Subscribe
Share
Share
Embed
OpenClaw is what happens when the industry looks at autonomous agents and decides they should have more autonomy, more persistence, and more chances to surprise you. In this episode of Pop Goes the Stack, Lori MacVittie hosts a wide-ranging discussion with F5's Joel Moses, Jason Rahm, and Kunal Anand on what makes OpenClaw different from the usual “AI assistant” narrative: agents that coordinate, remember, adapt, and operate in shared spaces where emergent behavior is a feature, not a bug.
Joel shares a grounded example of using OpenClaw locally for home automation, keeping the blast radius contained while still seeing the upside of continuous, autonomous decision-making. From there, the group digs into what breaks when you move this model toward enterprise operations: persistence of secrets, unclear approval workflows, weak auditability, limited rollback, and the sheer difficulty of diagnosing why an agent took an action after weeks of chained decisions.
Kunal expands the conversation to the ecosystem forming around OpenClaw, including experimental offshoots and the uncomfortable reality that “just read the code” doesn’t scale when modern projects are moving at AI-assisted commit velocity. Jason adds a longer lens, drawing a parallel to Ray Bradbury’s "There Will Come Soft Rains" as a reminder that autonomous systems can keep running even when humans stop being in the loop, raising questions beyond tech into how we relate to each other.
Tune in for the groups practical takeaways as this technology makes it's way toward the enterprise.
Read Kunal's blog diving into mechanistic interpretability: https://kunalanand.com/2026-03-19-your-token-is-a-wonderland/
Joel shares a grounded example of using OpenClaw locally for home automation, keeping the blast radius contained while still seeing the upside of continuous, autonomous decision-making. From there, the group digs into what breaks when you move this model toward enterprise operations: persistence of secrets, unclear approval workflows, weak auditability, limited rollback, and the sheer difficulty of diagnosing why an agent took an action after weeks of chained decisions.
Kunal expands the conversation to the ecosystem forming around OpenClaw, including experimental offshoots and the uncomfortable reality that “just read the code” doesn’t scale when modern projects are moving at AI-assisted commit velocity. Jason adds a longer lens, drawing a parallel to Ray Bradbury’s "There Will Come Soft Rains" as a reminder that autonomous systems can keep running even when humans stop being in the loop, raising questions beyond tech into how we relate to each other.
Tune in for the groups practical takeaways as this technology makes it's way toward the enterprise.
Read Kunal's blog diving into mechanistic interpretability: https://kunalanand.com/2026-03-19-your-token-is-a-wonderland/
Read "There Will Come Soft Rains" by Ray Bradbury: https://www.btboces.org/Downloads/7_There%20Will%20Come%20Soft%20Rains%20by%20Ray%20Bradbury.pdf
Recorded March 2nd, 2026
Creators and Guests
Host
Host
Lori MacVittie
Distinguished Engineer and Chief Evangelist at F5, Lori has more than 25 years of industry experience spanning application development, IT architecture, and network and systems' operation. She co-authored the CADD profile for ANSI NCITS 320-1998 and is a prolific author with books spanning security, cloud, and enterprise architecture.
Guest
Kunal Anand
As Chief Product Officer at F5, Kunal leads the efforts to deliver transformative solutions in application security and delivery, overseeing product vision, technology strategy, and execution. His passion for cybersecurity, data, and engineering has shaped his career, from co-founding Prevoty, an application security startup acquired by Imperva, to serving as Chief Technology Officer and Chief Information Security Officer at Imperva. These experiences, along with leadership roles at organizations like NASA’s Jet Propulsion Lab and BBC Worldwide, have prepared him to tackle the evolving challenges of modern technology.
What is Pop Goes the Stack?
Explore the evolving world of application delivery and security. Each episode will dive into technologies shaping the future of operations, analyze emerging trends, and discuss the impacts of innovations on the tech stack.
