Technology arrives in waves, and every wave creates giant new companies while killing incumbents that fail to adapt. Paul Adams argues we are at the "1999" moment of the AI curve, right at the beginning, and that the second and third-order effects, not the technology itself, are what will reshape how we live and work.
He walks through how Intercom bet the entire company on AI and rebuilt almost everything as a result. Gone is the stable medium-term strategy, replaced by a two-to-three month horizon; gone are clear roles, replaced by multitasking generalists who build to think; gone are the classic PM-designer-engineer triads, blown up in favour of small, temporary workstreams run by a directly responsible individual. Execution is now uncertain, so they build in code with customers rather than polishing Figma files, and Adams' closing point is blunt: the best ways of working don't exist yet, so you have to invent them yourself.
Auto-generated transcript - may contain errors.
Tap a timestamp to jump the video.
Good morning everybody. Thank you first to Brian for inviting me back. It's always a dangerous proposition. You never know what you're gonna get. I'm gonna talk about how we've completely changed almost everything about Intercom. Intercom, if you don't know us, we're a thirteen year old company, the customer service product and with the emergence of AI we changed how we work, our product focus, our roadmap, our strategy, the company that is Intercom today is unrecognizable from the company even twelve months ago.
So I'm going to talk about why. You might love these letters or you might hate these letters, but they're everywhere and we at Intercom are all in on AI. We have fully, fully bet the company. I'm going to first zoom way out and explain why and I'm going to get really into the details of what's changed, how we work now, how we build software.
We build software in entirely different ways to even twelve months ago. So if you look at technology and study how technology works, there are very, very consistent patterns throughout history and this exact same pattern is happening now with AI. It's the same thing, it's the next generation.
So technology happens in waves, people call these S curves because they kind of look like an S, and it's very, very predictable as one S curve ends and another S curve begins, and you can kind of break down these big technology cycles into these macro enormous kind of eras like transforming energy.
People say now we're transforming information or we're in the information age, it's probably lasting like going back eighty or one hundred years, certainly going to continue for many many more decades. You can kind of look at all of these technologies and study what happened when they were happening and what played out.
You can break these down too, you can break down these bigger waves down into smaller waves, so you can look at the information age and break it down into these other sub waves. These are the last three sub waves, the PC era, Bill Gates said he put a computer on everybody's desk, prior to that computers were enormous, big giant cupboards full of computers.
He put a computer on every desk. Then we had the Internet age or cloud that played out over kind of last ten, fifteen years, most recently mobile social over the last decade, and clearly we're entering the AI era. Each time these waves come about two things happen, one giant companies get created, enormous companies.
If you look at the most valuable companies in the world right now, they're internet companies. AI will do the same, enormous companies will get created. Also, companies you know and love will die. It happens every time, they fail to adapt, fail to change, and I'll explain why in a second.
These S curves, you might be quite familiar with this, these S curves are shaped like this because of how the technology disseminates through society. So there's always very slow initial adoption of the technology, it's not clear if it's valuable, there's usually a breakthrough invention that kicks it off, but people don't really know if it's hype, if it's not.
Sometimes people bet in these waves and they're not quite right. The metaverse for example was something that didn't quite work out the same way AI did, but there's always slow initial adoption, then the value becomes clear, people start using that technology, the use cases become really, really obvious, more and more people get on board, you have like early adopters and then later adopters, and eventually the innovation cycle ends.
So that's how these things work. The other thing that happens during these waves is that there's always cynics and there's always skeptics, every time. People love finding great arguments for why things fail, especially things that are hyped. It's nothing more fun if you're in Kyos or any other pub to talk about how AI is a load of ****, it's not working, it's not, it's over hyped, etcetera, etcetera.
It happens every time. Here's the back to the horse movement. When cars showed up, you know, horses were better than cars. They're faster, they can go over hills, cars have to stick to roads, horses are natural, they're cheaper, there are just as many reasons why horses are better than cars and lots and lots of people really really tried hard to convince people that cars were bad and horses were better.
Here's another example, these cycles always end. I'm going to give you a demonstration of that. I bet at anything that everyone has a black rectangle in their pocket. Right? And I bet you the next iPhone, this is actually Google Pixel phone, you wouldn't even know, the next iPhone is gonna look like this too.
