% wrong. They have the exact wrong mentality towards AI adoption. Because when you look at what those non-technical designers and those non-technical PMs are actually pushing, it's like really small changes. And I look at the small changes and I'm like,

actually, if you just invested in the right infrastructure, AI agents can do that now or will be able to do that in the very near future. So it's these teams that feel like they're creating leverage, but they're actually doing the exact opposite. They are blowing up their opportunity cost. and and so the opportunity cost here is right now they're like turning these non-technical designers into mediocre developers and that's stealing focus away from well actually how do i

create top tier designers in an AI age? How do I get them to adopt and do their responsibility, their part of the system, you know, at a higher and higher level? inside ai adoption but instead they're focused on how do i get these people to do low-level tasks in a totally different function it doesn't add up to me when you think about it on on this like systems level and once again there's this hyper focus of like

pushing code to production. And that's kind of led to a bunch of different mistakes. But I don't think people have the big picture on the opportunity cost. I think people are finally coming around to it. But, you know, the superpower of AI has been writing code. So it's like, well, let's just get everybody to write code. And it's like, no, no, no, no. Like, no, it's like, OK, well, how can I do things in the product management function that I wasn't?

able to do before that's more of the question that i i would be asking if i was in one of these teams That is the most fun or like meaningful or interesting question to ask right now. But it also seems like it's the hardest one. So like the part of this conversation I was really excited to get into is this, how do you reimagine what's possible? Both from a what we can build perspective, but also then to your point.

How do we reimagine what's possible with product, with design, with engineering to elevate them to the way things can go?

you see these teams kind of uh taking these almost like skunk work teams right and siphoning them off and i actually i am wholeheartedly behind that and you know at hubspot we kind of did this in the sense that um when we went multi-product We basically had to build a startup machine internally to create, validate, produce new products. And what we did is we siphoned that team off.

from the core product in engineering team at that time. We started from the entire company. We had a separate physical space. We had a totally different cadence, like all sorts of things. And that's because I think people forget. that when you're at a small stage trying to do something new, either build something brand new from the ground up or form a whole new way of working is that.

These processes, these things that have built up over time internally have been built up for the specific stage of stuff. And so you have to essentially replicate the constraints in the environment that a founder with a few folks might create outside of it to like force the right mechanisms. And so internally, they also funded this like startups. We would essentially seed bet. We'd give seed funding to a few ideas per year.

and uh that funding would be enough for like a few folks for a year and then at the end of the year they would come back for their series a funding to get the next year of funding based off the validation we would kill some and we would continue to fund some others and that's how uh the machine started found like the HubSpot CRM, the sales hub, stuff like that. And so I think you see a very similar parallel here, which is, look, people or teams are trying to not just...

disrupt themselves from the product. I think the Brex example you gave is like, what would an AI native version of Brex look like? as well as what would an AI native way of working is that you kind of have to create the container to allow them to take a completely different set of risks.

that you can't really take when you've got this other product. I think they're at like 700 million a year growing 50% year over year. I don't know, like they published a bunch of stuff because if you try to do it within that container, you're never going to take. uh the appropriate level of risk to actually find

the new thing, the innovative thing. And then once you find that new thing, that innovative thing, then you can think about, okay, how do I work this into the core machine? How do I transition this into the core machine? But you have to create that separate container, otherwise you're never going to find it.

Never. We're taking a quick break for a special feature, taking a look at how AI adoption is actually transforming the dev cycle with our friends and sponsor, Span. Steven Paletto, Field CTO at SPAN, shares how top companies use metrics for strategic inquiry, effectively drive performance improvement, and make smarter business decisions. And so as we've been rolling out Span with some of our customers, a common story that we'll hear.

is that one of the teams in the organization has particularly elongated cycle times. And maybe that's due to the fact that there's unclear expectations or collaboration practices around time to first review and how quickly...

code reviews get acknowledged. And without kind of having a single pane of glass to highlight this problem, it can be easy to just let the team continue operating as it does day to day. But these types of signals can then be used for targeted improvement to say, it seems like time to first review with this team is really long. Maybe we can help work with them to establish a tighter collaboration norm amongst the team.

