---
title: "The Thing You Tested Isn't The Thing That's Running feat. Rick Clark - [Dev]olution Podcast - Coder"
description: "Listen to the [Dev]olution Podcast from the team at Coder."
image: "https://i.ytimg.com/vi/0ktzsmmuTdY/hqdefault.jpg"
canonical: "https://coder.com/podcasts/devolution/the-thing-you-tested-isnt-the-thing-thats-running-feat-rick-clark"
---

[Back to podcast list](https://coder.com/podcasts/devolution)

# The Thing You Tested Isn't The Thing That's Running feat. Rick Clark

![Global Head of Cloud Advisory, UST](https://coder.com/_next/image?url=https%3A%2F%2Fwww.datocms-assets.com%2F19109%2F1788308008-rick-clark.png%3Ffit%3Dcrop%26fm%3Dwebp%26h%3D400%26w%3D400&w=2048&q=75)Rick ClarkAug 19 2026

[YouTube](https://youtu.be/0ktzsmmuTdY)[Spotify](https://open.spotify.com/episode/4YVS6Kj5sPeRIoMURtOvjp?si=pgfg7oQARG6zn4Rp3P-dxw)[Apple Podcasts](https://podcasts.apple.com/us/podcast/the-thing-you-tested-isnt-the-thing-thats-running/id1839475248?i=1000784427424)

![](https://coder.com/_next/image?url=%2F_next%2Fstatic%2Fmedia%2Fdevolution-masthead-bg.af392453.jpg&w=2048&q=75&dpl=dpl_6wsypvurgW3Wca4tyrJ6aVYnLrcb)

![](https://i.ytimg.com/vi/0ktzsmmuTdY/maxresdefault.jpg)

## Transcript

_This has been generated by AI and optimized by a human._

**Rick Clark (00:00:00):** A bug in an Agentic system isn't a defect waiting to be found. It's an emergence property of a live system. It didn't exist before. The bug doesn't exist till it happens.

**Nicky Pike (00:00:13):** This is \[Dev\]olution, bringing development back to speed, back to focus, back to freedom. I'm Nicky Pike. Okay, so everyone's talking about agents making developers faster. More velocity, more throughput, ship more code with fewer people. That's the whole conversation right now. But here's what nobody's talking about. What happens when those agents stop writing the software and start being the software running live in production? Because for 15 years, we've been signing a piece of paper that says the thing that we built is the thing that's running. We hand that to the auditors, we hand that to the regulators. Our whole house sits on it. So what happens when you drop an agent in there and the thing running in production makes a different decision every single time you ask it? Is the contract we've been signing for 15 years still true? Or have we been quietly lying to ourselves and just gotten away with it so far?

The guy with me today says that the SDLC, as we know it, does not survive this. So let's find out how the hell we got there. My guest today is Rick Clark, who leads cloud and AI advisory at UST, where he moves Fortune 500 CTOs from AI experiments to govern production grade adoption. His job is the part where someone has to be accountable for what the system actually does. This makes him exactly the right person to answer the question I just asked. He spent his whole career walking enterprises through the waves that broke the open source cloud platform. And he says Agentic is the next one, only faster and higher stakes than any before. Rick, welcome to the \[Dev\]olution, my friend.

**Rick Clark (00:01:43):** Thanks Nicky. It's great to be here.

**Nicky Pike (00:01:45):** Excellent. So we're just going to go ahead and jump into this. You have been at the front of basically every wave that's reshaped enterprise tech. Who bought two server, OpenStack, the Four Opens, and Pivotal Cloud Foundry World, and the platforms that gave teams a golden path from code to production. When you look back, is there a single moment where you went, "Oh my God, this changes everything." And do you think Agentic is going to give you that same feeling?

**Rick Clark (00:02:10):** Well, a single moment. So there's one particular moment. So I worked at a company called Geo, which was a greenfield telco startup in India. And we actually scaled from zero to a hundred million users in 170 days. The goal was 100 days, so we missed our target a little bit. But this was 2016. We started building it in 2013. What I realized then is that this wouldn't have been hard five years before. This would've been impossible. With cloud technology that we were using, which was a version of OpenStack, we were able to scale something. The levels that would've been completely impossible before. To me, that made me realize, wow, this changes everything that I could even do this. I wouldn't have known how to do this before. And now anyone could actually do this. Everything was open, everything was out there, everyone could do this.

So it kind of combined open source and cloud. Now, especially Agentic in production, non-deterministic production, it is a major change, maybe bigger than all of these. But the problem I see with it is we had a decade to learn cloud. We had decades to figure out open source.

We had so much time. Now we're seeing this compressed in 18 months. Now there's very little time. And at the beginning, I was actually very worried that we're going to mess this up and then we're going to throw the baby out with the bath water. I'm not worried about that anymore. I'm worried about the bath water that I'm sitting in. So I think that this isn't going away. The investment in it isn't going away, but we're still. I'm really worried that we're not building the right things or looking at it the right way, that we're letting technology outrun our ability to govern and manage it.

**Nicky Pike (00:03:58):** Yeah. Well, now you brought up something there. You brought up the open community. And for those that don't know, you helped or you wrote the Four Opens, right? That was the framework that still kind of runs the OpenStack world today. You got open source, open design, open development and open community. That was you saying 15 years ago that calling something open is basically meaningless unless there's a real structure underneath it that kind of holds it all together. Is this the same instinct that you're bringing to AI now, that the freedom that we love only works if we've got something underneath actually enforcing the rules?

**Rick Clark (00:04:30):** Yeah, that's actually a good way to think about it. So back when I did that, when I wrote those Four Opens, which literally I can't believe they've survived this long. It must have been 2009, 2010. That's

**Nicky Pike (00:04:40):** Good stuff, Rick, that tells you.

**Rick Clark (00:04:43):** There were lots of things being called open. Open APIs was open. So we wanted to define this is what openness actually means to us. And from what I understand, lots of. It's still used in the open info, which is now merged with Linux, the Linux Foundation, and it's used by over 600 open source projects. But openness wasn't a vibe. There was stuff, structural things underneath that we could say these things have to be here. I think that's similar. We need structure. We need things that can be managed in a centralized way and not pushed out to whatever team wants to do them at 2:00 AM in the morning. We need that structure or it's just fluff we're talking about. There's nothing real. We need the structure.

**Nicky Pike (00:05:26):** I absolutely agree with you. I mean, especially when we're talking about something like AI, which I know we're going to get into here in a little bit, but when you have something that can literally do things that we wouldn't have even though about two years ago, being able to have structure and rules to that, how it's going to act, I think is absolutely critical here. And the one thing that really stuns me about all of this, Rick, is you wrote the Four Opens, you've done all this stuff, but your degree is actually in anthropology. Do you think that studying people changes how you see a room full of engineers that are arguing about architecture and right now arguing about AI?

**Rick Clark (00:06:01):** First of all, it helps me be less pedantic. There's not one way of doing things. Different cultures discovered various ways to do things. Those are some of the differences. So I don't go in with technology being my religion, but also that the things that generally cause problems, and lots of people say this, it's almost never the technology, right? It's almost always culture and the things that go around with it, the organization. So being able to go in and maybe see things from a little level up and not just focus on the technology, I think helps me in those situations. I'm not an archeologist like I plan to be. I'm pretty happy applying it in the world of technology.