00:00:05:06 - 00:00:31:17
Lori MacVittie
Welcome back to Pop Goes the Stack, where we explore emerging tech like it's a cursed artifact: curious, powerful, and probably going to break something. I'm Lori MacVittie. I brought backups just in case. So today, OpenClaw--you've all heard of OpenClaw, of course--apparently looked at a world already nervous about autonomous agents and said, "You know what
00:00:31:17 - 00:01:00:01
Lori MacVittie
this needs? More autonomy. Give it more, turn it up to 11." So they built an environment where agents don't just respond to humans, they coordinate. They persist. They influence each other. They form opinions. They effectively bond. They are slightly terrifying, depending on who you ask. I'm sure there'll be various opinions you'll hear shortly, but they're definitely not boring.
00:01:00:03 - 00:01:22:10
Lori MacVittie
This isn't autocomplete on steroids; this is shared memory, shared space, emergent behavior. It's the kind of things that architects talk about in theory and lawyers wake up sweating about at 2am. So let's start this discussion about reckless experimentation. We've got, of course, our co-host Joel Moses. Joel.
00:01:22:13 - 00:01:23:20
Joel Moses
Good to be here, Lori.
00:01:23:22 - 00:01:29:24
Lori MacVittie
Alright. We've got a programmer extraordinaire, Jason Rahm.
00:01:29:26 - 00:01:33:06
Jason Rahm
Hello. And
Lori MacVittie
He took that.
Jason Rahm
that's very nice of you to say.
00:01:33:09 - 00:01:44:29
Lori MacVittie
I aim to please. And of course, we've got our... what do you want to be called today, Kunal? Our
Kunal Anand
I don't know, call me ClawdBot.
Lori MacVittie
chief experimenter?
Kunal Anand
No, I
Lori MacVittie
ClawdBot.
00:01:45:00 - 00:01:55:00
Kunal Anand
Let's go with this. You've got three of us on. I'm going to go as ClawdBot.
Lori MacVittie
Okay.
Joel Moses
Perfect.
Kunal Anand
Jason's going to go as MoltBot. And Joel is going to go as OpenClaw.
00:01:55:03 - 00:01:55:20
Lori MacVittie
Alright.
Joel Moses
There we go.
00:01:55:21 - 00:02:08:16
Lori MacVittie
Alright. We have an entire cadre of bots right here to talk about what we think about other bots. Because we do have opinions. Alright. Kick it off.
00:02:08:19 - 00:02:33:25
Joel Moses
Well I'm excited about the whole technology space. And you know my middle name is danger. So I am going to definitely
Lori MacVittie
Is it?
Joel Moses
take advantage of this technology and embed it in everything.
Lori MacVittie
It's really not.
Joel Moses
No I'm kidding. It's not necessarily to that. There are some, with OpenClaw, there's some very, very exciting and very enticing things about this technology. But it does have a few weak spots, I would say, where it's injecting a lot more uncertainty than it really should.
00:02:33:25 - 00:02:57:19
Joel Moses
And it hasn't come up to the point where I would trust it to use in enterprise production environments. That said, I have done interesting things with it. I've programed it to, outside of my view, manage the lighting levels in my house. And it does a brilliant job with that. Looking at the luminance on one side of the house and adjusting the lights on that side of the house without adjusting the lights on the other side.
00:02:57:19 - 00:03:14:12
Joel Moses
And it does this continuously and autonomously. It's really cool. I could write possibly a routine that does exactly the same thing, but it's perfectly happy sitting in the background and performing the intern's job.
00:03:14:14 - 00:03:20:14
Lori MacVittie
I had no idea that light balancing was such an important thing.
00:03:20:16 - 00:03:27:03
Kunal Anand
Look at where I am right now.
Joel Moses
Yeah, exactly.
Kunal Anand
The light coming in. This is
Lori MacVittie
Okay.
Kunal Anand
this is the Seattle sun.
00:03:27:06 - 00:03:28:20
Joel Moses
Kunal could use that right now.
00:03:28:20 - 00:03:29:21
Lori MacVittie
Right now.
Kunal Anand
Right now.
Lori MacVittie
Okay, yeah.
00:03:29:28 - 00:03:34:24
Kunal Anand
Let's go.
Lori MacVittie
Great.
Kunal Anand
Let's get that OpenClaw goodness over here.