And that'll help everybody feel like they're able to move faster. People have been working for years on quantifying engineering productivity. And the reality is there's no one single metric that you can boil it down to. It's a far too complicated problem space to try to reduce it down to something simple. The way we talk about it internally is using multiple different signals for the instigation of inquiry. So if you have something that's an outlier...

for instance, someone spending a lot on AI tools out of the norm relative to their peers, it can provoke curiosity. Like, why? What's going on there? When you... look at a metric for a given team, and you're able to benchmark that metric relative to the broader organization and relative to the broader industry, it can help you see when things are out of alignment or out of norm. And usually that indicates

some kind of opportunity for engagement inquiry improvement. And usually by following that curiosity, there's some learning or some insight. Span supports strategic inquiry by providing metrics across the software development lifecycle from a myriad of sources, including Gen.ai tools, source control, ticketing, calendars.

incident data, deployment data, and more. One of the things that we do is provide benchmark data. We also provide anomaly detection to help you see when trends have changed over time. So very helpful to understand. how you benchmark to similar companies or to other teams so that you can use those metrics as a signal and helps engineering leaders visualize where they stand relative to other teams inside the company and relative to the industry at large.

SPAN is the AI-native developer intelligence platform bringing clarity to engineering organizations with a holistic, human-centered approach to developer productivity. So if you want to get a complete picture of your engineering impact in health, drive high performance, and make smarter business decisions, go to span.app to learn more. That's S-P-A-N dot A-P-P.

To double-click on this idea, one of the principles that you've talked about around the fundamental changes for discovery and validation of product opportunities, and one of the things that you talk about is from exploring a couple different paths to exploring many paths.

part of what do I build and how do we organize the teams to build it question where with that phenomenon like have you what are ways that you're seeing maybe companies start to take more bets where they're exploring maybe going from exploring one or two different paths to now exploring

a larger multiple of that. I mean, like I think this HubSpot example of like you seed funded a handful of options and then that kind of spun out then to new products. I guess like what does that version look like now?

with this idea of like now that can sort of exponentially expand. Yeah, so this kind of goes back to the systems level view of EPD, which is there's been an over... focus on how do I do the same thing that I did before, like taking a feature from A to B in a much shorter period of time. and as a result that increases the number of features i'm just using that as a measurement right now but we could talk about in other ways i can produce more features within the same amount of time

But once again, the ultimate goal is to produce product that my customers adopt. In order to do that, actually, what I see a lot of teams doing is you just increase the velocity of shipping things that people don't adopt. So there's other ways to solve this. One of the core roles of product in, to some degree, design is essentially to make judgment calls.

Judgment calls when they're facing all sorts of different priorities, trade-offs, a fast-moving competitive market, positioning. How do I find the differentiation for our product in the market? You're making judgment calls and bets. on what to build. There's different ways to explore, validate, and de-risk those judgment calls.

A lot of people refer to this, you know, as the best product folks as having product sense because they have a higher hit rate on their judgment calls, right? And the way that they typically get to that is through a lot of reps. And I think with AI, you can essentially simulate a lot more reps and a lot more exploration paths in a much shorter period of time than you could do before. I might go from A to B on a feature in the same amount of time.

But if in the beginning of the product development process, one of the ways to de-risk and validate and get more judgment calls is rather than exploring two solutions, maybe explore 10 solutions, right? There's a diminishing returns for sure. It's not infinite, right? It's not like...

The hundredth solution I explore, I'm getting the same amount of value. But certainly most teams, if you really talk to them, they kind of narrow in on one solution right away. And then they kind of feel the pressure of, I just got to like pump this through the pipeline. Otherwise, we're never going to develop it.

That's kind of what most teams do. And so what we're doing and like what we're building at Reforge is we have a prototyping product called Reforge Build that's specifically designed for this phase of the process where a lot of other the AI app builders are around.

Let me build full stack applications. They're more for like the new entrepreneur wave, which is a really powerful thing. But when you're a product team, you're working on an existing product, like all that kind of stuff, you have a totally different set of constraints and problems you have to deal with.

And one of the things that you're trying to do early in that process is not only like, how do I take an idea, prototype from my existing product, but how do I explore a bunch of variations very quickly? And we've kind of built that into the product. You can do it other ways, but you know, these...

you can not only explore a bunch of things simultaneously in parallel, but a lot of times you can do it without like designer constraint, which was one of the main reasons before you'd only kind of explore one or two solutions. It's like you just never had enough designer. capacity to truly explore these things. But you can also validate these things faster as well, right? So there's a bunch of these tools like...

