This is getting real: Welcome to the AI revolution

AI has gone from possible to inevitable to certain. Now, it’s up to us to define what comes next.

At our first-ever AI customer service summit, Pioneer, I delivered a talk about how the generative AI revolution is impacting the software industry.

TL;DR: This AI transformation is getting real. It’s not just about sprinkling some AI ✨ sparkles in your interface or adding a thin layer on top of ChatGPT. That won’t cut it anymore.

We’re actually learning how to build innovative new tools with large language models (LLMs), and, in the process, completely rethinking products, roles, and indeed our industry as a whole. We’re witnessing a transition from “Software as a Service” to “Service as Software” – instead of selling seats for humans to do a job, we’re now selling the product that does the work itself. This transformation is already in full swing, and we’re barely two years into it.

Where do we go from here? Check out the full talk to find out.

What follows is a lightly edited transcript of the talk.


When tech got weird

I’m here to take us on a kind of drunken walk through technology – where we’ve just been and where we’re going. Tech is kind of amazing. Within the last four or five years, we’ve gone through a lot. That’s the best way I can put it. We certainly feel like that internally, and I’m guessing you all do too.

Even if we just go​​ back to 2019, this was what I would call sort of the late stages of SaaS. I think, if we’re honest with ourselves, we’d kind of ran out of ideas. Of course, we didn’t admit that. Instead, we told ourselves, we discovered an amazing new technology and whether they wanted it or not, as an industry, we rammed it down our customer’s throats. I am, of course, talking about dark mode. Literally, the most exciting thing you could do in tech would be take your UI, dip it in dark, bring it up again, and be like, ta-da! Also, it costs more money. That was literally where we were.

We were in a kind of weird ambiguous place and then 2020 came along, and with it came our old friend Covid. So Covid was an interesting time in the tech industry because the first thing that happened was, obviously, the lockdown. But no sooner did we lock down than we launched the original back-to-the-office movement, where we created new little virtual offices, which were actually tabs in a browser like this where you could drag your little people around and hear each other’s conversations. This is literally the sort of stuff we were building in early Covid. And it got weirder. At one point, Mark Zuckerberg took to the stage and said the future was going to be this.

Does everyone remember this? It was mad shit, right? And you thought, in 2021, we were going to get to a real dose of normality of sober times. Wrong.

2021 opened with the unveiling of NFTs, and I don’t know if you remember this era, but people were literally buying JPEGs of rocks for hundreds of thousands, or, in some cases, millions of dollars. And this was a thing we were all doing that we thought was somehow rational and on we went. In our heart of hearts, we knew what was coming. And what was coming was 2022.

After a run that had lasted for so long, we get to the top and we’re like, “Uh oh. Oh no.” Down things went. With it, came the doom and gloom that we all know. The layoff, the closures, the down rounds, all that negative shit.

ChatGPT frenzy

And then, on November 30th, a lad called Sam posted a tweet that simply says, chat.openai.com. And this popped up.

And we didn’t know what the hell it was, but it was more fun than dealing with layoffs. So we started to play with it. We asked it things like, “What can you do?” And it was answering questions like a human would, and that’s pretty smart. And we could ask even more interesting questions.

“2023 was, in so many ways, the year of the prototype. A lot of solutions trying to find a problem to match”

On the back of a bull market and then followed by an extreme bear market after layoffs and 1.3 million rocks and all that stuff, we went into 2023, which was, in a sense, many years in one. We used this AI tech to try out so many different types of prototypes. 2023 was, in so many ways, the year of the prototype. A lot of solutions trying to find a problem to match. To give you a simple example, I don’t know if we ever needed to hear a song about a customer service conference called Pioneer, but we’re going to.

Now I’m not saying it’s good, but that’s a single-sentence prompt, right? It’s just mad, right? I don’t know if anyone ever needed to see a dead celebrity fashion show, but you can do that too. All of this mad stuff is possible now. Why did someone do this? I don’t know. What could it be used for? None of that matters right now, it’s 2023. We’re just doing shit and finding out what’s going to happen.

“Fortune 500 companies, specifically, were told they needed to be seen to be adopting AI. Just do something, buy some AI shit, we need to put it into the next press release”

And then, in a slightly more useful way, there’s other tech. OpenAI did this demo where you could say, “Sketch out a website”, and you could basically take a photo of that, send it to chatGPT, and say, “Please build me this,” and chat GPT would without question, just build out the HTML and build your website. And that is interesting. I mean, no one’s canceling their Jira subscription thinking this is the future of software development. But it was definitely interesting that this stuff was starting to be possible.

