AI Agents Will Never Replace Humans | Matt Palmer
MTS
October 24, 2026
Matt Palmer, who works on Grok Bot at SpaceXSI, discusses the architecture behind multi-agent dynamics, context management, and security sandboxing.
Matt Palmer
When I talk about AI, AI is a tool. And at SpaceXSI, we make tools for engineers and builders and creatives. So I don't think software engineers will ever go away. I don't think creatives or artists will ever go away. But what I'm seeing is that I'm able to use these tools to get more and better things done. It is really being specific about what you want, who you're building for and the outcomes that you're trying to drive. And so maybe that's more creativity, systems design, empathy for the users that you're building for. I think maybe that's the most important thing. A model doesn't really have empathy for people.
And we're back. We are live with Matt Palmer of SpaceXSI, who works on, among other things, Grok Bot, one of my favorite products. I use Grok Bot every day, seriously. Every day to help figure out the most important things in the world to talk about. So thrilled to have you in the studio, Matt, welcome. Yeah, thanks so much. And thank you to the entire MTS team. Really also amazing new studio, congratulations. Thank you. It's only gonna get better.
So, it seems to me like one of the big differences between Grok Bot and, you know, Muse, Instinct, Dot, et cetera, is you can have multiple Grok Bots. So, what exactly is the point of having, like, a ton of different Grok Bots instead of just one? Yeah, you know, this is a question we get a lot. And I think our view on this has actually changed over the past few weeks. And that's especially true with introducing something like the primary bot, which is basically a new sort of proactive bot. So if you install Grok Bot today, if you open up Grok Bot for the first time, you'll see that you have a sort of primary bot. You can name this anything that you want to. But what we found is that almost everyone had this usage pattern where they're creating a chief of staff or a single point bot that would then manage these other bots. And internally, I think we designed Grok Bot to feel like an AI companion, such that you're creating these little co-workers, if you will. And it's very fun, the design is really great. But we really love the idea of having individual bots for individual use cases. And that's something I enjoy. But I think having this one bot that sort of serves as an inbox or a catch-all is something that most people do and that we've started to build into the product. So I like the pattern of one primary bot, now functionality in the app, and then specific bots for different use cases, which then, if you're going to create team bots or try and share them with folks on your team, turns out to be pretty helpful as well. Yeah, actually, I like the multiple bots a lot. Like on DOT, for example, you only get one DOT for some reason.
I actually named my DOT, I have my DOT named Grok DOT.
What have you learned about multi-agent dynamics from having lots of Grok bots interact with one another? What have I learned about multi-agent dynamics? Well, I think it's interesting, you know, kind of seeing the AI movement over the past couple of years and thinking a lot about like context management and what it means to give specific instructions to a specific agent. One of the advantages of Grok Bot really is that first, the harness or the code surrounding the model is very lightweight, which is one of the reasons it feels so fast and responsive. But another reason that it's good to have multiple bots is because you can have them focused on sort of multiple tasks. And so if you think about context management or the information that you're giving to AI, it really helps to have a bot with a specific set of skills that then is delegated to a certain task. So, you know, for example, if you mentioned you have bots that look at, like, X. Hey, my bot looks at X, I want to teach it what I like on X, and it has access to certain skills and tools for that specific use case. And then in that, we have very narrow context, a very narrow goal, and you can think about that as kind of like a sub-agent, whereas you have maybe a primary bot, you're saying, oh, hey, can you check in with my X bot and see what's going on today? That's actually kind of a multi-agent architecture, right? You have an orchestrator that's talking to a sub-agent and then going and accessing the platform to do something specific. So it kind of aligns with AI engineering best practices in some ways. Totally. You wrote this article called Chat Is All You Need, where you talked about kind of being bearish on visual interfaces for agent-driven software. And it seems to me like it's definitely stupid if you want a fitness tracker to make a whole graphical user interface, which you just don't need.
Read the other 21 minutes of this episode here.
$15 gets you this episode and 99 more. Credits never expire. After checkout, one link brings you back here to read it.
Not what you expected? Full refund within 14 days.
Thousands of transcripts fetched by people building searchable podcast archives
Fetch transcripts with code
The same credits work from a script or an AI agent. The demo key returns a sample episode in full, no card needed:
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090Markdown with the speakers named, for your notes, your knowledge base, or anything that makes HTTP calls.
From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.
Using your own key:
curl -H "x-api-key: YOUR_KEY" \
https://spoken.md/transcripts/1000797868535