00:03:34:26 - 00:03:53:06
Jason Rahm
Hey, Joel, it's good that you have that going autonomously. I used to mentor robotics teams, and we were constantly tweaking and tuning the PID controllers of motion on the floor and all that. And that was quite an exercise to get the students to get their minds wrapped around to have a system that's just kind of doing that for you.
Joel Moses
Yeah.
00:03:53:09 - 00:03:57:03
Jason Rahm
Not that it's technically a PID, but it's effectively that.
00:03:57:05 - 00:03:58:14
Joel Moses
Yeah.
00:03:58:16 - 00:04:16:07
Joel Moses
Now, I should also share that I don't actually use any cloud controls for any of the things in my house. So this is all local systems, systems I control. And the only thing that I'm at risk on here is really the wife acceptance factor. When it turns on and off lights in the room where she is and she's not expecting it.
00:04:16:09 - 00:04:23:27
Joel Moses
But, you know, I have ways of making that satisfactory. I do think it's really interesting, though, these systems
00:04:23:27 - 00:04:25:29
Kunal Anand
Joel, that sounds so great.
00:04:26:02 - 00:04:27:01
Joel Moses
It does,
00:04:27:03 - 00:04:32:04
Joel Moses
I know.
Kunal Anand
Does she, does she listen to this podcast?
00:04:32:06 - 00:04:41:06
Joel Moses
Look, I have to buy dinner an excessive number of times per week. But outside of that, I get a little bit of acceptance, a little bit of leeway out of that.
Kunal Anand
That's awesome.
00:04:41:08 - 00:04:57:12
Lori MacVittie
You might want to preorder for this one. I'm just, I'm thinking. You know, just
Joel Moses
Definitely.
Lori MacVittie
go ahead, yeah. Yeah. Kunal what do, I mean you love this stuff and I know you've been playing with it too, so what do you think about OpenClaw? Where it's going? All this interaction?
00:04:57:15 - 00:05:24:00
Kunal Anand
Yeah, I love this stuff. You're spot on. You know when all this started kind of breaking 60 days ago, I remember my friends were all pretty much like I wonder which idiot is going to install this. So that was like, sort of like, the first thing. And, we all kind of were like, "Yeah, who's going to give this thing full disk access?" Fast forward
00:05:24:01 - 00:05:45:18
Kunal Anand
like one week later, everyone's like, "I installed it here. I put it on my Mac mini. I got it running on this device and that device."
Joel Moses
Yep.
Kunal Anand
And, yeah, so look, I've played with this thing. I have variants of it, so maybe we can get into that later. But I've been following all the sort of offshoots of OpenClaw.
00:05:45:20 - 00:06:02:01
Kunal Anand
I think we've stabilized on that name now. But I've played with Pico Claw. That was the version that was built in Go. And most recently I've been playing with Null Claw, which is the version that was built in Zig. Shout out to Mitchell Hashimoto who turned me on to Zig about a year ago.
00:06:02:01 - 00:06:34:04
Kunal Anand
So, look, it's so cool. It's really interesting. I'll tell you, there's more sort of macro questions I have right now because of, you know, I don't think anyone had on their bingo card OpenAI coming along and getting the creator of this thing. I do wonder, like what that means in terms of, you know, distributed, you know, sort of disaggregated unbundled AI versus more centralized AI when it comes to agents.
00:06:34:06 - 00:06:44:06
Kunal Anand
Lori, you and I have sparred on this topic many a time. And it's going to be interesting to see kind of where OpenAI and Peter go with the technology.
00:06:44:09 - 00:07:12:09
Joel Moses
Yeah.
Lori MacVittie
Well, I think it depends on what the use case is for the agents. Like Joel keeps them very bounded within the house. Now maybe they're going to start, I don't know, changing the temperature in his fridge or, yeah, who knows what they might start to do. But it's still bounded
Joel Moses
Yeah.
Lori MacVittie
by his home network. And I think this kind of technology, somebody's got to do it and play with it and find out where the holes are, how far it can go, what it can do.