**Nicky Pike (00:06:43):** I think we're seeing that when we look at some of the debates that are going on in LinkedIn and across the internet right now. I mean, we watched very rapidly people saying AI is absolutely horrible. Nobody should touch it. Then we saw, well, it's coming for all our jobs. Now we're seeing some people starting to adopt it. I mean, we're seeing exactly what you're talking about there, where the culture or the ideas, they immediately kind of reject anything new. Then they slowly start picking them up. Where do you think we are in that cycle right now? Are we still a community full of detractors and people that don't think it's right? Or do you think that we're actually starting to pick this up? Because enterprises, they're kind of going full bore when it comes to AI.

**Rick Clark (00:07:22):** Yeah. Yeah. So this is not following the normal hype cycle trough of disillusionment. There's too much money and it's moving too fast. I think that we are going to have some slight trough of disillusionment. We're going to have some dropdown where some people become disillusioned just because they're being pressured. The technology's being pressured to show value and return on value so quickly when the technology is changing so fast and is immature. I don't think we're seeing the normal hype life cycle. This isn't going anywhere. It's also complex because it's become a political issue as well. People don't want AI data centers in their back. So now people's opinions are being influenced by things other than just the technology. I'm not sure where we are. It's basically what I'm saying because this isn't normal, but it's not going away. This is somewhere in our future. This is going to exist.

Just what are we going to make of it? How are we going to make sure that this is the technology that helps humanity or it just helps some people get richer? I don't know. The jury's out on that.

**Nicky Pike (00:08:28):** Oh, I think this is just like the dot-com, man. This is something that's going to literally change our lives. How? I don't think anybody knows yet. We're still following along. But I guess that kind of brings us to the challenge on the table, Rick. When we look, there was a 2026 TrueFoundry survey that included 200 enterprise AI leaders across the world who were already running agents in live production. And they found that 76% of them lacked unified logging across their models and their workflows, which means that most of these companies were shipping AI into production that they couldn't fully govern, they couldn't trace, or they couldn't defend in an audit. So the question I want to put forward to you, Rick, is this. We didn't just wake up here. You've been making the point that production has been sliding towards non-deterministic for roughly 15 years.

We've got microservices, feature flags, all of that. Did Agentic break the model or is it finally just kind of ripping the cover off of something that was already broken and we've just been patching over and putting our finger in the dyke on this thing?

**Rick Clark (00:09:26):** Yeah. I mean, it has been slowly drifting. I like that analogy, putting our finger in the dyke, because that's a human. And I think that's what's been happening. So we've been sliding to non-determinism. You mentioned there's auto scaling, feature flags, microservices. There's all sorts of ways we have been changing that over the last 15 years, but there's always a human with their finger in the dyke. There's always a human saying, "Don't look behind the curtain. Everything's fine." They make that true. They make it deterministic or they make it appear enough like it's deterministic that it doesn't break our operating model.

**Nicky Pike (00:10:06):** I kind of want to go back to that stat for a second. So 76% of the enterprise leaders that are running an agents in production, they can't fully trace what those agents are doing. Does that number surprise you or does it sound about right? It kind of surprised me.

**Rick Clark (00:10:19):** It surprises me because I think it's 100%. Yeah. Yeah, absolutely. I mean, what are they tracing? A trace requires you to know the shape of a system at the beginning. And if you don't know the shape, what is a trace? I think that we don't have the language or the model to govern agentic in production. And there are lots of people, vendors out there selling things, but I don't think any of them do it. So yeah, I think that if you're running agents in production now, unless you've got a crack squad that developed some agentic platform yourself that is doing things that no one else knows about, then no. You don't really know what your agents are doing in production. You can't know.

**Nicky Pike (00:11:03):** So we've got 76% that you think are being truthful and saying, "We have no idea." You think that other 24%, do you think they're actually lying or are they just kind of kidding themselves?

**Rick Clark (00:11:12):** They're not lying. I mean, it's interesting. I was at an AI dinner recently with a bunch of CIOs and all the CIOs when they were asked all said how great they were doing with AI. No problems here. We're on target. Until one person admitted that they were having a problem, then all of a sudden they all were open and honest about their problems. I think there's two things that's going on. No one wants to publicly say that they're not where they think they should be or where their board thinks they should be or where their CTO thinks they should be. And the other is if you're at a CTO/CIO level, you're operating on information that you're getting from your team who also probably telling you everything's fine. We got everything fine. Don't worry, we have guardrails. Don't worry we have evals. Everything's fine. And they're believing that.

And I think that's probably the most likely is that they believe it. They believe, hey, we can see everything because they don't know what they can't see. If you can't see it, you can't see it. So yeah, I think it's much closer to 100% if not exactly 100%.

**Nicky Pike (00:12:16):** That's an interesting phenomenon. And I mean, you work with a lot of C-suite guys. We do a lot of round tables in which we try to figure out these problems. And I've seen the same thing that you do, but this is why we closed door those. Because when you start talking to C-suite, it's either they're not getting full information from their team or they've even. A lot of places may have marketing and restrictions around what they can say because this is not only affecting the profitability of their company, but it's affecting how stockholders are seeing their company. And if they make the wrong statement, that could have some pretty dramatic effects on stock prices, which we've seen when we look at a lot of the companies today. So I do think it's interesting, but if you can close door and you actually get them talking, you know this because you talk to a lot of C-suite guys, getting those ideas out there is how we fix these problems.

Kind of covering them up and beating around the bush doesn't do us any good.

**Rick Clark (00:13:03):** Yeah, especially in that kind of forum you're talking about that you do is they've got other CIOs and CTOs there and they can help each other. And it is not useful to not tell the truth in those situations. But it's exactly what you say. They don't want to look bad. They don't want to damage their company. Those are great workshops that you run. Those are useful for people.

**Nicky Pike (00:13:25):** Well, I appreciate that. Well, let's go ahead.

**Rick Clark (00:13:28):** I've been to one, so I'm saying that from knowledge. Yes.

**Nicky Pike (00:13:31):** This is 100% true. We've had Rick in these and hey, spot alert, if you want to come to one of our round tables, you might see Rick in the future if I have anything to say about it. But this kind of leads us into the problem that we're talking about here, Rick. So there are three things that we've always kind of believed about production. And these are the same ones that you say kind of die the second that an agent shows up. Those things are that behavior comes from the code. A bug is a defect against the spec that we should be able to know in advance. And the thing that we tested is the thing that we deployed. How does each one of these fall when we start looking at this in terms of AI?

**Rick Clark (00:14:08):** Well, the behavior comes from the code. That falls because the behavior's not in the code anymore. It's in the model weights, it's in the prompt, it's in the tools, it's in the context. And those can change every single time. And tool selection, that will change every single time, even between milliseconds. So you can no longer look at what you wrote or even the tools available to the agent and know how it's going to behave in the future. And a bug is supposed to be a defect against a knowable spec, right? But there is no spec to defect against. It's same prop, different outputs, all of them valid. A bug in an Agentic system isn't a defect waiting to be found. It's an emergence property of a live system. It didn't exist before. That's my point is that the bug doesn't exist till it happens.