It's gonna be a black rectangle. The next Google Pixel will look the same, all phones will look the same, Android and iOS are almost identical, we're done. The innovation cycle is over. Apple I think are bringing out a folding phone, but we know that no one's going to actually buy that.
So they're all the same, black rectangles. That's how these things work, but what's really important to internalize about these waves isn't the wave itself. It's not like the phone, or even the car. It's what these waves do. It's what they do to society and how they change how we live and how we work.
I'm gonna give you one example. This is the combustion engine, a breakthrough invention that led to the car. The car, of course, changed how streets look. This is Manhattan, more or a decade apart. You might see little red circles. On the left, the red circle is the only car in the photograph.
On the right, the red circle is the only horse in the photograph. So these these completely change how society looks, operates, etc, but it's the second and third degree effects of these that actually really, really change how we think, work, live. So the car created suburbia and suburbia created the shopping mall.
So we've changed how we live, we've changed how we buy, and the suburbia, the car, the shopping mall also led to McDonald's, and how we eat, what we eat, all sorts of things like that. These are the second and third order effects of these technologies.
Happens every time. The Internet had second and third order effects. So did mobile and social. Not all of them are good by the way, some are good, some are bad, but they always happen and AI is going do the same thing. So it's kind of undeniable at the beginning of this curve, this new S curve that is AI.
Mobile is over, we're into the AI curve. Okay. So where are we in the curve? How far in are we? How long will it last? This is this is how I kind of think about it. Imagine it's nineteen ninety nine. I'm old enough to remember that. Some of you are not.
Imagine it's nineteen ninety nine, the Internet's just showing up, it's new, it's unclear what it is, someone from the future, let's say twenty twenty prior to AI, has been transported down to nineteen ninety nine to explain to you what the Internet is. Maybe they explained SEO to you.
On one hand, it would sound like absolute nonsense, like what are you talking about? On the other hand, know, people would be like, is that what you did with this amazing technology? Right? But it would be so foreign, you would not be able to understand what the hell search engine optimization is.
Then they showed you TikTok, and they showed you like short video clips, influencers, billions of people viewing these little videos, new celebrities, new trends. It would be like incomprehensible to you back in nineteen ninety nine that this is what the Internet technology would lead to.
Incomprehensible. That's where we are at with AI. We are at the very beginning. It is nineteen ninety nine. That's where we're at. So the questions really are how big will this be and how fast will it happen? You can kind of think about this as like if you really want to go there you can say like what's the McDonald's drive through of AI?
There's a version of it coming someday, but it's going be huge, it's going be enormous. There are all sorts of things AI can do today and we're at the very beginning of the AskCurve. AI can use computers, AI can think, it can see, we have agentic reasoning, AI now can do both system one and system two, it can basically think how humans think, it can reason.
We have robotics, something that's kind of exhilarating and terrifying is the combination of AI and robotics that is happening all over the world, many many people developing very sophisticated robots. So if you want to imagine what the SEO equivalent of is, we're building robots that we powered by AI, that is going to entirely change how we live and work.
So that's what's coming. My take on how big and how fast this will happen is that it's going to be very, very big. I personally think we're talking industrial revolution type scale, certainly as big as the internet, probably bigger, and I think it's going to happen very, very fast.
It is impossible to keep up with how fast AI is developing. It's going to happen very fast. So at Intercom, we bet the whole company on AI, the entire company like all chips in the table and we rewrote our entire strategy, completely changed how we work and we did it two years ago.
We did it two years ago. We saw a chatty PT emerge and we could see straight away LLMs are going to completely change customer service. It was so obvious. These are things common customer service tasks on the left and on the right are things that LLMs two years ago are good at, out of the box good at.
It's basically the same list. LLMs are really good at the things customer service reps do, but LLMs unlike humans speak any imaginable language. They're instant, they don't have to think, they work twenty fourseven, they don't take coffee breaks, they don't go on holidays, they don't sleep.
So if you have something that's just as good, but it's cheaper and it's better and it's faster, you're going to pick that solution. Businesses are going to pick these solutions over the things that they have today because they're better. That's where we're going.