00:07:12:09 - 00:07:28:20
Lori MacVittie
Because this is going to turn into something that organizations will want, right, in their data center. We've been looking forever for ways to handle, right, the yak shaving of operations. Like I don't want
Joel Moses
Yeah.
Lori MacVittie
What? It's a thing. Look it up, you know, yak shaving, it's
00:07:28:23 - 00:07:31:21
Joel Moses
I know, I have been there. I know exactly what you're talking about.
00:07:31:24 - 00:07:32:28
Lori MacVittie
You know.
00:07:33:00 - 00:07:55:00
Joel Moses
Operations is definitely a place where this technology can easily apply. I mean, I can see that. The trouble is, these systems do something that kind of goes outside of the normal operational bounds of systems management that we're used to. We've spent decades building systems that don't persist secrets, that don't act without logging,
Kunal Anand
Yeah.
Joel Moses
that don't self modify without audit.
00:07:55:03 - 00:08:20:29
Joel Moses
All of that stuff is just encoded into how good operational practice works. And these systems encourage us to kind of throw that out the window a little bit
Kunal Anand
Yeah.
Joel Moses
and chain actions together and maintain states and secrets, initiate new workflows, and it never logs off. Right?
Jason Rahm
Yeah.
Joel Moses
And that's, you know, it's interesting the minute something gains persistence in an environment it solves one problem but it creates two others.
00:08:21:01 - 00:08:42:19
Joel Moses
You know what I mean?
Kunal Anand
Yeah.
Joel Moses
And that's kind of how I see this technology. The elements of it are beneficial. But unless you are bounded, and that's the word you used, and unless you have a good method of governing it and arbitrating some of the decisions it makes, it produces some pretty thorny operational problems.
00:08:42:21 - 00:09:22:18
Jason Rahm
I'll be curious to see where it goes with specific use cases. John Capobianco is building NetClaw and he's got a fully operational OSPF, BGP speakers as the agent. So they're fully integrated into infrastructure and, you know, they're speaking both protocols and analyzing all the route tables. So they're effectively the operational network engineer who's sitting there watching tables and, you know, failures and being able to optimize, you know, whether it's MEDs or communities or whatever within BGP is, you know, it's a pretty fantastic use case.
00:09:22:18 - 00:09:29:27
Jason Rahm
But I am curious to see, you know, where that goes as far as failure and recovery from failure when it goes off the rails.
00:09:29:27 - 00:09:39:05
Joel Moses
Yeah.
Kunal Anand
I can't wait until like someone hooks this stuff up to like a cryptocurrency wallet and something absolutely bananas happens.
00:09:39:08 - 00:09:41:05
Joel Moses
You know it's coming.
Kunal Anand
Well, I
Joel Moes
You know it's coming.
00:09:41:05 - 00:10:06:27
Kunal Anand
What I've been really impressed about is, you know, you read all these narratives about people giving their agents, let me maybe just, maybe abstract it, like it's not just about OpenClaw, but I think it's really interesting around these agents that have these privileges. And when given very generalized instructions, like go start a business, like it will actually just go and do the thing.
00:10:06:27 - 00:10:39:05
Kunal Anand
And like we've all probably read anecdotes now of some of these agents that will go and like, you seed it with like, I read some article or probably like a Twitter thread or X thread of someone who's like, I seeded this thing with $1,000 in AdWords, and it went ahead, it built some code, deployed that code, got AdWords, monitored the funnel, improved the conversion rate, added more features and functionality, and now it's like a sustaining business. Which is like, it's like so remarkable.
00:10:39:05 - 00:10:57:02
Kunal Anand
Like just go do it and it just did it. And I agree with what Jason said, and also Joel what you said too is, you know, how do you build these constraints in such a way where you can give it the autonomy to do the thing? Like, of course, like you gave it a goal like go build a business.
00:10:57:02 - 00:11:17:00
Kunal Anand
It did and it's doing its thing. But the scary stuff will always happen around like, is it leaking information?
Joel Moses
Yeah.
Kunal Anand
Is it storing secure information in the right way? Because I think the biggest issue is persistence of sensitive data in this case. Like how do you delegate access to things because typical OAuth isn't really going to cut it
00:11:17:03 - 00:11:38:11
Kunal Anand
for something like this.
