Introduction

It was an exciting talk and I think gives you a little bit of a flavor of what's coming on all of our heads soon. So I'm going to try to find a place for us humans in this whole equation and hopefully we can live together in harmony with our AI overlords.

Speaker Background

My name is Artem Koren. I'm the Co-Founder and Chief Product Officer of Assembly AI.

We are an AI teammate product. We've been on the market since 2019. We've been doing AI since before AI was cool.

Over 1,000 companies use our product every day. We have customers in all geographies, focused primarily in US and adjacent, but also with a very heavy presence in UK and EU, and then all over the world, across Asia, LATAM.

many different kinds of verticals use our product which is something really interesting and special in the world of AI product and tech because historically you would figure out a function and then you would build a software to that function and AI is a lot more flexible than that.

And so as a result, we have customers who are construction companies, pharmaceutical companies, legal firms, financial firms, marketing firms, IT outsourcing firms, product firms, consulting firms, you get the idea. We have a huge variety of different verticals.

We also have a pretty wide variety of company maturities. So everything from individual users to small and medium sized businesses. all the way out to very large enterprise companies.

And we also have a span of functions in those companies. So HR departments, sales, IT departments, project departments use our product.

And I think we're not special or unique in that sense. I think you're going to start to see many, many more products that are so widely matrixed and can provide value in such a dynamic and fluid way.

Past Experience

So my background started in technology as I was a CTO and later CEO of a company that did financial trading and created trading engines back in the early days when data centers meant that you had to like carry a server somewhere and rack it up and plug the wires. None of this cloud stuff for the kids.

Later I transitioned into management consulting and I spent about a decade in and out of that environment, leading mostly operational consulting focused efforts in Fortune 500 companies. So technology transformations, resource process tech implementations and basically how to take an army of people who are trying to get their work done and shift them to a new kind of paradigm that is more productive effective for their company

And then I had a brief stint at a startup called Nusana. This was way before GPT, this was around 2010. And we were trying to use AI to identify cancer in biopsies, a little bit too early for our time.

So that's my background. So basically it's technology, product, and management consulting is my mix that I bring to my company.

Evolution of AI Perception

All right, so let's talk, let's get to the goods. You know, a few years ago, there were many AI enthusiasts. There are also many more AI detractors, and there were many, many, many more AI who cares.

I think that distribution has changed significantly over the past two years. And I think it's pretty obvious that modern AI technology, mostly driven by the paradigm of LLM today, we'll see what happens in the future, is extremely powerful.

And I think we're seeing You know, it used to be like every year, but now it's every month, every week. There's a new kind of model, a new kind of approach.

These models are getting smarter and smarter and smarter, faster and faster and faster. You've probably all seen the 01 preview that came out from OpenAI. And I think that's just a hint. That's just a hint, a very early, early preview of the kinds of things that will be possible much faster, much deeper, much smarter in the very, very not too distant future, one or two years.

So these LLMs are extremely powerful, but they all have the same flaw, each and every one. The custom trade ones, the foundational ones, the image ones, the text ones, they all have the same flaw, and that is they are a huge brain that's a very powerful, smart brain that has never met you.

And so if you go up to the most powerful LLM in the world today, plug to every NVIDIA GPU, and you say, Create me a sales proposal. It has no idea what you're talking about.

Who are you? Proposal for what? In what jurisdiction? In what legal contract framework? For what product? Under what pricing? With what background? To what customer?

I mean, it needs a few thousand questions answered before it can begin to try to do a decent job at answering that question, create me a sales proposal. And this is the smartest LLM out there, and there's no exception.

The Blind Giant Problem

And inside Assembly AI, we call this the blind giant.

And so as smart as these LLMs and related are going to get, they will never have, the all of the nuances to get you exactly what you need.

In some ways, I like to think of it as like it's kind of like a Michelin restaurant problem where like anybody can make a cheeseburger, but it takes like 1000 little details to make a Michelin star cheeseburger. And those are the details that these elements don't have because they've never made it.

Okay.

This is the problem that many, many foundational AI infrastructure companies are trying to solve today, which is the deep context problem. And they've started to look at this problem with something called the context window, which I'm sure many of you have heard of, which is basically like how much stuff you could throw at the LLM at any one time and have it think about it. but the context window will only get you so far and what you really need is an LLM that would intimately know you and your business for it to be effective.

Think about it as a difference between bringing someone from the street who you know has a 200 IQ and has spent 20 years doing exactly what you need them to do, but that you've never met and they don't know your company, versus having a very smart 200 IQ consultant who lives and works with your team. The difference in what they're able to achieve will be drastic.

Assembly AI's Approach

And that's what Assembly AI aims to do. And we do that by digital twinning primarily. I'll speak to that a little bit more.