The other thing that happened a lot was that people, Fortune 500 companies, specifically, were told they needed to be seen to be adopting AI. Just do something, buy some AI shit, we need to put it into the next press release. And it produced a lot of what’s called AI slop. You can imagine how you could say something like, “Hey, we should do graphics for our new restaurant menu where we serve duck and where we serve squid.” And sure enough, you’ll get some weird graphics. This is what’s become known as AI slop, just the massive outpouring of stuff that people are not really sure what it’s useful for.

The other thing that we saw a lot of in this period was what we call a “thin wrapper” idea. It’s when some hot new product gets really popular, but then, you scratch and sniff a bit at this breakthrough, and you’re like, “It’s actually just chatGPT again, isn’t it?” And this was a very common thing.

Beyond the hype

But amidst all of this 2023 drama, something very real was happening. We were actually learning how to build with LLMs. We were learning what it would look like to build in non-deterministic ways. With a lot of these projects, you don’t even know if it’s possible when you start. We were finding out what’s an actual real problem that’s addressable and customers themselves are starting to make a space in their life for the idea that something could be generated. And towards the tail end of the year, there were some interesting pointers to where we might go.

One example I really like is homework. There’s a famous educational study by a guy called Benjamin Bloom, who basically says, in a nutshell, being taught by an expert in your field, direct one-to-one, outperforms traditional education by two standard deviations. It’s called Bloom’s 2 sigma effect. Of course, that’s not a useful finding because who the hell can afford to do that? We could never possibly do that except for now we can. We can actually expose this type of technology to people and have really, really great tuition, and it’s competing with nothing. The alternative wasn’t to have a full-time lecturer helping your students learn. So all these things started to become interesting.

“I give this prompt to a product, and it pops out a fully illustrated storybook straight out of my daughter’s imagination”

Another one I liked because I used it a bit, was interior design, right? Again, you can literally send a photo of a room that you’re considering renovating, and you can ask for different ideas for how the room could go. And you can generate lots of ideas. Now, none of these designs are perfect, and I don’t think an interior designer is about to be put out of work here, but what I can do is use it to explore a solution.

Intercom.com, our website, the way the design team worked in this is they actually used a lot of generative artists to explore the solution and then, as they converged, they hired actual artists and painters to deliver the final graphics. I think that’s a really interesting way to partner with this technology. I think you’ll see more of that.

And then, a personal use case for me: I have a good imagination, but you know what kids are like – you run out of stories very quickly. Literally, I give this prompt to a product and give it all the context it needs and it pops out a fully illustrated storybook straight out of my daughter’s imagination, which is just something that wasn’t possible. I know we used to have fill-in-the-blanks stuff when we were kids, but this is breakthrough.

That was all consumer stuff. In our B2B land, if you’re doing project management software or ticket tracking or something like that, what’s in it for us? Well, the first wave of AI that we saw was what I call the sprinkling of AI. These started to appear in all our products, and what they promised to do was take a little task that somebody had to do and simplify it or speed it up or do it for them. And that was sort of the beginning of, “Huh, this stuff might be valuable.”

From thin wrappers to Service as Software

And that brings us roughly to 2024, where we are today. And the shift that we’re seeing happen in software, but honestly in society too, is this transition from Software as a Service, where we’ve spent most of our careers, to what’s now called Service as Software. You don’t sell seats to people to put humans in those seats to use the product to do the job. You sell the product that does the job. You’re selling work now. And that’s the transition we’re going to see happen in lots of places in software. Another way of putting that is that all this AI stuff is getting real.

Now, that’s not to say that it’s smooth sailing from here onwards. It won’t be. We are probably close to the apex of the hype curve.

I dunno if you’re familiar with these hype curve things, but basically, when something new comes out, everyone thinks it’s the best thing in the world, then they decide it’s not. And then they decide, eventually, it actually is pretty cool over again. We’re somewhere up here, which means two things I think you’re going to see: a lot of people questioning its value, and a lot of hard, skeptical, but sometimes important questions asked.

“Those thin wrappers will go away; that experimental budget will dry up”

Goldman Sachs produced this report saying, “Is it just too much spend and too little benefit here?” These are legitimate questions, right? Because unthinkable amounts of money have been thrown into this space. Tens or hundreds of billions have been thrown in here. A lot of the VCs won’t be seeing that money back. That’s the nature of startups. Most of them fail and then some of them are smash hits. We should expect that. Secondly, just the death of a lot of these companies. Those thin wrappers will go away; that experimental budget will dry up. People will generally start to realize these things didn’t work, never worked, and aren’t useful. We’ll see a lot of that disappear.