**Nicky Pike (00:15:03):** Well, and it may not even be a bug the next time. No,

**Rick Clark (00:15:05):** No. In fact, it could just be that one incident. So it is an emergent property of a live environment. And then the thing that you tested is, these are all sort of flavor of the same thing, but they're looked at differently by different groups. So the thing that you tested is the thing that you deployed. That falls because the artifact won't hold still. Model endpoints drift slightly, retrieval changes faster than code. The test you wrote yesterday is testing a different system than the one running today. So you can't say anything about, "Hey, I tested this. It worked in test."

That has nothing to do with what it does in production. You tested it at one moment in time with one test, with one set of tool choices and data and model weights.

**Nicky Pike (00:15:50):** Well, and I think there's a line there that we need to draw because this is one of the places where you and I kind of actually push on each other. When we're talking most places when they're talking AI right now, they're talking about it in terms of developer productivity. This is where agents write the software. And I would argue that when agents and AI are writing the software, that's still code at the end of the day. So I think a lot of those old guarantees that we were just talking about hold true. But where you kind of get different is when you say that the agents are the software that's actually running in production, that's where the floor drops out. So what actually changes and where do you think that distinction really matters? Because a lot of people out there I think are kind of blurring those two together when we're talking about agent assistant and agent generation versus agents being the production application.

**Rick Clark (00:16:35):** Yeah. And they're completely different problems. First of all, I think I agree completely that while there are intricacies, that if an agent is writing code for you, it is writing deterministic code probably, if that's what you have it doing. There's other problems like, well, how do I know it wrote what I intended for it to write? There are other problems there, but that's very different from this makes production non-deterministic, which is what agents in production do. And we have based everything we do, all of our processes, all of our audits, all governance in production around the fact that it is supposed to be deterministic. It is not supposed to be non-deterministic. And we kind of want to close our eyes and pretend like that's not happening. And I think that's why people are conflating the two. I got used to agents, they're writing code, they're doing searches for doing this.

What's the difference for them running in production? And I think we're just not thinking it through enough. I don't think that anyone's intentionally hiding it. I just think that we are getting used to agents. So now we're applying them other places without thinking about how we are going to govern them or what that means for the environment. All

**Nicky Pike (00:17:44):** This started when you wrote something on LinkedIn, which really kind of got us talking about this, which was the start of this interview. And we'll make sure that we link that LinkedIn post in the show notes. But within that LinkedIn article, you kind of made the point that production isn't really a place. It's a promise. And that promise has compliance and audit and security scanning all wrapped around it. So when we start talking about an agent in production that can make a different call from one millisecond to the next, which of those promises break first? I mean, who's the first person that's going to get the phone call because the agent's never doing the same thing twice or it's doing a change every time we ask it a different question? Or the same question, I should say.

**Rick Clark (00:18:24):** I would say audit breaks first. The audit promise breaks first. And the audit is what is attested to. So the thing we reviewed, that's the thing that's running. I signed my name to this as the CFO or as an executive of the company. That's what all compliance wraps around. It's what security scan certify. It's what we hand to a regulator. An agent that composes its behavior at runtime makes that all false the moment it's deployed, not eventually immediately. And the first call goes to whoever signed on that dotted line. Whoever said, "This is what we have in production. This is how it's going to run. I signed this auditor. I attest to this." That's the person who gets the first call. And

**Nicky Pike (00:19:07):** That seems interesting to me because if you look out there, we're seeing people put AI in practically everything. I would argue they're shoehornet in places where it doesn't even matter. But what you just said about audit billing first, that seems like a huge issue. I mean, for all companies, but especially for those that are hellly regulated, financial services or public sector or things of that nature. So how are they looking at putting agents into production when you think that audit fails first? How

**Rick Clark (00:19:35):** Are they looking at it or how should they be looking at it?

**Nicky Pike (00:19:37):** I'd say answer both.

**Rick Clark (00:19:39):** I think we're pretending. We're doing a little ostrich head in the sand thing. We're putting them into production. I think what they should be doing is isolating it at first, coming up with new operating models, isolating it to things like if you have a chatbot that answers questions, that's probably not a regulated system that says highly audited as something else. So let's say your online banking is different and it's treated differently than your high ask me for help chatbot. So there are places where you can put things and test things, but we also need to fix the operating model so that we can correctly manage and operate things that are probabilistic, non-deterministic and production. And until we do that, we shouldn't be putting those things anywhere that is audited. That's my opinion.

**Nicky Pike (00:20:30):** Well, even when we look at like the chatbot, because I completely get what you're saying, and I can't remember exactly which airline it was, so I'm not going to make one up. But there was an airline that actually got sued because their chatbot gave wrong information to a flyer that ended up costing them some money, costing them a plane ticket. And I think the courts came back and they ruled, "Well, no. Hey, Mr. Airline, you own the chatbot. If it gives the wrong information, you're still liable for it. That's a production issue as well." Again, non-deterministic, we can't really guarantee what an agent's going to say even in a chatbot. Is that going to be a concern going forward or is that just something that we're actually starting to be able to pare down a little bit? I

**Rick Clark (00:21:09):** Actually don't think so because that same non-deterministic bug exists in human agents. This is something that if you called their consumer helpline or their line to book a ticket and they gave you bad information, the company could be held to it the same way. So I think what this does is when it's done in a chatbot, we have some way to debug what happened rather than he said, she said. Hopefully. If they've gathered all the. If they could say, "This is why this happened." If they're correctly observing, if they have modern observability in their agentic system, hopefully they'll be able to see why it happened. Legally, I think it's a non-issue because it's exactly the same as humans. The more interesting question would be, what if things don't make mistakes? How does that change everything else?

**Nicky Pike (00:21:54):** If everything's right, what does that mean for us as well?

**Rick Clark (00:21:57):** Yeah. I mean, I hate calling customer service and being wrong.

**Nicky Pike (00:22:02):** Okay. So if you're just tuning in, let me catch you up on what Rick just told me, because it's the kind of thing that should make every CTO sit up a little bit straighter. We put a number on the table. 76% of enterprises running agents in production can't fully trace what those agents are doing. And Rick looked at that number and he said it was wrong. It's low. He thinks it's closer to 100%. And the only people who would argue are the ones who can't see what they can't see. Then he said the thing that really got me. When an agent goes sideways in production, the first promise that breaks is the audit. The signature on the dotted line, and the first phone call goes to whoever signed it. So here's where we're headed after the break. If your test can't tell you what production does anymore, what replaces them?

Rick's got an answer. He calls it the behavioral envelope. And it starts with knowing an agent off the same way that you would know that a friend is off. You get into why our old observability stack is probably lying to you. And Rick tells the story of the company with 57 secrets managers and what happens when you point AI at that kind of chaos? If you run anything in production, this next part is the part you can't skip. Stay with us. Let's talk about how we actually solve this. So the SDLC has been the backbone of enterprise software for decades. After everything that we've just unpacked here, what do you think is the actual answer? I mean, is the production contract still good? Or is this something that you actually think we kind of need to tear up and just start over again if AI is going to be at the forefront of this?