So there was a slide in the previous presentation that it's the automation versus co-pilot and the difference in productivity. And I think certainly the numbers are correct, that if you can end-to-end automate a function, the switching costs and all that, there's a big advantage. But there's a lot of humans walking around that know things.

And how, you know, one way is to create like droids that can fly and shoot lasers. And that's like very powerful. But there's also something to be said for the Iron Man suit itself.

And so why not make something that makes everyone Iron Man at what it is that they're doing, whether it's sales, marketing, legal, HR product. IT, et cetera. And augmented work intelligence is that kind of a Iron Man suit.

1The idea behind it is rooted in what we call digital twinning. And that is, first, allowing AI to learn as much about you and your work environment and your business dynamic as possible. Really have a deeply understand

what you're doing, why you're doing it, how you're doing it, who you relate to in the company, what are all your team members doing, how are those team members relating to each other, how are the teams relating to themselves, what's the organization doing, what's the strategic alignment in that organization, et cetera. So really starting to get that information. And then once it knows you intimately enough, it can start to give you things before you think you need them.

It can start to do the hard lifting for you in terms of things like document creation, et cetera. And so this is the concept and that's the, I guess you can call it the co-pilot concept. But I think it's really more than that. It's not something that's kind of like waiting for you to ask a question. It's something that is an avatar that very organically binds to you and then lets you get things done in amazing ways, much more powerfully than you could before.

Product Features

And so when we started, we started as an automated meeting attendance technology. And so we spent the last almost six years perfecting that. layer, which is the attendance layer.

And so we do a really great job of getting into meetings and following you along during your day and hanging out with you and all the meetings that you have. And the reason that we do meetings and not email and not Slack is if you reflect on what you talk about and how you talk about in meetings versus what your inbox looks like. And then you ask the question like, which one is more me and my work? I think the answer is very obvious. Same goes for Slack and other things.

With just one sentence to chat GPT, you can start to get some really useful things out of it, like just plain old chat GPT. But imagine if it's not one sentence that we have about you, but thousands and thousands and thousands of sentences all contextualized in... other people's conversations. And so we actually build thousands and thousands and thousands of data points about you as the product hangs out with you at your meetings. And there's nothing you need to do. If you invite it like a human being, it shows up like a human being on your calls. Google Meet, Microsoft Teams, Zoom, WebEx. And it just is a participant in the meeting. It just hangs out. But as it hangs out, it's learning all about you. And then it's able to do amazing things.

So what's behind kind of the product? How does it do it? So it starts with attendX, which is the attendance layer that I mentioned. It's very important to be able to very fluidly hang out with you as you do your work, specifically in meeting context. And so we make that happen.

The next thing that happens is called work streams. And so this is a technology that can take the flow of all your meeting conversations across a week and distill it down to atomic units of projects or customers or partners, effectively units that are results oriented, right? Because you're not having meetings to have meetings, you're having the meetings because you need to get something done and the meeting was necessary for that. And so this technology figures out what that for that is and starts to thread together contextually relevant information across all of the meetings that you've had over time.

The next piece of this solution is KeyMaker. So KeyMaker, now that we've attended your meetings, we've organized the information around what you're working on into projects or customers or partners or otherwise. Keymaker, knowing everything we know about you, can tell you, you know what? Now that you've had this meeting, and we know who you are and why you were there and what you're trying to do. Here are like the four, five, six, seven most impactful things to work on.

So for example, you maybe just had a meeting with a customer and in prior conversations, the customer mentioned they're looking at a competitor and maybe in this conversation, they mentioned which competitor and they're thinking about a different product. So Keymaker might pipe up and say after the meeting, you know, it'd be great if you send this customer a side-by-side comparison of your product versus the other one. So that's pretty cool, and that's already very helpful. And nobody had to say during the meeting, like, hey, we should send, right? Like, it will figure that out. And by the way, the insight it gives you, in this case, I'm assuming you're like the salesperson, but if you were like a project coordinator or a technical expert on the call, your insights would be completely different.

And then finally, how cool would it be that when it tells you like, why don't you make this side-by-side comparison, I think you all know what's coming. What would be really cool? If you can push the button and it makes it. And that's called SendBloom 2.0. We have that. That works. And so you literally have a button that says work on this. You click it and then you can ask it to create the artifact for you.

So these are kind of the three major capabilities that power our augmented worker intelligence solution. And effectively, it's the big difference of just getting something generic from ChatGPT versus getting something completely imbued with all the information about your customers, your coworkers, yourself,

um and being able to generate all kinds of various artifacts from project scope documents to contract agreements to project plans to candidate assessments for hr i mean it's it's really up to your imagination we don't have like a limited set of documents that we generate

Conclusion

So that's Assembly AI, that's augmented worker intelligence, and that's a little bit about digital twinning.

I kind of ran around a lot of different topics. I hope you found it interesting.

Thank you.