Project Lobster: Building an AI Assistant with Agency and Memory

Añadido: 2026-06-03

[00:00:01][music] >> I remember when I when I started using Claw Pilot and was like this is a really cool product.

[00:00:10]But it's different than the idea I had.

[00:00:12]The idea I had was I want this running in the cloud. I think that's the place where we realized we could both help each other out.

[00:00:19]>> I was looking at it and I also wanted to make it work in the cloud. There's all this you know, I want to have a VM in the cloud, you know, install it there, keep it running in the cloud. It doesn't feel like a good system, right?

[00:00:29]So it's like if you're if you're going to make the cloud-based version and I connect it you're going to get all the powers of Claw Pilot. I'm going to get the always on. So let's just collaborate.

[00:00:39]>> I don't know that people fully grasp what it means to have a personal assistant that is has agency and judgment.

[00:00:47]>> Yeah.

[00:00:47]>> That you trust to do certain things for you.

[00:00:51]>> Mhm.

[00:00:51]>> Without you having to babysit something or push a button, right? Like [snorts] that's sort of the the real big difference between this and I think you know, including open claw and every other AI tool out there, it requires you to sit in front of it and type some things or say some things and hit enter and wait for it to complete the task. This is totally different. I kind of think that these things are going to help us be better people in a way. And I think that if you can teach these assistants like I These are the things I aspire to do but I'm never going to get around to.

[00:01:23]>> Yeah.

[00:01:23]>> And then it can like start to fill in the gaps and it just like it helps you be better at what you're doing. I think that's exactly aligned to our company mission, you know, like if you think if if you actually read the words, empower every person organization achieve more, like that's what we want for customers and people and this technology is at the point now where you can make it happen.

[00:01:46]>> So we're actually swamped with all the you know, PRs people are making asking for help. They're starting to help each other. It's kind of the community is kind of growing around it. I'm glad we're collaborating with security, with, you know, our organizations, you know, guardrails to make sure this is properly secure for the customers.

[00:02:04]>> The stakes are pretty high here. So, like, what are some of the things that we need to do to make sure that when people go and get their assistant, you know, they feel confident that like we've done most, you know, it's like we've we've tackled all the the known unknowns that that that like live in this area.

[00:02:22]>> There's different ways to approach security, privacy, compliance, uh and generally our processes internally as well.

[00:02:31]One way is to think of it as, "Okay, this is a blocker for me to roll out."

[00:02:35]The other way to think about it is, "I invite you to hack this in so many ways."

[00:02:40]>> Yeah.

[00:02:41]>> Please be creative and find so many attack vectors.

[00:02:44]>> Yeah.

[00:02:44]>> So, we don't see this with actual customers, right? Then we have like a robust RCA process where, you know, we don't want to see this ever again, right? We have an issue, we harden it, and then we go on.

[00:02:55]And then that we apply to the new features we build as well. You know, if some feature comes in, we have a a set of people who are around us making sure they review this not because it's a blocker again, just because we want to feel com-