We have one called Reforged Research that's in AI interviewer that you can just like send this thing, you have the prototype and you can send the AI interviewer to 100 folks and you'll wake up the next morning with 20 pretty rich responses.

and you'll be able to iterate again in 24 hours and then do the same thing right like those cycles are just much faster so once again it's changing the frame of i think most people have like i used to be able to produce a feature go from a to b in one month And you could either look at that as like, okay, well now I can produce four features in that same time period.

But the other dimension to think about is your hit rate on the adoption of those features. And as a result, when you look at it through that frame, you might solve for a totally different set of things. Now, ultimately, I think it's going to be a little bit of a blend of both.

right like we're also we're going to work faster we're going to get through that pipeline faster but we're also going to be explore variations but if you optimize only on one dimension you're just going to end up you're going to end up with some problems one of the dilemmas that starts to come to my mind is the question of

I guess, like, am I building the right thing? Or what do I what do I build at a high level? And I think one of the parts that I want to explore a little bit is like zooming back up to the kind of first question, which is how do I with my EPD team rethink what we should actually build like with

the dynamics and what's possible now with different products. Reforge sort of represents an incredible example of this because of your shift from education company to education plus the suite of tools that you've launched over the last few quarters. It sort of represents a new look on what is now possible.

for being able to provide and support and enhance product management. How do you begin to even rethink what's possible? What does that conversation look like within EPD teams or within the people that are helping make that happen?

as well as where you think your specific category is going to go. What you have mainly seen in the spaces is that incumbents have essentially raced into multi-product expansion into known categories with known solutions because they can produce those things with much smaller teams and much faster periods of time before they were faced with a constraint and opportunity cost that they no longer constraint we just have to name one company

Figma, who now like with the recent acquisition of Weavey, they've gone from a two product company to like, I don't even know how many products they have at this point in a very short period of time. And there's tons of examples of this. there's applying ai to known categories known problems known solutions and what they're really doing there is just like leveraging distribution and trying to cut off known attack vectors from other companies and and other startups startups don't have that game

It's way too competitive to go into known categories and known solutions. I'll just give you one example. I've looked at 15, 16 different versions of... AI session replay companies in the past four months. and i see three new ones in every single yc cohort like every single one it's not just that category it's almost every single known category it's not a winning formula for for startups and actually i

the oxygen on that first strategy for incumbents is also going to run out. They will eventually have to move on to these things. And so trying to boil down, like essentially the question you're asking is like, how do I build something super innovative?

there's no formula for that i think you can once going back you can create the right environment and constraints for it uh which is you know building small teams that are taking huge risks if it doesn't feel risky then that's probably a signal that you're not taking a big enough bet um i think another signal on this on like are you taking a big enough bet is assessing like

Does this feel easy to build or does this feel hard to build? If it feels easy to build, you're probably not solving a hard enough problem. We live in a competitive environment. and everything gets arbitraged away. And so if you're taking an incremental bet, that thing just gets commoditized super quickly. And so people have to build harder and harder.

things right and so in in the nature of ai the way that ai works is it's really good at accelerating you on things that have been built in the past right because it has that in its like training data set uh but You've got to start solving new, unique, hard problems, things that you don't necessarily see a solution for yet. But if you find a way to solve it, it has an outsized outcome.

And that's ultimately the judgment. And this is really hard for larger companies to lean into. How do you lean into things where you don't see a known solution? And it also takes a special DNA of a... person. You can't just take a typical product manager, a typical tech lead, a typical designer off of a product that is 100 million ARR, right? And put them in this environment and say,

Find a problem you have no idea what the solution is for and just see if you can find a solution. It takes more of that early stage like founder type of mentality and DNA. And so people are just good at different things. So once again, I think you can create the right conditions for it. versus have a formula. And then you have like signals to understand, are you taking big enough bets? And then the question is, is also if you're an incumbent, how many bets can you possibly take?

Because not all of them are going to hit, right? So, you know, at HubSpot, we seeded three or four a year before we kind of figured out the machine. And then that's the risk of a startup is you typically can only take one bet. And all your eggs are in that basket. And that's why it's like you either win or you lose. It's binary outcome. It's very it's very exponential.