**Rick Clark (00:23:30):** I wouldn't say tear it up. And you're always going to have deterministic software that's running too. And you still have to build things. I think that what you end up doing is what we need to do is stop pretending that all these pre-production checks are making us safe and we can shrink those down to maybe we're testing intent. Maybe we're running our evals on agents. We're ensuring the guardrails work. But compliance has to migrate from artifact attestations collected out of your CI/CD to behavioral evidence collected out of production. So it has to change, but I wouldn't say that we're tearing it up. But certainly the compliance part, the attestation part, that has to change. At least for

Agentic systems. I mean, we're always going to have deterministic systems. There's just things that need to be deterministic. The future isn't all agentic AI.

**Nicky Pike (00:24:26):** Well, I mean, so you brought up the pre-production portion of that there. If the test that we do can't promise us what production does anymore, production itself, it seems becomes the only source of truth there. And I think we need to break that down because what does that actually mean? You mentioned the behavioral aspects of this. What does it mean, Rick, to run on a behavioral envelope instead of the green check mark that we're used to from CI?

**Rick Clark (00:24:51):** Yeah. Well, the green check mark certifies your binary at a point in time. This artifact, this test suite, this moment, it's a gate. The whole model is that you pass the gate and then you're done. The envelope, a behavioral envelope, it abandons that binary pass/fail. It's a boundary of expected behavior asserted continuously against live traffic. So what tools are the agent allowed to reach for? What does this decision pattern normally look like? Is it behaving normally? How far can it wander from normal? And how do we define that before someone gets told or it's stopped? A gate can be passed. An envelope, you're either in it or you're out of it. Verification stops being a gate you pass and becomes a property you maintain. Verification is a property, not a gate. So you have to constantly monitor that property. You

**Nicky Pike (00:25:43):** Made the statement that you don't need a brain scan to know that a friend is off, right? You can just kind of tell that they're not right. And I think this applies to agents as well. How do we build that same instinct for an agent in production? I mean, this isn't something that we can just put on a dashboard. So are we kidding ourselves by even kind of thinking that we can tell when agents are being quote unquote right or wrong? Well,

**Rick Clark (00:26:06):** Let's never ever underestimate the need for people to look at dashboards with green check marks, first of all. And you can put it on the dashboard. You would just call it behavioral fingerprinting and have the percentage or a checker or something. But you know that your friend is off because you have accumulated years of baseline. You know they're normal. If a friend is all of a sudden acting funny, you might sense that they have a relationship problem at home or something else is going on. You're not reading one vital sign. You don't even know exactly why you feel that way. But it's a pattern and you're pattern matching. And this is outside of his normal patterns or her normal patterns, whatever the case may be. So we have to do that same sort of thing. So for an agent, we have to capture traces that are decision trees.

You don't know the shape of it ahead of time. You have to trace this fan out graph of a decision tree, not just request paths. There's new signal classes, confidence, behavioral drift alongside the normal ones like latency errors, envelope SLOs next to your latency SLOs, continuous verification instead of point in time certification. And that's not vibes, it's instrumentation and it's real and it's buildable. Doesn't exist yet, but it's buildable. And yes, I'm fuzzing over a lot of how we do this, but if we can do it in our brains, if we can know that a friend is off. Another thing I like to use is the weather. I can predict the weather, but I can know the weather if I stick my head outside and see if it's raining. Every second it changes. The weather can change every second. So the only way to really know is to experience.

And so experiencing the behavior in production is the only way we'll know. But we have to have some model that says this is the behavioral envelope. We have to define what that means. We have to say if it acts out of this so many times or if it goes too far, then we need to stop. We need to reign it in. Maybe we need to do something. But that's all stuff I think the industry has to figure out. And we're putting this stuff in production and we haven't figured out how to monitor it basically. We don't know how to observe this stuff well. It's getting better and hotel has all the primitives there, but the industry is just catching up and it's already in production. It's a scary time. We're moving very, very fast.

**Nicky Pike (00:28:30):** I want to go back to that analogy about your friend's not right. I get that, but to your own point, we've been around the friend for a long time. We know what the expectation is. When we're talking about agents, this is somebody we just yet. We don't have that normalization. We don't have that baseline. So we're a practitioner, a CTO that's listening to us right now. How do you ship that? How do you know what normal looks like? Are you saying that we just have to put it in production or test and let it run for a while so we establish it? Or is there a better way to do this for them? Well,

**Rick Clark (00:29:02):** Kind of. I mean, kind of saying we have to put it in production to test it, but we don't have to take outputs from it. So this is where I say we need a new operating model. So everyone does their blue greens, their canary deploys, right? Well, now you maybe need to put an agent in production and you feed it a mirror of live traffic and you watch what it does. Then you canary it in real and you watch what it does. So yet you have to gather that. You have got to gather that behavioral evidence. And I think this is where our SDLC has to change. This is where the way we deploy has to change. If we're deploying a change in an agent, that has to be, we have to record some behavior. And how much behavior? As much as we need. And we have to figure out what that means.

That is not me.

**Nicky Pike (00:29:49):** That makes my brain hurt when we start looking at all the. And I've used this term before. We're seeing a multiverse in technology right now where there's infinite number of variabilities. So God, I hope somebody Comes in that can understand this a little bit better.

**Rick Clark (00:30:02):** Yeah. There are people that live their whole life for these hard math problems and figuring these things out, so someone will figure it out.

**Nicky Pike (00:30:07):** Going back to your observability statement, that's the one that kind of worries me because you made the statement that traces don't work the way that they used to because a trace used to have that shape that we could predict, but now it's a fan out. It's that multiverse, right? It's models, it's prompts, it's tool choices. And that's a lot for us to keep up with. I mean, if the honest answer is to keep up with that wealth of information and all the variability, it feels like we'll need agents to tell us what other agents are doing. I mean, doesn't that mean we're kind of just adding in another non-deterministic thing to explain the first non-deterministic thing that we're trying to do? I mean, where does this actually bottom out in your mind?

**Rick Clark (00:30:45):** That is exactly true that we are adding a non-deterministic thing, but I think that's where you end up. I think you have agents flagging things as they're happening and then there's some threshold and then a human looks at it. But I think that you're talking about massive amounts of data and there's no way to parse that except through an agent, right?

Except the humans aren't going to be able to do it. We're going to have to restrain the restrainer. We're going to have to make sure that the agent that is watching the agent behaves properly. It's funny, agentic is the problem, but agentic also solves the problem in a lot of cases.

**Nicky Pike (00:31:24):** Is this the first time that you've kind of seen that where, like you said, agentic is the problem, but agentic's also the answer to the problem. I mean, have you seen this shape anywhere else in the technology since you've been around?

**Rick Clark (00:31:35):** No, I haven't. It is normally unusual that the technology solves the problem the technology creates. I haven't seen that before. But I'm glad we have that ability at least because I don't know how we would solve it otherwise. We're talking about an amount of data. Humans can go in and ask questions. And if you look at what Honeycomb has done with their agents and their system, you can go in and ask questions and you'll be able to ask behavioral questions. But that's not the same as catching it and verifying that it's constantly doing the thing it was supposed to do. And in order to do that, let's say it was Geo and they're at 500 million users now. Could you imagine the traffic they get? No one's watching all those users, making sure that the agents that they're talking to are doing the right thing.

It has got to be other agents unless you're at a scale so small that you shouldn't be listening to this podcast.