So we very quickly realized we have got to bet the company. This is not a choice. We're default dead unless we bet the entire company on AI, so we did it. It was risky at the time because it could have been another metaverse, thankfully it wasn't.
It was a good decision, but it wasn't obvious that it would be at the time. And we changed how we think about strategy. So I'm going walk through this old way, new way. The old way is how we used to be, how we used to work and think, I think it's how a lot of people still work and think, I'm going to talk about the new way at Intercom, how we think and work now.
So the old way, we used have a very stable strategy, we're quite proud of that fact, a very kind of clear vision and medium term strategy, and people used to say to us, hey a lot changes at Intercom really fast, but it's really good we have a stable strategy.
We all know where we're going. We've completely changed that. We do not have a medium term strategy. I do not know what our roadmap is for October. No idea. I have no idea what we're building on October. We have a very fast changing, ever changing short term strategy and the reason is because LLMs change so quickly, the capabilities change so quickly, it just doesn't make any sense.
We kept changing things too often to embrace the fact that we don't know and start to work in an entirely different way. AI just changes too fast. That's the very first thing we did. We have a short term strategy, we have a vision, a mission, kind of multi year kind of hey, we're trying to go that way, but our actual company strategy roadmap is like two to three months set.
That's number one. What else? Well if you think about strategy and changing strategy, you've got to realize that great strategy is a set of connected things, and so if you change one thing this is a famous diagram from Southwest Airlines. Southwest Airlines were like the predecessor to Ryanair in the States invented all of the stuff Ryanair do so well now.
Quick date turnaround, you have to pay for meals on the flights, all sorts of stuff like that, but it's an interconnected system and the reason it works is all of the things work together, the designs work together, there's like the compound value from how these things work, but if you change one part of it, you change and kind of break the whole thing.
So if you're going to start changing your strategy, you need to start thinking about all these Because a business is just a mutually reinforcing activities, you can't just swap out one piece, you got to rethink the whole thing. So when we're thinking how we do strategy, we're thinking about how do we build product, what product do we sell, what's the buyer, who's the buyer, how should we charge for it, what brands do we need to attract people to this product, what sales motion should we use,
like all these things are interconnected. When you start changing one, you've got to start changing the other. So you can't basically look at this kind of set of connected things and decide this part would be our AI part. Everything else will kind of stay the same, but we're kind of like going to do the AI stuff in our company over here.
It doesn't work like that. You have to reimagine your entire business from a blank page. You've got to think that this has to be built in, you can't just bolt it on. And so I often say to people like, you need to become an AI first company.
Everyone here, I can't think of any imaginable company that doesn't need today to become an AI first company. Now some people think and say often to me, well that's obvious for you to say, you're Intercom, you're customer service, you're building Fin, an AI agent for customer service, that's like the obvious kind of extension and progression of your industry.
I'm saying okay, you don't think it applies to you. Imagine it's nineteen ninety nine because that's where we're at on the S curve and someone said you need to become an internet first company. That's where we're at. Many people at that time rejected this notion entirely.
They said, we've an amazing brilliant successful business, we'll do the internet thing over here. This even happened, I worked at Google, I worked in the mobile team at Google in London, it happened then too, the mobile curve was taking off, Google had a mobile team and the mobile team was in London not California because it wasn't important enough to be in California.
This is how these things work. So you have to become an AI first company, I don't think people have a choice. You need to lean into it and understand what that means. So it's an entirely new mindset, you can't just bolt this stuff on.
So we changed our mindset and we think AI first about everything. Alright, what's next? Here are roles that I'm sure are very common to you and common in your company, product manager, designer, engineer, it's very clear what these people do for the most part, you can add other roles too, researcher, data scientist, product analyst, all sorts of different types of things.
It's very clear what these people do, clear roles, responsibilities, job ladders, job descriptions, etc. That world is gone. This world is gone. It's already gone. This is a new world. Everyone can do everyone else's job. At Intercom, we have all sorts of permutations.
We have PMs who build prototypes and code. We have designers, vibe coding marketing pages, we have engineers doing customer discovery. Everyone's role is blurring because people now have all the tools to do the other things. So all this world is kind of merging together and it's like a one way door.