I'm wondering if you have maybe a couple of examples that you're observing of people increasingly working on harder and harder things. What would be an incredible goal or an outcome for people listening is like, okay, how do I change my frame of thinking to go from duplicating competitor features

on cracking a harder and harder class of problems for our specific maybe category that we're in. And I think I'm trying to understand, like maybe illustrate that a little bit for the types of thinking that people are doing. What does that look like to... build out something maybe that shifts you out of what your normal product is and you're not trying to just eat the features of your competitors around you.

I'm trying to think example. I mean, of course, most of the examples are going to be in early stage startups versus incumbents right now because of the trend I mentioned of like the incumbents, they saw the... short-term opportunity of racing into known categories with known solutions for the most part. You can also look at like there's companies that have done a good job of laddering this up over time. So captions is like an interesting one.

they started off as uh one of these like tools for creators that like did the transcript cleaned up the ums and ahs the eye contact all kind of stuff like they would have been labeled as like you know a gpt rapper right But they used it as like a wedge to gain traction in an audience. And now they've leveled it up to harder and harder problems. They sort of saw it. And so this was maybe three, four months ago. I can't remember.

they actually built uh and trained their own model for a very specific set of use cases it does incredibly well with human emotion and like taking up taking a photo or taking a short something and what they're now doing with that trained model and they've now tapped into this whole new market a couple things where they're getting traction is a bunch of like media studios that you use it to prototype you know commercials and movies and like all that type of stuff essentially

de-risking incredibly high, you know, production stuff in a much higher fidelity way where they would used to have like these sketches on a whiteboard, right? And they were also finding like use cases in marketing. And so, and they'll take that and they'll probably... leverage it up to an even harder use case to go solve.

That was like somebody who didn't like stop at the features and the wrappers and the things that they started with. And they were like, okay, like if we take another step function leap here and try to solve something that we can't do with the models out of the box today, what would we do? you know, how would we do that? And they sort of took that leap, which I found pretty compelling. Another one maybe off the top of my head is Shopify.

They have this internal tool, support tool that they ended up building. This guy, David Wertz, who's a VP of product there. Rather than just building like a support agent, they viewed support as actually our customers are entrepreneurs, they're founders, they're creating e-commerce stores. And rather than just giving them like a support agent, we want to give them an advisor.

like a business advisor that's how they approached that and that framing for them led them to building this incredible multimodal agent that not only helped people do things in the product but advised them on a bunch of business knowledge along the way and um like i saw this demo where uh e-commerce store owner kind of came in they're non-technical and uh they're like how do i get my site

to go to this domain that I registered. Something kind of complicated because you got to go change domain settings. Their domain was like hosted on Cloudflare. This agent was like, We need to like change these domain settings. And it went and looked and it was like, oh, your domain is on Cloudflare. Can you go open up Cloudflare? And then the agent followed him to Cloudflare. And it was like, oh, you need to.

put X in this box and then you put X in this box and it's like, cool, then I'll click back, right? Like something that would take a business owner, non-technical business owner, an incredibly long period of time, very frustrating, like all that kind of stuff.

And along the way, it's also giving them advice on what they might do next to optimize their store. And I thought that was a really interesting framing. They could have just taken the approach of, I'm going to... install a fin type, you know, support chat bot, but they kind of reframed the whole thing and was like, actually, how can we deliver a level of support value that wasn't possible before, but is now possible, you know, with AI?

I think the Shopify one is such an excellent example because I think about some of the things that I've learned from some of the blog posts that you put out and one of them, the shifts that you highlight for customer expectations is give me a place to create, provide the tool, provide the platform to do the work for me. or advise me on

what needs to happen and take away some of the tasks to do that. The customer goal is not to set up the Shopify store, but the customer goal is to have a successful business.

help me set up the store to help me run a successful business. And I think that is such an interesting category elevation to really truly understand what is the core motivation of people and the outcomes that the people that are using your product are going after. And then how do you build something that is going after that hire?

order of thinking. A hundred percent. You know, Shopify is one of these interesting businesses that has relatively low retention based on signups, but those that they do retain create. an enormous amount of economic value just based on their pricing model and all that other stuff. So if you move that outcome just a tiny bit, huge lever on the business.