**Nicky Pike (00:32:25):** I think we need to bring that down to something a little bit more concrete. I mean, you sit in C-suite offices all day long. If a CTO is watching this and they can stand up exactly one new instrument this quarter to address the agentic thing, what's the single thing that you would tell them that they need to capture or measure first? What's the very first stepping point for them?

**Rick Clark (00:32:45):** It's difficult to choose a single first, but it has to be, based on the definition of first, there has to be one. I would say observability. It's hard. I want to say multiple things. I really do because I think there's multiple things you need. But if you don't have a modern observability stack, unless you ask novel questions of production, then you're not even close to being able to deploy AI. AI written code, let alone agentic applications running in production.

So if you can't see it, you shouldn't run it. So a modern observability stack, and that's not the three pillars of old. Logs, traces, your dashboards, all that. It is having wide rich events that you don't have to predict in advance the questions you're going to have. You can ask it novel questions. And Honeycomb, who I advise and work with, they do a great job at this. If there are multiple companies doing this that can do this at scale and do it fast enough, talk to one of them. Don't just let your old monitoring vendor relabel themselves observability and you think that you actually have observability because you probably don't.

**Nicky Pike (00:33:56):** I do agree with you. I think observability has got to be the first place because you can't govern, you can't control, you can't put security around something that you don't know what it's doing. And in the world of AI, we're talking about a very different type of observability. This isn't what we've seen with humans where we can log, "Hey, a human accessed this on this date." Now we've also got to answer the question, why? What prompted it to go access this? What did it do with that information to make the decision that it made? And that's a very different observability. And I think that's the main measure there. You got to know what you're looking at and be able to not necessarily predict, but look and see what it's doing before you can ever put any type of guardrails around it.

**Rick Clark (00:34:35):** The way that those tools are used today in normal non-deterministic or deterministic production is quite often they're after the fact. You're not managing live production, but when you have agentic in production, you have to do this live. I mean, sure, you want to be able to see things so that you can go back, but your agents have to be able to see things. So if you have agents that are making sure that are guarding your behavioral envelope, that are making sure these things don't go crazy, start hallucinating, it needs to see what's happening. And if you can't see what's happening, it can't see what's happening. So it's the same observability platform that should be feeding those agents. And if you don't have that, then you shouldn't have agents in production yet. Maybe as a chatbot, but you shouldn't have agents doing serious things in production unless you can actually observe them.

**Nicky Pike (00:35:28):** Yeah, I agree. And I mean, like you said, there's so much more to look at. How do we sandbox the agents? How do we make sure the agents are not accessing or exfiltrating things that we don't want them to? And if I was a CTO, I'm listening to this, I'm starting to sweat bullets because there's a lot of companies out there that got agents in production or about to be, and kind of the platform decisions are already behind them. What is the first thing you're going to tell them? We know observability. That should be something that they look at at the very start. But what can they change on a Monday morning that has nothing to do with the tooling that they already bought? What should they be looking at when they're even starting to consider bringing AI into production?

**Rick Clark (00:36:07):** They should look at their development processes. First of all, how they're developing, how they're measuring them. They should look at architectural oversight. So one of the things that's happened over the last 15 years as we've shifted everything left is that enterprise architecture has become rare or non-existent. And we did that because we thought, "Hey, this gets in the way of developing productivity." We shifted everything left. We shifted all decisions out to the edges. But that doesn't work with AI because AI shouldn't be making those decisions. You now need to have some method of AI knowing what is right and what is wrong. You have to teach it or it's going to just start doing wrong things. If you measure development based on lines of code and deploys, I'll tell you, AI is going to look great. It can write a lot of lines really fast and it could deploy a thousand times a day.

So look at my DORA metrics, man. I'm great. You can't look at things that way anymore. So sort of rethinking how you measure development, not about lines of code, looking at how do we know we have a coherent architecture? What are the rules? What are the rules that agents need to follow when they write code? That stuff needs to be codified and in a way that AI can get to it and verify it. I mean, you've played with AI. It can do some crazy things if you don't tell it what to do. It will just do something. It will try to do what you say, even if it doesn't have the right tools to do it.

**Nicky Pike (00:37:42):** That's a great point because yeah, it has all this capability, but you're right. Now this is changing. I'm starting to see this in a lot of the frontier models now, but it doesn't have the sense to kind of stop and go like, "Hey, what did you actually mean by that question? Or what did you mean by that directive?" It's just going to say, "Okay, you gave me a directive. I'm going to go do it. And if you leave me a lot of open ground, I'll figure it out." Now talking about the architecture, you actually brought this up in one of the round tables. You told this story about a company that had 57 secret managers in production. And you said that they did this because they let the developer do what was ever easiest for them before a deadline. But when AI shows up, it's going to read all that chaos' truth and it's going to do what AI does.

It amplifies it. So this goes back to your point about centralization. What do you think has to be centralized? And what's the line between the guardrails that work and that actually hold versus rules that an agent's just going to work its way around, Rick?

**Rick Clark (00:38:38):** So you're asking me a very tricky question there at the end.

**Nicky Pike (00:38:43):** That's what I'm here for, my friend. I don't give you easy questions. We're counting on you to be able to answer this stuff for us.

**Rick Clark (00:38:49):** So this is when I imagine the future is that I imagine your SDLC, what you call CI/CD, is now a series of agents. And one of them is an enterprise architecture agent. And you're going to declare what you want. The enterprise architecture agent is going to say, these are the things that you should do for that. If nothing matches, then it'll be escalated to a human. That will create everything you need, assign the rules, create your guardrails. I imagine that an agent's going to do it, but a specific agent where that agent's intelligence is owned by a centralized enterprise architecture or whatever you decide to call it team that says, "Hey, there are these three use cases we have for secrets managers. If you need a globally accessible secrets manager, then this is what you use. This service is what you use. If you're just storing something locally, if this is highly secure, there could be maybe two or three different use cases.

And it knows what those rules are. And if it doesn't know a rule, it's going to kick it out and then a human is going to be involved. I don't think we're to the place where AI is going to be coming up with novel architectural rules. I think the humans have to be the guard there, and they have to make sure that this agent is enforcing those rules that they've created inside of the CI CD pipeline.

**Nicky Pike (00:40:15):** I had a conversation with Luca Galante who leads platform engineering, arguably the largest community of platform engineers out there. And there's this belief that a lot of companies out there have platforms and they're like, "Hey, we can just add on to accommodate the AI that we plan on bringing in." But that's not true, right? Because your 57 secret manager's point I think highlights this specifically. You have to rethink how your platform's going to work here because we talk about this with developer productivity all the time. AI amplifies whatever you give it. If you give it that kind of chaos, it's only going to multiply that by a hundred X. You've got to go back to your point, make good architecture decisions and really think through how you want AI to work here. I mean, what would you tell people that are kind of looking at this and going, "I think we're good.

We've got a platform. Let's just shoot for the moon and see what happens." I

**Rick Clark (00:41:05):** Think that we are at the cusp of something major here. And if you don't want to be displaced, then you better modernize. And you can think, ah, we're fine. Everything we've been doing is fine. But I think you're endangering your company's future by doing that. I really do because it is not fine. What you hear is, we'll create the mess, but then there'll be agents, AI will just fix the mess. No, it won't. And even if you had agents that could fix the mess, how should they fix it? You have got to have coherent architecture and I just don't see it anymore. It is exceptionally rare that I go in somewhere and they have a really stringent architectural process. And I know why we got rid of that stuff. It was slow. It was review boards. It was all sorts of things that slowed everything down.