It's not a maybe we'll have it different in our company. Nope, you need to become an AI first company. See things are blurring. I'll show you a simple example of this. This is a blog that Des, our co founder built. It's a fully working blog front and back end.
Des is not an engineer, Des vibe coded this himself. He built it himself, no engineers involved whatsoever, and he built it for twenty eight dollars and a few solid hours over a few weekends. Right? We had this entire micro sub industries who build blogs for people.
So this whole world is changing, and again, it's a mindset shift. In intercom, we have people and what they say is, I use these tools, you can pick whatever tool, cursor, lovable, replet, bolt, there are so many, so many tools, but you build in these tools because you build to think.
I write a lot to think, these days we build to think. It's way faster. When you start building things, you can build things. Anyone here, anyone here, I can't code whatsoever. Maybe back in the day I did some HTML, but I can't code, I can't build anything.
Now I can, I use Replit, I use Lovable, I use Claude, I can build stuff, it's an amazing experience? So that's the third thing. The world of clear roles and responsibilities in our industry is gone, we're now in a world of multitasking generalists.
Generalists who can multitask will win this next era of technology and software in business. Okay, what else? How do you build this software? So here's how we typically build software and this has software's been built for like a decade. There's a fixed state of technical capability, it's very clear what the technology can do, so you can kind of build anything you can think of.
It's very rare these days for like a SaaS company or a consumer app company to think that's an amazing idea but we couldn't build it, it's just too hard from a computer science point, you can't build it. That doesn't happen. You can pretty much build anything you want, it's very clear how to build it.
So you kind of pick the highest value areas, whether it's features, new product lines, whatever, and execution is certain. If you have enough resources and enough time execute, you can do it, you know you can do it. So execution is certain and when a feature goes live, if you've done a good job, that's kind of the end of the project, it means you've solved the problem.
Building AI first products is completely different. This took me a long time to really truly internalize. It's completely different. It's not a little bit different, it's totally different. First of all, the capabilities are constantly expanding. Every few months there's a new model from a different company that can do new things, and the only way really can truly know how it works and how the new things that can be done with it is to build in it.
So you're never really sure, you have to like explore this new technology and you're never really sure what's happening at the edges because it's generative. You don't really know if it works, if it doesn't work until you build it and try it and start using it and start pulling it apart.
So you're never really sure what's possible and when start to explore it, you're never really sure how good it will get. So for example like again, I'll use Finn as the example, we use Finn as our kind of number one product, it's our best selling ever product by a mile by the way, the spoiler there is this is working, it's our best selling product by a mile, but we don't know if it's good.
We will build new types of things that Finn can do, can Fin do these types of customer queries, can it talk to these systems and bring back something sensible, like you can build all that stuff, but you don't know if it works. Like agents and LLMs in general are amazing at giving you answers that look incredibly accurate and correct, but not always right.
There's always little mistakes and little things, so you got to just keep building and explore and it could be really good or might not be that good at all. So you don't know how valuable things will be, you don't know how valuable features will be, and you don't know how to design it in an appropriate way.
So this why you've got to get building it and building it in code and putting it live. So this whole process is one where execution is uncertain, you're not entirely sure what you're building at times, you're not entirely sure if it works. We've stopped creating like nice mock ups, high fidelity design, we've stopped, none of that happens anymore.
We build like really rough stuff, rough mock ups, designers are in tools building things, using Lovable, using Replit, using Cloud Code, that's what they're doing, it's a waste of time to make nice beautiful Figma files, a waste of time because once you get to actually building the product you realize, **** it doesn't work that way.
It doesn't work that way, we've got to change how we think about the product. So that's how we do it, the execution is really uncertain and once the feature is live, you're still not sure if it's any good. You've got to monitor it, you've got to look, you've got to talk to customers, you've got to see the actual results that generating, and you have to obsess about it.
Once it's in the market you've to obsess about it. So you might ask, okay, how can you build a product when you're so uncertain about everything? Does it work? Is it good? Is it valuable? How can you build a product? This is how we do it. We do it with our customers.
Actually with them. Some of them are like, what do you do with them? I don't mean to use a research, I mean we build it with them. They're together with us. We visit them, they visit us. This doesn't mean like take all these customers' requests, enter it like an enterprise business and build all the stuff they ask for, that's not what this is.