The other thing we should expect is that a lot of founders who had a kind of failing company have decided to, as a last-ditch gamble, rebrand it as an AI company and see if that works. That won’t work either. A lot of this is just normal. This is what we call the sort of creative/destruction period. We’ve gone through a massive change in tech, and because of that, we should expect a lot of turbulence. It’s as if someone’s just ripped up the tablecloth and we need to reset the table entirely. And it’s not the case that just because you have an idea for a good AI product and you can ship it, you’re going to somehow be a part of this new wave. There are too many things changing. The technology capabilities are changing, and we don’t yet know what the right shape is.

This became obvious to me earlier this year. I do a little bit of angel investing, and on a Monday morning, someone pitched me a product that helps SDRs, which are sales development representatives. They’re people who effectively cold email, trying to generate business. “We help them write outbound emails.” And I was like, “Oh, that’s interesting.” I could see how that’d be useful. I needed to check if it wasn’t a thin wrapper, if it was actually something useful. But it sounded useful. On Wednesday, I was still mulling it over, and I got another pitch where they take a list of the emails they need to hit and they’ll write the emails entirely and do it all from there. And I was like, “Oh, that’s a better idea.” If this thing works, we don’t need the other thing. So, maybe that’s the actual winner in the space.

I kind of had the checkbook out ready to dance, and then on Friday, a third email came in: “Actually, we’re going to just replace your team. You just tell us what type of customers you want to reach. We’ll work out everything else. We’re going to work out how to get to them, and what their email address is. We’ll send the emails, we’ll schedule the appointments, set up the sales team, we’re going to do it all.” Huh.

Only one of these things is going to work. This thing is easier to build and might have initial success because you can put that on the market pretty quickly. This thing is a monster, but if you can do it, it’s clearly going to dominate the other two.

Rethinking products in the age of AI

This is what I mean by the shape of the solution that we’re building in AI. We have to work out if we are building a tool that helps the human do a job or a single task like summarize, generate or whatever, or if we are taking a lump of work off the humans and saying, “Hey, how about we’ll just do all that and you carry on with the other stuff?” Or maybe we are looking at org-level things and saying, “We’re just going to do all that and you don’t ever have to worry about it.”

We’re going to have to take big old massive products that we all use, whether it’s project management or expense tracking, and we’re going to have to decompose them into all of the important tasks that actually happen, and for every given task or workflow, we have to ask questions. Can this be done and then verified by humans? This is the human-in-the-loop thing. Or can it be heavily augmented, like a copilot-style thing, where it’s like, “We’ll help you do most of the work” or, “We can do it completely?” As in, you’re never going to think about this any more than you think about where your electricity comes from. Or, if we can’t do any of that, can we just apply some sprinkles of AI on a purely creative thing, and maybe we can just help by helping to brainstorm or something like that?

And then, when we’re finished doing all this, we have to ask: What do future products look like? And most importantly, what actually changes? Generally speaking, a product is a product and a business is a combination of things. It’s a product, who buys it, who uses it, how much you charge for it, etc. Every one of these will change in a post-AI world.

“The dominoes all sort of fall and the whole landscape is going to get rewritten”

What the product actually says it does – is it work or is it seats to do work? Who uses the product will change. Who’s buying the product? Often, the economic buyer of AI products might be the head of operations or the head of finance. It might not always be the traditional end user. The pricing will change. Paul spoke earlier about our outcome-based pricing. You’re selling work, so you should probably charge to the work versus charge to the seat. The performance. In the future, the questions people will be asking are, “How good is your agent at resolving conversations” or something like that. The measure is not if you have the best UI or the nicest integrations.

It’s going to shift a bit. And a lot more of these things will change. The dominoes all sort of fall and the whole landscape is going to get rewritten. This will happen everywhere, but it’s not going to happen immediately.

AI will be as big as electricity

A lot of people say AI is as big as the internet or cloud or mobile or something like that. My preferred current metaphor is it’s actually about as big as electricity, which sounds grandiose, but it’s the parallel that I like.

Electricity didn’t happen overnight. In 1879, Edison filed a patent for the light bulb. In 1881, you had the first demo of electricity here in the country. In 1882, the first power station was built here. They also did a funny thing, which has a weird parallel to today. They passed the Electricity Act, and the Electricity Act, contrary to what it might sound like, made sure that no one used electricity. It was the original bureaucracy back then. It was illegal to make, illegal to sell. In 1892, they unwound all of the bureaucracy and said, “You know what? People can make and sell electricity, it’s fine.” It was the 1900s before things like London lit up.

And then, the second-order effects. The nine-to-five that we all know and love came from the fact that electricity existed. We had to follow the sun before that. Late-night shopping and shift work, all of these things were only possible because of the rollout of electricity. That’s the second and perhaps third-order effects from one dude filing a patent for a light bulb in 1879.