Regardless of the past, it's what we need for the future. And you need to do it. If you can't figure it out, you need to bring someone in to help you figure it out because if you don't have the architectural integrity figured out, AI is going to make your whole business unmanageable. That's my belief. This is do or die, man. If you think you're okay, it's a screen door in a submarine, man.

**Nicky Pike (00:42:18):** I love that analogy. I absolutely do. And 100%. I mean, you've got to start looking at that stuff because if you don't, that's how you end up being one of the news stories that ends up on LinkedIn or in the news about how AI did something wrong. And to that point, you kind of keep landing on this idea, Rick, that rules can't be set by the AI. They can't live out the edges. They have to be central for any of this to work. And that does cut very hard against the 15 years of the push decisions to the teams that we've all gotten used to. Do you think that the autonomy era is over or is there a version where we can get both of these things at the same time? I

**Rick Clark (00:42:55):** Think we made some mistakes over the last 15 years. I think that we made mistakes that burdened developers. Developers had to now write data and understand infrastructure. They had to understand database. They had to understand things that they never had to understand before. I think we can fix this. And they have autonomy of intent. You did not need autonomy of choosing the cloud or building your own database to hold PII. That's what we've created is that this autonomy to make decisions early on so far left makes it so that you have chaos when you go all the way right. I'm saying we should standardize everything that the developer shouldn't have to think about. A good example is there's going to be standards around PII data. You as a developer shouldn't have to know what are the regulations? How is PII stored? I shouldn't have to know how to set up a database and put permissions.

I should just be able to say I have PII and the agent can fix that. I still have autonomy as to what I store, what I write, my business logic, my intent. My job as a developer is to translate business intent into something that a technical system can make as true. You still have that, but you don't need to be choosing containers. You never needed to be doing those things. Every time you weren't writing business logic, if you're in a business unit as a developer, anytime you're not writing business logic, you're not helping the company. Your job is writing that business logic. We shouldn't have burdened them with this ever in the first place. So taking this back and giving it to agents so they can declare what they want. Don't make them figure it out. Make them just say, "I need to store PII." And then everything happens in the platform.

I think the platform has to be much more intent driven and that the developers need to be declaring things, but this does not make them less autonomous. It just makes them have less busy work. They're still writing whatever they want at the end of the day.

**Nicky Pike (00:44:58):** One of the things I find interesting is I think me and you with that conversation, we just kind of blurred the lines again. We went back and forth between agents in production back to what developers need and how they use a platform to make their code get out there. So it's understandable that a lot of people have blurring these lines. So we talked about that on the developer's side. Where does that translate to, okay, the agent now is the application. How do we look at that in a framework that we can actually secure something that, again, we have no idea what it's going to do? I won't say no idea, but we suspect that it could do different things at different times.

**Rick Clark (00:45:34):** Let's say we have an agent developing another agent. First of all, it should declare things the same way that a developer would because it shouldn't have to decide. We can't let agents decide, this is how I'm going to do security for this application today. Those things have to be standard. But we also have new things we have to figure out. If we're going to be verifying in production, that means we need the behavioral envelope. That means we need a process where this is put into production and real traffic is mirrored off onto it and then we are collecting data. Is this behaving correctly? And that is not in production yet. That is part of the process of flowing through. We're going to need new ways to do that, but that's only if it's non-deterministic, right? If it's deterministic, maybe that's a step that it skips.

I think that we should be treating who writes the code the same. It doesn't matter who they are, whether it's an agent, an agent or a person. But the pipeline it goes through is going to be a little different if an agentic probabilistic thing comes out at the end. And I think there's places in CI/CD where we can insert some of that stuff. And a bunch of smart people need to sit down and think about what does this need to look like? But what is ideal? What does good look like? Where do we want to be in five years so that we haven't made an absolute mess of everything? And do we need to start building that? And there are people working on guardrails and other things, but getting the entire picture before we start putting these things in production. I mean, especially in important things, things that matter.

I think we have to get that whole picture.

**Nicky Pike (00:47:13):** I think I just heard you right. The statement is we don't know yet. I think it's almost like you're doing a call to arms here, Rick, to say, "Hey, everybody's paying attention to how AI is going to make software development faster. It's going to make those that are non-technical come in and be able to create software and unleash all these good ideas." And where I think you're going is, okay, that's all fine and dandy, but you're already talking to some companies that are putting AI into production. You know this is fixing to punch us in the face and you're saying, "Hey, we should probably get a bunch of people, really smart people together to start figuring this out because we don't know right now."

**Rick Clark (00:47:48):** I'm specifically not coming out and saying, "Here's a problem. Guess what? I have the solution. Just buy my startup product." No, I don't have that. And I think it's more than one person who's going to come up with the ideas to fix this. It's going to be a trial by fire and we're going to figure out what not to do by the charred bodies that we see on the side of the road, which is not the way I want to do it. But yeah, that's how it's going to be.

**Nicky Pike (00:48:13):** But it's not all that different than every other technology we've seen, right? Everybody left bodies on the side of the highway. But I do agree with you. The speed at which we're seeing this is huge.

**Rick Clark (00:48:22):** So I was involved in Linux, right? And to bring Linux into a bank, Linux existed for a long time before any bank ran it. What we have now is that there is this idea that your CEO and chairman of the board has that you can save money. You can save money and affect the bottom line and affect earnings by putting AI in. Why isn't this done yet? I actually was talking to an AI leader at a company last week and he said on his first day, the CIO came to him and said, "You're six months too late." They expect results so quickly. So we are charging forward with disregard in ways that we hadn't. Even with cloud, there was people like, "What about the data?" There were naysayers for a long time with Linux, with open source, with cloud. Now we have major enterprises ready to put this stuff in production before we figured any of it out.

They're afraid not to. They're afraid they're going to fall behind. They're afraid that they're going to leave money on the table. They're afraid that a competitor is going to take market from them. So the main difference to me is that there isn't the reticence to move to this new technology. These are companies that used to be risk averse that are doing something very risky, and that's a difference in behavior that didn't used to exist. So that's how I see this as different now, is that we are just saying, "La, la, la, la, we can't hear you in charging forward because we're afraid not to."

**Nicky Pike (00:49:52):** And that literally, it does scare me. Now I'm huge in AI. I do think this is going to change the world, but what you just described is absolutely true in what we're seeing out there, which is this is a technology. We are putting MENSA level intelligence into our systems, but that intelligence has zero experience, zero self-control. And we're going full forward. When we looked at things like Kubernetes, even the hypervisors in the public cloud, people really thought about it. Well, maybe not the cloud so much. People rushed into that because they though it would save them money. But again, we took those things that took five to 10 years. We're compressing that into 18 months. And it makes me wonder, Rick, are we going to see a pile of bodies this time rather than just a few left on the side?

**Rick Clark (00:50:37):** I think it's going to be piles of companies. I think that there are companies that are going to make major mistakes. I think in 10 years, I think we'll be able to talk about, just like we'd say, remember Sears? We're going to be able to talk about companies like, "Well, what happened? They were killed in it when they mishandled AI and the move to AI." I think we're going to see lots of those companies.