This is what can we make, is it valuable to you, does it work, alright here's the things it's generating, do they make any sense to you? Is that what you thought would happen? So we're like deep, deep in partnership with customers. We've got these really tight feedback loops.
You could say it's a really commercial focus because we want them to succeed. The products of no value to them then it's of no value to us. So that's how we work and all these new types of roles are emerging. This is a role called forward deployed engineer, this is at OpenAI, we have very similar roles too.
Here's another role from Anthropic, a demo designer, someone who builds demonstrations of what the technology can and can't do. These new types of roles. If you go and look at Anthropic, OpenAI, companies like that, even Intercom and look at the roles, the types of people we are hiring, we are hiring all sorts of different types of jobs.
We are not hiring your standard UX designer, your standard product manager, we are looking for new types of skills, new types of ways of working, people are open minded to know that like, hey you might be a product manager and you might start building in code, or you might be an engineer and you might start talking a lot to customers.
So there's a whole new world emerging. So that's the kind of next thing here, product execution is certain in the old world, in the new world it's very very uncertain. So you can kind of pause here, I have two more, you can pause here and think these are entirely different companies already, entirely different.
The company on the left and how that company might operate, and the company on the right and how it might operate are just totally different. They're like unrecognizable from each other. Okay, so what's next? There's two more. First made this slide in twenty nineteen and I did it at UX London, I was trying to convince the UX industry, my background is UX originally, so was trying to get the UX industry to stop navel gazing, kind of get your **** together, stop complaining about a seat on the table,
mature, grow up a bit as an industry, so a bit of a rant. It didn't work by the way, the US industry is still ******, but I tried. I kind of showed people this and was telling them we made all this **** up.
People have to realise we made all this up over the course of the last ten, fifteen years and if you make it up you can just delete it. You can just stop doing it. You don't have to do any of this stuff. You can just stop doing it, which is a great segue to these jobs.
We have a PM, a designer, an engineer, these people typically work in a triad and as companies grow they kind of add a different team, and here's a third team, and when these teams start to kind of aggregate people typically create groups. Those groups are led by a triad too.
You can add in research and data science and analytics and stuff in here too, is like different types of roles and marketing, but this idea of a triad is very, very common standard practice in technology. A PM and a designer and an engineering team, and that's how we build software, then you can have multiple groups and those groups typically roll up to a VP and the VP might roll up to the executive team, something like that, very, very common, very standard.
However, in a world where you have a short term strategy because there's nothing else makes sense, in a world where the product is changing rapidly, in a world where product execution is uncertain, you have to ask does any of this make sense anymore?
Is this the best way to build software? At Intercom we resoundingly concluded absolutely not. It's way too slow, laborious, complex, way too slow. So we blew the entire thing up. This happened maybe six to nine months ago, we blew the entire thing up which is a kind of a risky, crazy decision to make.
We blew it up and we said let's go back to first principles. How should we build software in this age? What do we need to do? We need to move fast, we need to work on the most important thing, and we need to follow the work rather than like follow the role, do the thing your job is supposed to say on paper, just follow the work, do the thing we think needs to be done.
So we now work in work streams. The work stream has a clear outcome, very, very clear outcome, work streams end, so at some point in time the work streams over, we did the thing we set out to do, could we ship a product, it could be update something, test a new architecture, different types of things.
So work streams have a very clear outcome. We have a DRI, a directly responsible individual. The DRI runs the work stream, they're the single accountable person and they have a very, very small team, very small, three, four people, sometimes smaller, very, very small.
We assemble the teams on the fly, some teams have PMs, some don't, some have one designer, some have three, some have none, we assemble the teams on the fly based on what we need. Every work stream works differently, we are not trying to like create some master process here where like they all have to be the same.
Have probably OCD when it comes to things should work in a structured way, you have to abandon that. These all work differently, people are learning on the fly, and the process for building software for the work streams is whatever you want. Do whatever you think you should.
Just make sure you're moving fast, you're exploring the space, we're testing with customers, we're learning if it works as fast as possible, so just do whatever you think we should do. So that's how we work. It's going really well. They really work. We're building software faster than I've ever seen in a very long time.
