Engram emerges from stealth with continual learning AI that cuts inference costs by training models to know your world

Anyway, thank you so much for taking the time to come chat with us.

Great. Great to meet you.

Have a great rest of your day. We'll talk to you soon.

Cheers.

Have a good one. Let me tell you about Figma. Agents, meet the canvas. Your AI agents can now create and modify your Figma files with design system context. And Jack Morris from Engram is in the waiting room. He's co-founder and head of research. Jack, how are you doing? Welcome to the show.

Hi. Yeah. Uh, nice to meet you. It's great to be on the show. I was actually just watching it in another tab, so this is kind of

Here you are. Fix that.

Um,

great to meet you. uh tell us a little bit about yourself, tell us about the company. Uh you're emerging from stealth uh with a whole lot of venture capital. What's the strategy and what's the product?

Yeah, sure. Um my name is Jack. I'm a co-founder and I guess technically the head of research at Engram.

We came out of Stealth last week after eight months or so of working on our product and ideulating with our design partners. Yeah, we raised money for a bunch of VCs. The product is

is mogged mogged.

Let's hit the gong. Let's hit the gong for that.

Grateful for the opportunity, but I was hoping you would hit the gong.

Yeah, we just did a b, you know, baby. Yeah, big one.

Congratulations.

Yeah. And thanks to all of our partners and thank you so much for funding us. Um, our product is a new type of AI. So, I think we have a pretty different vision from a lot of the frontier labs which are sort of working on like one model per lab and trying to make that model smarter every month. I think there's another way to think about it which is that the model doesn't need to get smarter every month. It needs to know you better. And so we're working on like a whole different stack which is a way to train models that that train themselves to like know your world better and like adjust to the things that you say. So it's like new ways of training, new ways of running the models. I think like to give a concrete example, I assume you know you all are very tech forward. You probably have agents doing things like preparing

preparing you for the show and like giving you reports every morning. And if you actually look at what the models, the agents are doing, they're probably like reading the same files a lot to get context about what your show is and what you do. Like literally probably every night, they're probably like reading from scratch. What is TVPN and who are you two and you know who's been on the show recently

and it's

No, we're in the pre-training now. Come on, give us some credit.

Oh yeah, you are in the pre-training move so fast.

Your point 100% stance,

but yes. Yeah, I I think you're lucky because you're in the pre-training, but I think most people are not in the pre so many documents that aren't and you have to feed those in every time. Is this the solution to continual learning? Is that the correct uh buzzword for this strategy or is this a different fork in the road, a different path?

Uh I think it's the correct buzzword. I think a lot of people use the phrase to meet a bunch of they cracked it in eight months. They cracked it in eight months. It only took them eight months.

Let's go. Oh, we decided to name ourselves something different. But I think the

I think of continual learning is basically this problem of how do you keep the same model but actually update it like rewire it every single day to learn more about what you're doing and we're working on that problem.

What's the what's the sweet spot customer enterprise AI that can mean fortune 500 companies that can mean a very data inensive company. There's also whole categories of enterprises that have uh a whole host of of AI rappers and application layer companies duking it out. I'm thinking of legal, medical. Um where where do you see the product having the earliest signs of product market fit?

Yeah, I'm glad you said earliest because I think like there's there's two halves to the vision. one is the long-term vision which is that the model will get to know you better and understand everything about you kind of like a person does like your you know co-orker um and it'll be able to like generalize and do things better than the current models but I think the current customers and like the way we're finding early success is by making the models a lot cheaper because like essentially they know everything about you already and instead of reading like a hundred files to write a summary of what you need to do tomorrow they read you four files or something like that.

So, our early enterprise partners that we've been working with are Microsoft, Notion, and Harvey.

And I think they all

you guys with the sound effects, I'm like so flattered. I wasn't sure if there would be any.

Um,

they they're nice because they have these like massive

uh workspaces of context and like they're,

you know, early adopters of AI. And I think these are the places where we can like reduce costs the fastest. the soonest because the workflows really are just that repetitive.

That's great. Well, thank you so much for coming on and breaking it down. Appreciate you taking the time and have a great rest of your day.

I know you will be back on I'm going to guess two times this year. That's my guess. Two times

for sure.

We'd love to have you back and chop it up more. Have a great rest of your day.