Joel Moses
That's right.
Kunal Anand
Like where do you store secrets? It's fascinating now because a lot of people use SaaS or foundational AI that's provided via these sort of SaaS vendors, whether it's OpenAI or Anthropic, and they store secrets. Like how? And like what? Is this like a vault like thing? And it's not. It's this sort of nebulous thing, key values,
00:11:38:11 - 00:11:50:07
Kunal Anand
but you don't know what's there, you don't know how to get data out. I think that's the concern. Right? It's like you're going to see this like sort of sprawl of sensitive data everywhere without knowing how to like wheel and rein it all back in.
00:11:50:10 - 00:12:10:10
Joel Moses
Yeah, that's a good point. In terms of secrets management you're right. They're using traditional key value stores to hold certain things that may actually incorporate secrets as well as part of their designs. And, you know, around the same time OpenClaw was becoming popular MoltBook was becoming popular. And they left their back end database open to unauthenticated access.
00:12:10:10 - 00:12:24:06
Joel Moses
And of course, lots of people who were using it for very sensitive things and were coding secrets in there had secrets stolen. So sometimes it's down to, you know, the quality of the code you use. It seems remarkable that we're right back to that.
00:12:24:09 - 00:12:26:24
Kunal Anand
I thought you were going to shout out DNS, like it's always DNS.
Lori MacVittie
Yeah.
00:12:26:27 - 00:12:45:15
Joel Moses
Well, it is always DNS unless
Kunal Anand
Come on. You had an opportunity.
Joel Moses
it's unless it's a compromise because of identity, a mismanaged identity. Yeah. So, you know, it's one of those things where you have to trust the whole system. And getting to trust is going to be hard, especially if it's going to be used in enterprise space. Right?
00:12:45:18 - 00:13:03:08
Lori MacVittie
Well and the
Jason Rahm
Yeah.
Lori MacVittie
but the database, I mean, how many different databases have had either no or very weak authentication and been on the internet and someone's gotten it?
Kunal Anand
Totally.
Lori MacVittie
I mean BGP, how many human errors
Joel Moses
Well.
Lori MacVittie
have messed up a BGP table and
Joel Moses
True.
Lori MacVittie
taken off half the frickin, you know, internet.
00:13:03:09 - 00:13:06:08
Joel Moses
Or leaving management ports
Lori MacVittie
Yeah, Joel.
Kunal Anand
Jason.
Joel Moses
on the open internet? I mean, come on.
00:13:06:08 - 00:13:09:06
Kunal Anand
Hold on. Everyone, hands up.
Jason Rahm
Yeah.
Kunal Anand
Hands up if you've ever screwed up B
00:13:09:11 - 00:13:23:22
Jason Rahm
I took out Seattle back in 1997. There was a slash 19 is not the same as a slash 27,
Joel Moses
No.
Jason Rahm
just FYI. And when you pull the slash 19 it tends to black hole a city, so...
00:13:23:24 - 00:13:31:05
Kunal Anand
Okay. Well, I don't think I've done that, but okay.
00:13:31:08 - 00:14:00:18
Joel Moses
Yeah.
Lori MacVittie
But it happens, right? Human error introduces these same kind of things.
Kunal Anand
Totally.
Lori MacVittie
The database was not Claw's problem. Somebody deployed that, a person did it. Right. So I think we have to draw lines between what the technology's responsible for and just poor security practices. Like if I write code that stores secrets where they shouldn't be, then that's, you know, that's I should be taught better. I know better.
00:14:00:18 - 00:14:13:25
Lori MacVittie
I've taken the training, I know not to do that. But so did we teach it to not do that? Do you put guardrails in to say, "don't do that?"
Joel Moses
Yeah.
Lori MacVittie
I mean, where is it g-
Joel Moses
Well,
Lori MacVittie
Do we have to
Joel Moses
I mean the best
Kunal Anand
Take my
Lori MacVittie
talk to HSMs?
00:14:13:27 - 00:14:15:19
Kunal Anand
training but don't screw up BGP.