are in. So you were talking about how captions, they didn't stop at features and being a GPT wrapper, but they took a step function leap. I want to frame the operating environment right now and talk about the great distribution shift and the role that this plays in terms of how people make decisions on what they build.

short-term and long-term. Can you talk to us a little bit about this distribution shift that's happening and how it maybe is informing important product strategy decisions? And I guess maybe what it means for engineering leaders, like what they need to know about this distribution shift going on and how it can...

what they build and then how they build it with their team? The hard thing is not necessarily always the technology. One of my good friends, Casey Winters, he's trying to build this more AI native version of LinkedIn is probably the easiest way to describe it. It's called SuperMe or superme.ai, where you have like an AI profile version of you that others can interact with. The AIs can actually interact with each other. So you could actually go there. You could ask.

me all sorts of versions of growth questions and stuff. And then if you ask a marketplace question, it's going to bring up advice from Casey because he's a deeper expert.

on marketplaces something like that no are they inventing like new technology no the hard problem for them is it's a marketplace network problem around liquidity they're applying ai to create a unique experience and there's a different hard problem a different risky problem that they need to solve so it's just a good reminder that it's not always a technology thing that we need to solve as like the hard thing the risky thing that we need to go try

The distribution shift is incredibly important because there's this concept that I talked about like 10 years ago called product channel fit, where the best products mold their product to fit with the distribution channel, not vice versa. That's always the mistake.

we have no control over the channels and where that we can tap into where our customers are living those channels define their own rules we can't change those rules but the thing that we can change is our product and how we mold to them This is most clear in the pre-AI version of products that tapped into SEO, UGC SEO. I think the Trip Advisors, the Pinterests, like those of the world.

They shaped their product experiences, their activation experiences, all of that type of stuff to mold and tap into that channel. which is what drove it. So you have to think about that. Those two things have to combine. So that gets to the distribution shift, which is with most technology, there's technology shifts, but then there's technology shifts with distribution shifts. And the ones with distribution shifts end up being the most powerful.

It doesn't always happen in conjunction, but when it does happen, it can completely change the landscape. And of course, previous versions of this would, of course, be things like social. Mobile is another one. Obviously, search is another one. A couple of years ago, We had this technology shift of AI, but we had not had the distribution shift yet. And if you look at history, that's pretty normal. The distribution shift typically comes about two years after the technology shift.

because it takes time to apply that technology for consumers to change their habits, start to aggregate on a different platform, a different set of channels, like all those types of things. But when there is a distribution shift, there's huge winners and there's huge losers, right? The winners are the ones.

that find a way to take advantage of that new distribution platform, early mold their products to the new behaviors, the new environments, the new rules that those channels set. And the losers are the ones that tend to wait way too long to do that. And so they either miss out of the boat or they get disrupted. This happens with every single cycle, every, every single cycle. Now, the key thing is about these cycles that they follow a pattern.

The pattern is typically one is that a competitive environment forms around a huge new channel or distribution opportunity. And there's typically five to seven players that tend to battle it out for it. So in social. It was Facebook. Google made a bunch of attempts at it. You had things like Friendster, MySpace. You had a bunch of others. And they were all throwing tons of money at it. You know, in search, there was all sorts of different search engines besides Google at the time.

In mobile, you had not just iOS, Android, Windows made an attempt at it. There was multiple, Facebook made an attempt at it, if people remember that. So anyways, there was all sorts of things. So that step one is like, there's consensus, there's a huge new potential.

channel people are pouring bucks into it so then there comes step two one of those people figures out what is going to be the moat the defensibility mechanisms that separates themselves from everybody else and uh and then once they figure out that moat they then go on to the next step which is they tend to go in an open phase

they either invite developers content other people to contribute to their platform in some way there's a value exchange it's like if you contribute to my platform i'm going to give you something in return typically it's in the form of distribution potentially additional monetization But then that eventually goes into a closed phase where once they have escape velocity, because all these people brought additional users and usage and moat, they then start closing it down to monetize.

And they close it down by levying a tax like a transaction fee. They start pushing down organic. reach in order to push you towards like ads or they just change the rules entirely and they absorb all the most interesting use cases and shut the platform down. This has happened over and over. Facebook went through this.