Like I said, Finna is our fastest growing product ever by a long way. It's hugely successful. DRIs especially work. When you have DRIs and really great brilliant DRIs, cut through all the crap, all the decision making ********, they're like get everyone together, consensus culture, everything, you cut through all of that.
That's how we do it, but there's a big giant caveat. One is it's not sustainable. We can't work like this forever, we kind of know we can't work like this forever. We don't know what's next, like how will Intercom work? I know that we need to change work streams over the course of the summer, maybe a lot of it will stay the same, but I'd stand there probably in December and give you a new talk about how we're working now.
I don't know what we're going to do next, I think it's going to be some kind of hybrid of the old way and the new way, because as you keep building all of this new software really, really fast, you have to maintain it, and to maintain it you need ownership.
The work streams don't have ownership. We have a different set of people and different little processes for closing bugs and things like that and keeping the quality high, but we know we need to maintain it. So we are going to have some kind of hybrid probably in the new world, but we really believe you've got to overcorrect if you really, really want to succeed in this new world.
So triads, PM design engineering, groups who own product areas and have direct ownership of that, gone. Workstreams that end is how we build software now. So one last thing, you might ask why did you bet the whole company on AI? You ripped up, you're a very successful company, why did you rip it all up?
What we did is pretty radical and here's why, you can kind of tell like with my AI speech earlier maybe we believed we didn't have a choice, I don't think you have a choice either by the way, you can ask me about that at the round table later.
We're a thirteen year old company, fourteen year old company maybe. People would say we're an incumbent, people in our industry in customer service say you are an incumbent company, a big company, a thousand person company, you're here a long time. So we're trying to reinvent ourselves, we're trying to make sure that not only do we survive this new era, but that we win.
We think we can win our category, we could think that Fin can be the best AI agent for customer service in the world. So we're trying to win the category, so we have to reinvent ourselves. But this is a thing we're like deeply, deeply paranoid about and you should be too.
Our competitors, we don't think about other incumbents all that much, Zendesk, Salesforce, don't really think about them. Our competitors are young, they're like a year old, they're AI native, they're not reinventing themselves, they're not talking about triads and whatever, they're AI native. They don't have to change to be AI first.
They're unencumbered by history. They don't have any history, they don't care. They don't care about the UX industry in twenty nineteen, they don't care, they're probably in school. They're wonderfully naive. They're wonderfully naive like we were back in the day. And if you're me, that is incredibly scary.
That's an incredibly scary proposition because these people are fast, hungry, ambitious, they don't care about all the ******** that we have to drag around with us. So that's why we did it. Our next competitors might not be these companies, they probably won't be, they could be anyone.
Know, when you think back to like nineteen ninety nine, early two thousand's and look at huge companies like now, Google were not the only search engine. There was Yahoo, Ask Jeeves, Butler, AltaVista, there were many, many search engines, twenty search engines, but Google won for very specific reasons, but it wasn't clear at all who was competing with who and who the winner was going to be, and that's where we are now.
Again, it's nineteen ninety nine, our competitors might be young companies doing customer service, our competitors might be OpenAI or Intropic, companies that we partner with today. They might go up the stock and start building customer service products. Maybe companies like Intercom who are application companies need to go down the stock and building models.
No one knows. No one knows how this is going to play out. So we're deeply paranoid and that's why we changed everything. So this is the new world, this is the new way that we work. I'd highly encourage you to do something like this.
You can decide how radical you want it to be, and then here is my closing advice, one last thing to take away. The best ways of working in the AI era do not exist yet. For word and intercom, these are not the best ways.
This isn't like the new way that's going to certainly be the best way forevermore. We're at the very early stage at a S curve. The best ways don't exist yet. The new types of roles, new types of triads, relationships, they don't exist yet.
And so the only way you can make sure that you win in this new era is to invent the stuff yourself. You can wait. You can wait for other companies to work out that you should have a triad or there's a thing called UX, you can wait or you can do it yourself.
So we're going to do it ourselves, we think we have to because otherwise we fall behind, we have no chance of winning the category. So that's it, thank you so much. If you want, I have slides, you can get the slides, just mail me.
If you have questions come to the round table or you can mail me too and I'll send them on. Thank you.