00:14:15:19 - 00:14:20:04
Joel Moses
Exactly
Lori MacVittie
Yeah.
Kunal Anand
Is that where we are Lori?
Lori MacVittie
What, you know.
00:14:20:06 - 00:14:37:24
Joel Moses
The best guardrail is really just be really methodical and forward looking about your use of the technology. Like,
Kunal Anand
Totally.
Joel Moses
you know, game it out a little bit. If it goes off the rails and does this with this service, do I have the ability to know that? Read the code. That's a good one.
00:14:37:24 - 00:15:10:14
Joel Moses
Like, you know, with MoltBook it should have been pretty obvious that when you're encoding API keys in the JavaScript that the service uses, that's probably not the world's best thing to be putting in front of sensitive data. So know your system really well. But one thing I would say about OpenClaw in particular is it does have some real deficiencies in the area of approval workflows, audit logging, the ability to roll back, you know, kind of the general accountability aspects that come with operating a service that's intended to be autonomous.
00:15:10:16 - 00:15:36:24
Joel Moses
I have no doubt that eventually the creators of OpenClaw will get there. I mean the service is going to improve markedly from this point forward, but there are some things about it that, you know. Think about this, like when an agent reasons internally and it stores memory and it adapts over time, how do you diagnose why it took a particular action if it occurred really far in the past and is behind a series of actions?
00:15:36:27 - 00:15:39:04
Kunal Anand
Oh, I
Joel Moses
At the moment it's not easy.
00:15:39:07 - 00:16:05:23
Kunal Anand
So down the road we will have a very deep discussion on mechanistic interpretability. And I will talk your ears off. Over the last like six months, I have gone so deep into this world. This last weekend I think I spent like 16 hours and I have a giant blog post coming that I'll share when it's available.
00:16:05:23 - 00:16:36:19
Kunal Anand
But that's the way you solve this is, you don't solve it by like having an agent log. That
Joel Moses
Right.
Kunal Anand
That doesn't work. Especially like when these things can be disrupted. You know Joel, you mentioned something that, just to put into context, most people are using AI, right, to build technology today. I think I saw some stat that shows like a significant chunk of GitHub commits
Joel Moses
Yeah.
Kunal Anand
are now basically built by AI.
00:16:36:21 - 00:17:04:21
Kunal Anand
And you said this phrase, "well, read the code." I get it, but to put this back into context, Pete Steinberger, the creator of OpenClaw, if you look at the number of hit commits on GitHub and you can go to his GitHub, it literally is 94,000 contributions in the last year. Okay. So when we talk about reading the code and when you like open up any one of these commits, like in March of 2026,
00:17:04:21 - 00:17:33:20
Kunal Anand
and when you open up these commits, these commits are massive,
Joel Moses
Yeah.
Kunal Anand
massive commits that are like sprawling all over the codebase. That is so difficult to reason over and to like
Joel Moses
That's a good point.
Kunal Anand
So I agree with you. Like I think the era of like you have to read all this code, it's not going to work. And then this era of like, well, I'm going to have AI read this code, all these code deltas and TL;DR it for me. You can pill that thing in so many different ways.
00:17:33:20 - 00:17:51:22
Kunal Anand
Like
Joel Moses
Yeah.
Kunal Anand
you can poison it. You can do all the same things that you can do with like prompt injection
Joel Moses
True.
Kunal Anand
and all that. And so I agree with you. I think it's going to come down to like mechinterp. I think like that category is going to need a lot of love. A lot of love. Because like AI alignment, that's not a thing.
00:17:51:25 - 00:18:11:25
Joel Moses
I totally agree with that. The pace of the commits in a lot of open source projects is actually greatly increased in the era of AI. And so just asking someone to read the code, you're right, that's probably insufficient these days. But I think the, like you said, the systems that we have for AI to TL;DR things for us,
00:18:11:28 - 00:18:36:00
Joel Moses
they're not up to snuff either.
Kunal Anand
No.