Facebook figured out it was about the social graph. How do I get as many friends on here as possible? They opened up the platform. All these apps and games came on there. They drove a ton more adoption, a ton more usage. The value exchange at the beginning was like, hey, I'm going to give you this canvas. You develop whatever you want. We just want these ads on the right hand side. We're going to give you distribution. And then they started.

basically taking that all back over time they first levied attacks and then they were like actually no you got to use our ad system and then eventually they just absorbed all of the best apps and shut down organic distribution you know apple's been through the same stage google played a much longer cycle over like 20 years around this to the point that search something on google and like 80 of the screen is filled with with ads right

All of these typical pieces. And then most recently on LinkedIn, you saw this, which was like they invited a bunch of all these content creators, gave them a bunch of distribution, organic distribution. They introduced thought leader ads and they've been shutting down organic distribution over time to push people.

happens over and over again. And so months and months ago, I predicted that OpenAI was going to be the winner and they would launch some form of app platform or distribution to basically king themselves. And that's exactly what happened. Basically, now we went from phase one, that competitive environment, to phase two, they identified the moat.

around like it's about memory and context. And now we're in phase three, the open phase, which is, well, not quite yet, but they've announced the app platform. They're planning on opening it up at the end of the year. And this is so important because they're going to have a billion mile by the end of the year. no matter what company or product you develop,

It's most likely that your target, a huge portion of your target audience has shifted their time to spending time on ChatGPT over some other platform. And so as a way, you have to participate in that platform in some way, shape or form. The folks that are able to mold their products to it earlier are going to be the biggest winners because the biggest mistake of these shifts is that people with existing products, what they do is they'll copy and paste their existing product into the new channel.

versus rethinking their product from the ground up of like okay what are users doing in this new environment how are they behaving differently and how do i shape my product to fit that you know in social we saw all the web game developers they just copied and pasted it into so none of them won it was like and other ones that got all the traction and grew up from the ground up.

same with mobile they people would copy try to copy their web app into the mobile experience none of them wanted it was all the folks that thought through the mobile experience um through first principles and so that's both the opportunity and the challenge at the same time and this is not a marketing problem this is a pro product problem. And that is something that EPD leaders have to understand.

When you talk about the opportunity and then the challenge there, it can seem like you can fall into this trap where you really try to optimize, this is the new distribution channel, this is how we're going to build and shape things for new users. So then how do you survive? from phase two to phase three, where maybe you're benefiting from the distribution and then it's becoming closed. What can an EPD leader strategically think about before that shift happens?

I would actually say right now we're in a moment in time that it's a good thing to have in the back of your mind, but it shouldn't be the focus. Because the first job is like... you see this massive wave coming you just have to catch the damn wave like that is your first job right so until you do that

It does not matter. So your first job is just catch the wave, get in the ocean, paddle out there, try to get on the board, catch the wave. Once you're kind of surfing that wave, then you should immediately, like the moment you stand up on that board and you're going, then you got immediately.

start thinking, okay, I know this end phase is coming, the closed phase. How am I going to de-risk myself from that? And that part, people need to be thinking about, okay, well, how am I capturing unique and proprietary data so that, you know, users can't get this exact thing like just through the horizontal platform of like ChatGPT. Or I'm using the channel as like a starting point, but I've built out very deep, specialized workflow for my customers.

specific use case that the platform is never going to build out. Or if they do build it out, it's going to be far inferior to mine. these are all the questions that you have to start asking yourselves and start building towards to both leverage the platform as well as kind of de-risk yourself a lot of people take this mentality of i know this phase is coming so i'm not going to play the game and i guarantee you that that is the losing strategy yes it's a prisoner's dilemma

We all know how it ends. We all have to play the game anyways. And the other thing is like these companies aren't evil either. You could take that kind of stance either. It's just the nature of companies and competition. You know, like they have to monetize. They have to keep growing over time.

Every product is built off of the back of another like distribution channel, another major distribution channel. This is just the cycles, it's the waves. And so you just kind of, it's a game and you got to play the game. And the deeper you know the rules, the better you can play the game.

That's so great. So far, just a few categories of things we've covered so far. So we kind of started to poke around into the changing dynamics of EPD teams and getting more reps, delivering complete adopted products. We started to talk about like elevate to higher class problems and some ideas around how to do that. And then this really clear view of the shift and some of the decisions that you can do and the choices that you make. I want to spend a couple minutes to talk about.