We’re not even two years into AI, right? And look at what’s changing. We’re still kind of illuminating the streets and wondering what comes next. But it‘s pretty clear what’s ahead.

Let’s talk about some examples of things we’re going to see. Earlier, I showed you an example of, say, sketching a napkin and turning it into a little bit of code. This is a product called create XYZ.

You literally give it a text description of an app that you would like to build, say which model you’d like to build it with, and say “go,” and it will go and build, start to finish, an idea for a movie recommendation AI engine that actually works. You can literally go from idea to working prototype just like that. Does anyone think software engineering is going to be the same in the future? It definitely won’t be.

Another, perhaps even more boring industry – legal. Harvey is a tool that literally takes what used to take months of time to do, analysis of legal docs, see what’s compliant with what, etc., and it reduces it to minutes. From months to minutes. And does anyone think law will be the same? In medicine, tools like AI Doc literally monitor every single scan that comes into a hospital and dynamically order and prioritize them to make sure that the time spent in the hospital is spent as efficiently as possible, ultimately delivering far better outcomes.

These are all the sort of examples of how this will change. It’s not going to be live today. The future is here, but it is not evenly distributed yet.

What does it mean for customer support?

What does this mean for us? Well, it’s reasonable for us to wonder what the hell this means for our jobs. What does this mean for employment? Does my role change? Does my title change? Do I still exist?

When the spreadsheet was released in 1979, the first spreadsheet was called VisiCalc. The big complaint at the time was that the spreadsheet was going to kill jobs. And it actually did kill jobs. Not denying that. Bookkeepers and account clerks went down and have gone down further since.

But that’s not the full story. It actually made jobs, a lot more other jobs, higher-order jobs, higher-paying jobs. And that’s the sort of thing that I suspect we’ll continue to see in our world.

So, what does it mean for customer service again? Well, Paul showed earlier that we went from no automation to some automation, deterministic. We’ve been playing this game of AI bot where we went from nothing to programmable, informational, dynamic or personal, and action-based. And there’s a lot more to come in this space. There’ll be proactive bots. Bots that get to know you, that ask your name and engage with you. Bots that spot when you’re doing something wrong and make sure you do it right. All of this stuff will be obvious, right? We shouldn’t assume that just because the customer is silent, they’re happy. They could be very, very frustrated, but not bother to contact support.

The bots will break out of these windows. They’ll break out of chats. We could see video bots and voice bots, right? These things are all on the horizon. At Intercom, we have demos of pieces of this. If we don’t have animated avatars providing one-to-one customer support, it’s no longer because we can’t do it. It’s not a technical issue anymore.

“Imagine listening to literally everything, right? That’s going to change how businesses understand their customer in a really deep way”

And the other thing that I think is going to change businesses forever is the ability to analyze so much conversation. We’ve never seen what a business looks like when it can listen to every single thing, analyze everything, and from it, build models of what’s happening and where the concerns are. It used to be the case that customer feedback was a game of triage where you did your best stab at it. Imagine listening to literally everything, right? That’s going to change how businesses understand their customer in a really deep way.

Let me just wrap up with a few thoughts here. This is a time of change for customer service. I’m immensely grateful for you all to be joining us on this. One thing we all have in common here is that what we are going through in our industry is a massive change for us. There are all of these existential questions about the future of what AI-first customer support software looks like. What does customer support look like in this new world?

“For as long as we’ve had news, we’ve had news about machines taking away jobs. This story is as old as time”

Times of change are really scary and there’s a temptation I often find in our industry, but also outside, to be skeptical or cynical or hesitant or resistant. And it’s not new. Did you know, for example, the typewriter was once blamed for causing wars? Did you know that as soon as we built cars, we had a back-to-the-horse movement? Did you know that, for as long as we’ve had news, we’ve had news about machines taking away jobs?

This story is as old as time, but employment has only ever gone one way. We keep innovating and we keep employing. The tech gets better, the jobs get better, and we should believe in ourselves. Let’s be optimistic. The big changes will mean massive improvements as long as we’re willing to be learners.

I love this quote by Eric Hoffer who says that in times of change, the learners inherit the earth, while those who believe themselves learned will find themselves perfectly prepared for a world that simply no longer exists. If you came here today, you’re keen to learn. And in that regard, this next generation of customer service is yours to write.

As for us, our motivation is basically what it’s been since we started in 2011. We want to make a better internet. We want to make it a great place to be a customer. We want to make it a great place to run a business. That’s what we were passionate about. And I think when you all go back to your roles, that’s what you’re passionate about doing for your companies too. This time of change is our chance to write the next chapter of customer service. And we, at Intercom, are really excited to write that chapter with you. Thank you all for coming out today.

Intercom Blog CTA IoP Podcast Horizontal