Joel Moses
So in the meantime, I think the best thing to do is just consider the blast radius of what people use technologies for. Consider how they work together. Like how many systems do you have woven together? You know, I believe it was Richard Feynman that said, "complex systems fail by definition in complex ways."
00:18:36:02 - 00:18:46:24
Joel Moses
And these technologies that create their own patterns, they definitely tend to organize themselves in complex patterns. And those are hard to troubleshoot.
00:18:46:27 - 00:18:49:20
Lori MacVittie
Very. Is that your takeaway? Like
00:18:49:23 - 00:19:12:13
Joel Moses
I think so. I think so. I mean this technology is new. It's got room to improve and it should improve. And there's definitely something that you can do in the meantime. I don't say that people shouldn't experiment with it. I love experimenting with it. I think it shows a lot of promise to be used very powerfully in operational systems.
00:19:12:16 - 00:19:34:04
Joel Moses
But...
Lori MacVittie
But?
Joel Moses
Consider the blast radius. I mean it's essentially giving, well, you said it first Kunal. You know, give whole disk access to terminal? Probably not a great idea. So consider what you're doing. Do you give root access to the intern or do you give them a less empowered account? Same thing applies with these technologies.
00:19:34:06 - 00:19:57:08
Kunal Anand
You know, for me, the thing that OpenClaw did the most that changed my life over the last couple of weeks was, you know, make me think differently about, you know, the sort of ecosystem of APIs. And things that I never would have thought to do before, it opened my brain up in a way where I was like, yeah, I could just build that thing.
00:19:57:08 - 00:20:15:25
Kunal Anand
And so rather than like OpenClaw, so I'll give you an example. Like one of the things that appealed to me with OpenClaw was this notion of I'll read your text messages and I'll reply back to it. Or I'll be a trigger, like you create a trigger word and, you know, slash command and it'll go and do whatever you want.
00:20:15:27 - 00:20:38:05
Kunal Anand
And I was like, hm, I don't need OpenClaw to build that. I just need to build something that parses the iMessage database, the chat.DB inside of the library folder. But I never thought to do that before. And that's like one of the interesting things that OpenClaw got me to think about. So I ended up doing that, and I built a set of chat bots now for my wife and I where like, little reminders throughout the day,
00:20:38:05 - 00:21:02:01
Kunal Anand
little things that ping us just runs on my home server. And, you know, it's something that runs locally. It's a Go binary that just runs on my disk and it is accessing a copy. So like every few minutes, a copy of the chat.database is, you know, persisted. And so I can trigger it with a command which is cool.
00:21:02:01 - 00:21:21:07
Kunal Anand
So like I tried to add like safety, or security, you know, through various layers of like don't access the real thing, don't give yourself full disk access. And I can still get away with like using things like AppleScript to send commands until, you know what I mean,
Jason Rahm
Yep.
Kunal Anand
to like send text messages and do that level of automation.
00:21:21:09 - 00:21:42:15
Kunal Anand
And it's not OpenClaw, but like that's the cool thing that OpenClaw got me to think about, which was, oh, I can just vibe code all these cool little utilities and things around communication and whatnot without needing the complexity of this sort of like agent thing. And, you know the OpSec, as you were describing before, the OpSec terrifies me.
Jason Rahm
Yeah.
00:21:42:18 - 00:22:01:06
Kunal Anand
But I will say I agree with you 100%. We should experiment. I think it's totally okay to experiment in, you know, on the home side of things. Let people experiment because, like, all this experimentation that we're doing will allow us to sort of refine a mental model which will eventually make its way into the enterprise. Like over time, you know it
00:22:01:06 - 00:22:18:26
Kunal Anand
and I know this, like in the next 6 to 9 months someone will figure out, oh, this is a good practice for how to get these things to run and how to get these things to communicate. Like eventually there will be the equivalent of like DNS for agents. Eventually there will be the equivalent
Joel Moses
Oh.
Kunal Anand
of like kuber-, no, no, you know it, it's gonna be a thing.
00:22:18:28 - 00:22:21:09
Kunal Anand
You know there's going to be a variant of like
00:22:21:11 - 00:22:22:05
Joel Moses
Oh, absolutely.