With these things, how do we build the EPD functions that can deliver and execute on the different topics that we've talked about?

And so I wonder if maybe we could poke around like some of the experiments that you're seeing that are maybe proving out to be really interesting ways of working for EPD together or structures or models or approaches to just make this workflow of EPD better, faster, stronger, more... complete adapted products, maybe observations or things that you're seeing there that are really working on how we build these things and how we bring these things to reality.

Yeah. So if you're an existing company with any sort of scale, I think it goes back to what we were talking about earlier is you should split your initiatives into two. You should basically find some containers to take incredibly high risk bets, give them a lot of autonomy, let them do those things outside. of the risks of maybe like the core product core product team all that kind of stuff and yeah do you have to create

clear separation. It's like a death by a thousand cuts if you don't create that clear separation. But then you also have to, you have to do a track list. You also have to be working on your existing core team and your core machine as well.

I think one of the biggest things that I've seen, and I'll frame the problem, and then I think everybody listening will probably have a ton of ideas on solutions and stuff, which is I think it's no secret to folks that you... both move faster and build better product when it's constrained to a small group.

of individuals with complementary skill sets. And that's why the EPD kind of triad formed, which is like complementary skill sets, small group. The problem is when you actually look and like map out. follow like literally follow uh an initiative from task to task to task it is not constrained to this core group right and when you do constrain into this core group the cycles move really fast they get to know work together all kinds of stuff because what happens is

When a company, as they evolve over time, as they go from, okay, I've got this core triad, EPD, that are working really nicely together. And then they start surrounding them with all sorts of specialists, user research, different forms of design, PMM. all that kind of stuff. The challenge then becomes all these folks essentially become inboxes for different tasks. And each of these inboxes have a huge backlog in turnaround time.

So I'm a PM or a designer and I've got a question about my users. It's like I send an email to the user research team. They have a whole backlog of stuff. Maybe they get back to me in two weeks. Right? It's the same thing with all of these specializations. PMM. Hey, we're going to launch this thing. And it's like literally every cycle of iteration is like a week or two. And so what... This triad has turned into a lot of companies is like an octagon.

uh of a bunch of folks and there's all these delays of like inboxes outboxes like iteration cycles right so for me as a pm where i'm trying to make this judgment call and to make that judgment call I'm going to make the best judgment call by collecting and looking and exploring all of this different data and having these faster iteration cycles on it. The problem is if the iteration cycles are so slow, I'm just like either not going to do it.

Or I'm just not going to get as much exploration to make the best judgment call possible. And then that all kind of flows downstream. And you've got examples of this across the EPD triad. I think the biggest thing that I see teams doing is that the opportunity isn't to turn your designers.

into these mediocre engineers, as we were talking about before, or your PMs into these mediocre engineers, it's actually how do you eliminate this octagon of specialists so that you can move faster now everybody's going to hear that first and be like how do i eliminate these specialist roles and that is not what i'm saying so let me just be very clear the role of the specialist needs to shift from

I become an inbox for these specialized tasks to I build the tools and the systems that these EPD triangles can self-serve into it. And actually, I can focus my time on much messier, much long ranging, like strategic things that these EPD teams are not going to do or they're not equipped to do. There is no reason at this point in technology that if you need to understand a question that you might have qualitative data on.

in support or some survey that you did to go to a support ops person or whatever with tools like reforge insights that aggregates and stuff like there's no reason that that should exist anymore Put these tools, put these systems in place to allow these teams to self-serve into it. That is the job of the specialist now is to build these systems and not to be an inbox. If you're an inbox, I think it's just not going to last. And so that's...

where I would spend my time and that's where I'm seeing the best teams kind of eliminate friction and get to that ultimate output that we talked about before which is just like product that customers adopt which is like how do I constrain these cycles and these workflows to the more of the core triad versus this octagon of specialists. Does that make sense?

It absolutely makes sense and I'm laughing because I'm just thinking about all of the things that I'm an inbox for. And I'm like, wow, I really need to start to think about how do I extract that as a system and build the different elements around it. On that real quick, a huge portion of people's jobs is like if you're an inbox doing that task, I think you're actually doing it wrong these days.