**Nicky Pike (00:50:56):** A lot of self-inflicted ones. I don't disagree with you. All right, buddy, let's move into the rapid fire portion of this. I'm going to ask you some questions. Don't think too much about them. Just come out and tell me what you're thinking, what's on the top of your head. So agents in production at a bank doing something that actually moves money. Are we ready or absolutely not?

**Rick Clark (00:51:16):** I mean, only if it's in my favor. Move money into my account is fine. No, absolutely not ready. And the auditors would say that. The regulators would say that.

**Nicky Pike (00:51:23):** Yep. And I think a lot of the countries, because we're seeing a lot of regulation coming in, so I think they would agree with that. Observability or observability washing. When a vendor says the word to you, observability, what's your gut feeling on this? Are they being real or is this just rebranding to put AI on it?

**Rick Clark (00:51:38):** That depends what the company that says it. But quite often it is cloud washing. I mean observability washing, which is where that work comes from, of course. If they're selling three pillars and they've just rebranded it, observability, then that's just observability washing. If you can't ask a novel question of production and get the answer immediately, then you have monitoring and not observability. If you go to the same three people every time there's a problem, then you have monitoring and not observability. So there's a list of five things, but almost everyone does observability washing. If you don't have a super fast column in your data store that you're storing all your data in, then you don't have real observability because nothing can do it as fast as that, at least at this point in time.

**Nicky Pike (00:52:28):** All right. Well, that brings me to a follow-up and I'd love to get your opinion on this because I just thought of this as you were talking. A lot of the observability companies out there are calling themselves AI now because they put a chatbot that will look over their logs and answer questions. Is that true observability of AI? Because again, I think that observability with AI in the mix goes far beyond what we traditionally thought. So is that still observability washing if their answer to AI is, "Hey, we've now got AI bot that you can go ask questions about."

**Rick Clark (00:52:58):** That's still observability washing. It wouldn't solve all the problems. You might solve the problem where you have a novel question you can ask about production. It might solve that, but it won't solve all of them. And in the end, we need something that can actually attest to what's going on in production live. And none of those things can do that. None of those things are going to be good enough. None of those three pillar models. No.

**Nicky Pike (00:53:22):** Yeah, because I see that as a feature to help with the metrics and the logging fatigue that everybody's going through, which don't get me wrong, 100% absolutely needed. But I do not think that it actually goes through and helps with true observability of what AI and agents are doing for you out there. All right. 2: AM developer or the AI developer. Which one scares you the most in a fragile system?

**Rick Clark (00:53:45):** Oh, the AI developer. Just because of the speed. I mean, they can quickly write so much that I can't know what it's doing. Yeah, I'm much more. And a 2AM developer still has the fear of being fired or arrested and basic human morals and things like that. So yeah, I'm much more scared of the AI than I am the 2AM developer.

**Nicky Pike (00:54:06):** I'm going to add that to my list, right? At least the 2AM developer still has a conscience. AI has no conscience. He might be drunk, but if he has a conscience somewhere. Yep. At least he's got a conscience where AI doesn't. You tell it to do something, it will break every rule it can to do what you actually ask it to do. All right. One word for what we're doing right now. Racing agents into production before we figured out how to cover them.

**Rick Clark (00:54:29):** I would say it's uninsurable. If I had to say one word, it's like we can't predict what it's going to do. We can't insure it. There's no really one good word, like a bunch of bad words.

**Nicky Pike (00:54:41):** All right, buddy. Strix Halo box or NVIDIA Spark on your desk. Pick one.

**Rick Clark (00:54:46):** Okay. Well, I have got a Strix Halo box sitting right here, but cost is a part of it. Would I have a spark if I had an extra 20 grand or so to drop in one? Probably. But Strix Halo is great. I love my AMD Strix Halo box.

**Nicky Pike (00:55:02):** All right. So we'll pick up from that. All right, Strix Halo, it is. I don't know. We'll see if that causes a little bit of argument and debate in the comments when we see that, because I imagine that this is one of those things that people have a strong feeling about. All right, so predictions, my friend. Put a number on it. What do you think the year is that the artifact we tested is the artifact we buoyed stops being the promise that any serious enterprise can make with a straight face?

**Rick Clark (00:55:28):** I'm going to say 2028. And here's the reason why. It's not about when the technology changes. It's about when the auditors decide to stop accepting it and the auditors and the regulators. So I think it's 2028. I think we have a couple years to catch up. The technology's changed, so it should be now, but I think we have time to fix it.

**Nicky Pike (00:55:50):** I'm going to be honest, when you first said 2028, I was going to say that's a ballsy move when we look at how much has changed just in the last hell six months to be able to go out two years from now. But I can't disagree with you. I think what you said makes sense. I mean, things are going to continue to change. We've got to get time for enterprises to catch up and see what happens.

**Rick Clark (00:56:07):** And the law. I mean, so a lot of this stuff is laws that now don't make sense. So for the regulators to catch up, which is what almost all auditing is based on, right? It's going to take a couple of years.

**Nicky Pike (00:56:20):** Yep. And we don't know. I mean, that's another one of those points of we've got to be honest and say we don't know, which is regulations, EU regulations, US, state regulations, we don't know what kind of impact that's going to have. And I think that enterprises have to be considered in that. Take what happened with Anthropic and the government. You never know when a regulatory changes may shift the whole foundation of what you're building on. You've got to be prepared for that and be thinking about, okay, what happens if this no longer is an option to me for whatever reason? Rick, we're both kind of old guys here. We've watched this movie before. We've seen it with cloud. We've seen it with Kubernetes. Companies rush in, they get burned and they walk it back. We've talked about this. The thing that scares me about this is the speed of this.

We've seen 10 years of hard lessons just getting compressed into what, like 18 months? Because everybody's terrified of falling behind. You said it yourself, the FOMO. I mean, you live through these earlier waves up close. Does this read right to you? And who do you think is the cautionary tale here? The company that we're all going to be pointing out in the next two years in 2028, as you said.

**Rick Clark (00:57:25):** Oh, that is interesting. What company will we be pointing out? So if I look at cloud, we still don't really have cloud completely right yet. We rush. People rush to get on it. And then they found out, oh, hey, wait, this isn't cheap like I expected and didn't do any research about. Oh wait, now our workloads have changed and this isn't flexible. We're still figuring out cloud. So compressing that into 18 months, we don't have the time to actually get our heads around it properly. As far as cautionary, I think probably one of the SaaS companies is going to mishandle this. And I don't want to give names, but I think one of the SaaS companies is going to mishandle this and they'll disappear. The Sears thing we were talking about earlier. I don't know who it is yet though. I don't know who it is.

Maybe it'll be Facebook. I don't know.

**Nicky Pike (00:58:18):** It'll be interesting. I've got my popcorn sitting here. I am watching the frontier wars happen in real time. I suspect that somewhere in the not too distant future, we're going to see books written about the frontier wars, the capabilities and who comes out on top, because it's kind of fun to watch the big heads kind of poke at each other and see who's actually going to come out on top. But I think we're starting to see the emergence of some winners and losers in that. Do you agree with that?