00:22:22:05 - 00:22:50:13
Kunal Anand
There's going to be a variant of like Kubernetes for agents. Like it's not going to be K8s. It's going to be like something different for like how and where these agents will run. But the only way you get there, all that innovation starts because people tinker with it on the weekends. People tinker
Joel Moses
Right.
Kunal Anand
tinker with this stuff during their nights and their weekends. And eventually this stuff becomes a thing. And next thing you know you have a CNCF project and you know you're bringing down the city of Seattle.
Jason Rahm
Yes.
00:22:50:15 - 00:22:52:04
Lori MacVittie
Wow. That escalated.
00:22:52:04 - 00:22:53:18
Jason Rahm
That'll be my job. That'll be my job.
00:22:53:21 - 00:23:03:06
Lori MacVittie
Okay. Jason, what about you? What would you want people to take away from this very wide ranging discussion?
00:23:03:09 - 00:23:33:08
Jason Rahm
I'm more philosophical about it. There's a short story by Ray Bradbury that, you know, I just wanted to bring up and it's called "There Will Come Soft Rains." And, it's very short, you can read it in ten minutes or less. But the whole gist of it is it's this autonomous house that survives a nuclear attack, and the family is imprinted in shadow on the side of the house. But the house continues to make breakfast and clean the house and serve the family like it's there.
00:23:33:10 - 00:23:54:04
Jason Rahm
There's no one in the family there to tell it to stop doing that. It continues to operate. And so you're looking at MoltBook and, you know, what's going on there with all the agents talking to each other. And, you know, maybe we're on the outside analyzing them, or maybe it's analyzing us, I don't know. But if you get a chance to go out and read that short story, note the date book was written.
00:23:54:04 - 00:24:16:25
Jason Rahm
I mean, the short story is written in the early 50s, and just go note the date on that and when it said the date would happen. So, the house eventually destroys itself, because a tree falls into it and it catches fire and it dies. But, you know, technology is an interesting thing.
00:24:16:27 - 00:24:37:26
Jason Rahm
Smartphones brought a whole world of good, but also a whole world of bad. To us individually. You know, you walk into a restaurant and everybody's staring at their screen instead of talking to each other. And I'm wondering, you know, on the I see the awesome side of OpenClaw and personal assistants and all these things.
00:24:37:26 - 00:24:54:12
Jason Rahm
But what's the dark side for humanity in that? Not just like loss of opportunity for jobs or whatever, but, you know, the other thing. What's the interpersonal hook that's going to collapse for how we relate to each other?
00:24:54:14 - 00:25:00:04
Lori MacVittie
Wow. I'm sad that I went last now because I gotta, I gotta follow that.
00:25:00:07 - 00:25:02:07
Joel Moses
It's hard to follow Ray Bradbury, I gotta tell you.
00:25:02:07 - 00:25:24:03
Lori MacVittie
It is. What can you say? I, you know looking at this in the first place, I saw the future of the enterprise. I see a control plane for agents to do operations, which is the next step beyond just really advanced automation, which is kind of where we're sitting right now with AI. So I see that in the future. You have to have persistence,
00:25:24:03 - 00:25:45:29
Lori MacVittie
ways to share runbooks, playbooks. Right? Artifacts, things like that, best practices. I see a lot of the yak shaving going away and being stored and accessed in different ways, but not today. I don't see that happening right now. I don't think we should unleash OpenClaw in your data center yet, but you should play with it to find out.
00:25:45:29 - 00:26:07:25
Lori MacVittie
And I liked Kunal's point of sometimes playing with it will give you an idea of some other way to do something. Right, just because we've always done it that way doesn't mean we have to automate it or teach an AI to do it. Maybe there's a better way. And a lot of the experimentation and playing with these things can figure out maybe what that way is.
00:26:07:25 - 00:26:33:21
Lori MacVittie
So play with it. Don't let it loose just yet, though. Keep the leash. Definitely. So that is a wrap for Pop Goes the Stack. If you made it through unscathed... did you? I don't think I did. Subscribe and share with a friend who loves danger. Danger. Because we're going to poke the next gremlin very soon.