And this is really hard to do. You should be sitting back and saying, wait a second, if I'm going to do this task, how do I record this task in a way or capture the process and the knowledge in a way that that becomes enough context for the right AI setup to do on its own? Literally the behavior, you got it. But a lot of us feel like we're probably on this hamster wheel.

of just like doing those tasks. And it's very hard to get off that hamster wheel. But that's a huge portion of our jobs today, especially if you're these specialists, which is like, how do I encode this system and this knowledge in the right way for AI so that

I stop becoming an inbox and I start creating the system for others. Is there maybe a final vision for the future of how EPD works together? If you had to imagine your perfect flow? Is there maybe a final vision to leave us with for this system works in this kind of way? No, because I think there was never a perfect ideal system before. And actually, the chase of that ideal system, I think, leads to suboptimal outcomes. And the reason for that is that...

There are different types of product problems and different types of product work. And each type of problem and each type of work require a different system. Growth product work required... a highly iterative experimental data-driven process that was very different than feature outcome and feature development that was very different from scaling work that was very different from how product market fit expansion like how do i go multi-product like innovative

Each one of them over time, like we had codified this in our product strategy program, required a different approach, a different set of tools and stuff to do right. And the biggest mistakes is like you take a growth process and a growth system and apply it to like a feature.

process or vice versa, right? Like that's kind of thing. And so I think roughly those shape of problems stays the same. You've got product market fit work, you then have feature work, you then have growth work, you have scaling work, and then you have product market fit expansion. And I think all five of those

There is a different, like the system and process will change for all of them. And I don't know if I necessarily have like the answers for all of them. They probably share a level of attributes, but. I actually think the growth system is going to look a lot different than the feature development work and stuff. The old experimental system, the thing that I codified 12 years ago, 13 years ago, in my growth machine framework.

That's going to change. That's changing quite a bit right now. And I think the same is on the others. So I think we would need a whole nother hour just to take those five and talk through them piece by piece and talk about some of the differences and changes. And now I'm saying that I'm like, man, that was... that's probably a good blog post I should write. Well, we look forward to reading that.

Brian, I want to be mindful. I know we're two minutes over time. I have a couple rapid-fire questions. I've got like two minutes, and then I've got a whole company meeting I've got to get to. All right, we're going to jump into as many rapid-fire questions as we can in maybe 90 seconds. Okay, what are you reading or listening to right now?

Well, I'm not reading a ton because I have kids and my brain can't absorb that much. I do do a lot of listening. This is a cliche answer, but Invest Like the Best podcast. I actually just, I think the host of that, Patrick O'Shaughnessy, is one of the best interviewers in technology. He asks some amazing questions.

I just learn a lot just from the questions that he asks, not just the answers. And so I kind of admire that just from a content creator perspective. A trend you're seeing or following that's not mainstream yet. So I had this friend, Justin Mares. He's the founder of this company called TruMed. He's always been on the front lines of like all this like crazy health stuff.

And, you know, he's like invested in these companies that basically analyzes and tracks your pee for health. And so like that, like stuff like that is stuff that I'm following and I'm always interested in because I think there's just going to be all sorts of new.

ways to track and analyze our health that we haven't even seen yet. A quote that you want to leave us with or a mantra that you've been living by. I love this quote, which is miserable people overthink and underact, right? So energy goes where attention flows. It's a good reminder, especially in this day and age where there's so much change going on. I can easily find myself in these cycles of just like constantly thinking, analyzing and stuff.

And sometimes I just need a kick in the butt to be like, you know, I just get back to doing, put one step in front of the other. That's kind of what makes things happen. Brain. Thank you for helping us answer the unending question of what to build and how to build it. I just can't begin to share how much this really, I think, helps shape the future of our community. So thank you.

Yeah, feel free to reach out to me on LinkedIn or other places, folks that are listening. Look forward to the questions. If you're listening to this and you're wondering, how can I connect with other engineering leaders in my city? Pull up your phone right now and go to elc.community. Click our chapters page. You can see that on the menu on the left. Find your local chapter and click join.

We're hosting virtual and in-person events all the time. And this is the best way to help you get involved, expand your network in your city and support your leadership and career growth. So pull up your phone, head to elc.community, join your local chapter and get involved. A huge thank you to all of our local leaders who make community happen. And thank you for listening to the Engineering Leadership Podcast.