Trajectory raises $15M from Conviction to build AI agents that continuously learn on the job

great rest of your day. Thank you so much for

you guys. It was a pleasure to be on.

We'll talk to you soon. Have a good one.

Up next, we have Ronach from Trajectories, the co-founder and CEO. We've been keeping them waiting too long. We have the whole team. Let's bring them in.

Give us the news. What happened?

How much did you raise? Tell us about the company.

Sorry for keeping you waiting. Please introduce yourselves.

Awesome. Great to meet you guys and and thanks for having us on. So good. Yeah, I can start. My name is Michael. Uh before this I was on the foundation research at deep mind build a lot of cool things in robotics and very excited about trajectory.

Amazing.

Amazing. What's up? My name's Arjun. Before this I was at Apple working on vision pro and then multimodal foundation models.

It was good hearing a lot of tea uh right before this.

Well John also is like the number one user in the entire world of the vision pro.

Yeah I watch full movies pro regularly.

All of his blog post. It's great. He's he's got some great on this. Well, I was saying this John. This John absolutely love this John. No, no, no. I'm like still a regular user of the Apple Vision Pro, which is

I'll call John on a Friday night. It's like I got to I'm in the Vision Pro. Give me a second.

We should chat. We should chat.

Now, everyone else in our office is using monitors and everything. Sometimes Arjun is just like they're working in the office. But great to meet you guys. I'm Ron. I was at Windsurf before training one. Uh and then this led to the 2 billion acquisition to deep mind. So went over there uh as you guys probably heard on the news and then left all the acquisition money on the table to start trajectory.

Okay. But you raised some new money. How much did you raise?

What's the deal?

Yeah. So we raised 15 million from conviction.

Amazing.

Fantastic.

And we're building

we're building the platform for continual learning. Uh we're working with some awesome companies. Harvey, Dakagon, Clay, Rogo, More. Um, and basically building agents that learn online. So all of that that usage, the edits, the retries, everything they're using in an AI product, using that to make AI smarter every single day.

Yeah.

Amazing.

Um, so, uh, yeah, like 15 million seems like, you know, great amount of money to start a business with, but it doesn't seem like enough to jump straight to training a frontier model at the mega scale. So I imagine your approach is a little bit more uh capital light. Uh how how how much of your approach is in the deployment phase with the companies that you're partnering with versus pure research versus uh something that's going to happen around fine-tuning something like what can you tell us about how you actually deploy a model that is capable of continual learning?

That's a great question. I would say the way we're we're thinking about this is every lab is focused on building this like smart PhD student, right? And how to make this PhD student smarter and smarter, but it's always day one of their job.

Sure.

And it's always that they're going to make the same exact mistake every time.

What we're interested in is building

like building an agent that is 10 years of experience on the job. It doesn't need to be a PhD student. It doesn't need to be like frontier level math or anything like that, but it needs to know your primitives. It needs to understand your business. and it needs to really really understand your rewards. And with that, like just doing very very light post training, okay, can get us very very far. So we have models that are 10 times smaller than the frontier models that are able to beat them.

But also, that's not what we're most excited about. What we're most excited about is that not only are we beating them today, but we're improving 1% every day. Yeah.

And that is continual learning. That's the most exciting part about continual learning.

Very cool. Is part of this too, you know, you mentioned some of the application layer companies that are your guys' customers. Is part of this helping sort of reduce their dependency on uh on the model providers and and help change maybe the the kind of leverage equation there.

Yeah, I think all of our customers have different motivations. Uh but I think really at the the the main thing they're really attracted to is this idea that like what they're doing and their bet is that product like craft product expertise domain expertise is going to win and is going to hold the taste is going to hold for the years to come. Uh and they want to find ways to actually use that and work that into a product. right now kind of their only lever is like prompt optimization, putting prompts in and we view ourselves as like a research lab in their like back pocket like giving them the tools so that they can modify the models, modify the harnesses and have them modify together so that they can do what they do best and that's make beautiful products.

Amazing. Well, congratulations on the progress.

So great to meet you guys.

Thank you so much for taking the time to come on the show.

I'm sure you'll be back on this year.

Yeah, any day now.

Fingers crossed. Yeah. Yeah.

Have a great more things coming up. Yeah. Yeah. So excited to see y'all soon. Hopefully

was posted in San Francisco.

Is that correct?

Yes. Only in San Francisco for now. More to come soon. Um we're doing our five days of trajectory right now. Check us out and so many cool stuff to come.

Amazing.

Amazing. Thanks so much. Have a great weekend for coming on. Congrats.

We'll talk to you soon.

Amazing. Talk soon. Bye

bye. Up next we have Zayn Mount Castle from Pico Grid. A good friend. Interviewed him for a video about the Gundo years ago. I'm very excited to catch up with Zayn if we have him in the waiting room. We've been keeping him waiting. Uh and so hopefully Zayn from Pico Grid will be able to join in just a minute. We'll see. Uh let us know who is in the waiting room. Um we're waiting. We'll give him just a second. Um

wish we had more time with the the Grubinator.

Yeah. So fun.

Yeah. Yeah. Yeah. We can we can go. I mean he's a professional uh professional podcaster. Dithering of course is available in the strategy bundle twice a week. Uh we also have to say a little shout out to Jackie Shapiro from the Bronx. Uh apparently she's one of our newest fans. Thank you so much for being part of the audience and tuning in. It's an honor to have you in the audience. Um

what else is going on? We have uh Jamie from Pace coming on in just a few minutes. And then

says what we've all been thinking.

Oh, what is that? Can you die from a lack of being called big dog?

It's a big question.

It's a big question.

Do you get called big dog too much?

No, not very often. I don't know. I think you put yourself in the right there's certain spaces where you might get called big dog.

Uh and you have to you have to attend. You have to open yourself to the opportunity. You have to open the door to being called big dog. Yeah. Like if everyone knows your name, they're just going to call you your name. You have to be uh you have to be walking around in a certain space.

Deep Fates. This was the post we had earlier. I fear not the man who vibe coded 50 new apps, but the man who vibe coded one new app 50 times. It's a banger post. I think he meant to say, but the man who vibe coded the same app 50 Yeah. times.

Yeah. The same new app 50 times. Something like that.

Uh this was good.

What else going on

over on Reddit? R/CFA. Would it be considered insider trading if I'm on a hunting trip or safari with a CEO of a large firm and they get mauled by hyenas and I start buying put options on their firm? Let's say I go to Tanzania on a

This does not sound hypothetical

with the CEO of a very prominent firm in the United States when we suddenly get overran by a pack of a dozen hungry hyenas. However, I and my youthfulness are of course quicker than this old man and I manage to escape and hide behind a rock while the hyenas maul him.

If instead of helping him fend them off, I open up Robin Hood and start buying put options on his company, would this be insider trading? There's absolutely no way this can be priced into the stock predicted. Oh, but the material non-public info, blah blah blah. Okay, but what if before I buy the puts, I post a video of him getting attacked on my public Instagram story of what then? Thoughts? Purely hypothetical.

Does not sound hypothetical. Well, it all depends on what the market thinks about the CEO because it's possible that the market sees as a bullish signal that