**Rick Clark (00:58:48):** Yeah. Yeah, I think we're starting to. Yeah.

**Nicky Pike (00:58:49):** All right, buddy, this is the last one and we ask it to everyone. So you helped give the world Ubuntu Server, right? This became the backbone of the whole cloud native era. You wrote the governance framework that people still run open communities on. And now you're telling our whole entire industry that production itself has to be rethough, may have to be reinvented. After all of that, after all of those waves and all of the benefits that you provided this industry, what does it mean to you to be a coder?

**Rick Clark (00:59:18):** Coder is someone who makes promises that a machine can keep. And I think that that job was easy when the machine couldn't think for itself. I think that making promises that a machine is supposed to keep is about to get a lot harder. And that's really what we're doing at the end of the day is we're saying, "You business guy, you want this to happen? Well, I can make the machine do this." And now that's a harder thing to say, but we're still coders. Even if we're just coding intent, it doesn't matter. We're still coders.

**Nicky Pike (00:59:53):** That actually may be one of the best answers I've ever seen because your answer gives a little bit of the apocalyptic Took view of it. It also gives us a little bit of hope of what it may be. I think that may end up being a viral moment. Buddy, it has been a pleasure. Is there anything that you would like to give, any parting thoughts that you want to give before we back out of this?

**Rick Clark (01:00:12):** The only things I might have seen like I'm a little anti-AI, I am not. I just want to make sure. In fact, I'm the opposite. I don't want us to make mistakes so that people are trying to move away from AI. I want this movement, this changing of an era to happen in a reasonable way so that we actually help ourselves and help humanity and help our companies.

**Nicky Pike (01:00:39):** I love everything about that because the idea of us slowing down and thinking about this, that's not going to happen. We know that. But to your point, there's some things that we got to start getting ahead of before we're rushing like we are with AI right now. Well, man, I thank you for being on. Rick, if they want to reach out to you on LinkedIn, once we post that article that started all of this, any problems with that, with us putting your LinkedIn post out there?

**Rick Clark (01:01:02):** No, no, I'd love that. And anyone that wants to can contact me. I'm happy to talk about these things.

**Nicky Pike (01:01:07):** All right. Well, I appreciate you taking the time, Rick, and until next time.

**Rick Clark (01:01:10):** Thanks for having me.

**Nicky Pike (01:01:12):** Thank you for listening to \[Dev\]olution. If you've got something for us to decode, let me know. You can message me, Nicky Pike, on LinkedIn, or join our Discord community and drop it there. And seriously, don't forget to subscribe. You do not want to miss what's next.

Show more

## Featured speakers

![Global Head of Cloud Advisory, UST](https://coder.com/_next/image?url=https%3A%2F%2Fwww.datocms-assets.com%2F19109%2F1788308008-rick-clark.png%3Ffit%3Dcrop%26fm%3Dwebp%26h%3D400%26w%3D400&w=2048&q=75)

Rick Clark

Global Head of Cloud Advisory, UST

![Field CTO, Coder](https://coder.com/_next/image?url=https%3A%2F%2Fwww.datocms-assets.com%2F19109%2F1746230577-nicky-pike.jpg%3Ffit%3Dcrop%26fm%3Dwebp%26h%3D400%26w%3D400&w=2048&q=75)

Nicky Pike

Field CTO, Coder

## Watch more

[![ 96% of Companies Run AI Agents. Only 21% Can Control Them](https://coder.com/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2FqTRwwcx4QQI%2Fhqdefault.jpg&w=2048&q=75)](https://coder.com/podcasts/devolution/96percent-of-companies-run-ai-agents-only-21percent-can-control-them)[![ Open Weight AI Isn't the Risk You Think It Is](https://coder.com/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2FrFFbzoKGMtc%2Fhqdefault.jpg&w=2048&q=75)](https://coder.com/podcasts/devolution/open-weight-ai-isnt-the-risk-you-think-it-is)[![Why Developers Are Ditching AI and Talking to Humans Again](https://coder.com/_next/image?url=https%3A%2F%2Fi.ytimg.com%2Fvi%2F-DYZyPLovjc%2Fhqdefault.jpg&w=2048&q=75)](https://coder.com/podcasts/devolution/why-developers-are-ditching-ai-and-talking-to-humans-again)

## Additional content

Read more in these blog posts.

## Additional content

Read more in these blog posts.

[![A blog post thumbnail image with the title and small purple icon](https://coder.com/_next/image?url=https%3A%2F%2Fwww.datocms-assets.com%2F19109%2F1784830943-blog_when-anyone-can-build.png%3Ffit%3Dcrop%26fm%3Dwebp%26h%3D324%26w%3D604&w=2048&q=75)
AI • Jul 23 2026 • 3 min read

### When Anyone Can Build: Gene Kim on What Vibe Coding Breaks First

![Nicky Pike](https://coder.com/_next/image?url=https%3A%2F%2Fwww.datocms-assets.com%2F19109%2F1748473544-nicky-pike-linkedin-image.jpeg%3Ffit%3Dcrop%26fm%3Djpg%26h%3D100%26w%3D100&w=2048&q=75)Nicky Pike

Read more](https://coder.com/blog/when-anyone-can-build-gene-kim-on-what-vibe-coding-breaks-first)[![A blog post thumbnail with the title over a light background](https://coder.com/_next/image?url=https%3A%2F%2Fwww.datocms-assets.com%2F19109%2F1782244428-blog_five-months-to-patch-one-day-to-weaponize.png%3Ffit%3Dcrop%26fm%3Dwebp%26h%3D324%26w%3D604&w=2048&q=75)
Jun 23 2026 • 3 min read

### Five Months to Patch, One Day to Weaponize: CVE Remediation Has a Math Problem

![Nicky Pike](https://coder.com/_next/image?url=https%3A%2F%2Fwww.datocms-assets.com%2F19109%2F1748473544-nicky-pike-linkedin-image.jpeg%3Ffit%3Dcrop%26fm%3Djpg%26h%3D100%26w%3D100&w=2048&q=75)Nicky Pike

Read more](https://coder.com/blog/cve-remediation-dashaun-carter)[![Blog thumbnail image with the title over a dark background with a purple icon.](https://coder.com/_next/image?url=https%3A%2F%2Fwww.datocms-assets.com%2F19109%2F1781801403-blog_everyone-said-ai-would-kill-developer-community.png%3Ffit%3Dcrop%26fm%3Dwebp%26h%3D324%26w%3D604&w=2048&q=75)
AI • Jun 18 2026 • 4 min read

### Everyone Said AI Would Kill Developer Community. The Opposite is Happening.

![Nicky Pike](https://coder.com/_next/image?url=https%3A%2F%2Fwww.datocms-assets.com%2F19109%2F1748473544-nicky-pike-linkedin-image.jpeg%3Ffit%3Dcrop%26fm%3Djpg%26h%3D100%26w%3D100&w=2048&q=75)Nicky Pike

Read more](https://coder.com/blog/ai-and-developer-community-pauline-narvas)

## Subscribe

[YouTube](https://youtu.be/0ktzsmmuTdY)[Spotify](https://open.spotify.com/episode/4YVS6Kj5sPeRIoMURtOvjp?si=pgfg7oQARG6zn4Rp3P-dxw)[Apple Podcasts](https://podcasts.apple.com/us/podcast/the-thing-you-tested-isnt-the-thing-thats-running/id1839475248?i=1000784427424)