Logs Are All You Need: Rethinking Observability with AI Agents

Sherwood Callaway: [00:00:00] To, to do sort of a ge- genetic Pareto, to apply the idea of genetic Pareto to, like, an agent trajectory-

Demetrios: Yeah ...

Sherwood Callaway: or to, uh, some agentic task, what that might look like is, like, spinning up 100 versions- Exactly ... of the agent in parallel to s- with the same input

Demetrios: Golden. Thank you, sir.

Sherwood Callaway: Thank you.

Demetrios: All right, dude, where do we even start? So much to do I feel

Sherwood Callaway: like, yeah.

Demetrios: What are you trying to do? You trying to dethrone Datadog?

Sherwood Callaway: That's basically it. That's like the short version. And along the way there are another, there will be other bodies.

Demetrios: Yeah.

Sherwood Callaway: But, um- To reach

Demetrios: the final boss.

Sherwood Callaway: To reach the final boss, yeah. But the, the, I guess for the viewer, I mean, we should explain what Cazabi is. The short version is I'm working on this company called Cazabi, and, uh, Cazabi is an AI native observability platform [00:01:00] specifically built for engineering teams that want to move fast. Uh, I think there's been a lot of changes to the way that we do, uh, software development over the last year or two, and one part of that life cycle that feels distinctly not fast and slow and painful is the process of finding and fixing issues in production.

Sherwood Callaway: Mm. I felt like, you know, I'm, I have a background in observability and infrastructure and have hated observability tools for a long time. Yeah. Or had a very love-hate relationship with them. You know, as my development workflow changed over the last couple of years, uh, it very quickly became clear to me that, like, this particular part of the software development life cycle needed to be disrupted in the way that we've disrupted, like, creating new features.

Sherwood Callaway: Um- Mm-hmm ... so.

Demetrios: And it's observing agents- Mm ... or it's observing what? Like, systems, like traditional systems?

Sherwood Callaway: Yeah, that's a good question, and actually it's one of the first questions I basically get every time I talk about Cazabi. Yeah. It's like, um, like is this a [00:02:00] BrainTrust or an Arize or a LangSmith competitor or a Raindrop competitor?

Demetrios: Yeah.

Sherwood Callaway: Um-

Demetrios: Which all of those you were talking about last time we did the podcast. You were like, "Yeah, we're using Phoenix, we're using LangSmith." That's true,

Sherwood Callaway: yeah,

Demetrios: yeah. It's cool, but it's, you know, we're also using Datadog, and so you were using all of these different tools, so you're very well equipped to know what the pains are of using the tools.

Sherwood Callaway: Yeah. The sh- the official answer is that we do AI for observability, not observability for AI.

Demetrios: Oh, okay.

Sherwood Callaway: W- w- I started the observability team at Brex, um, like in 2019. Mm-hmm. And, uh, for most of my career I've been focused on DevOps infrastructure, making developers more productive, uh, site reliability, and, uh, have a ton of familiarity with Datadog, and bring it to every company that I work at.

Sherwood Callaway: But I feel like it's just really not meeting me where I am as with regards to, like, s- my software development workflow. Like, running all these coding agents in [00:03:00] parallel and, like, running background coding agents and, like, not even looking at code, just no look merging stuff.

Demetrios: That's the, "Looks good to me."

Sherwood Callaway: Yeah. It's, it's just a, we- we're building in a different way today. Yeah. And so I think that Datadog is, um...

Demetrios: But how are you doing it different then?

Sherwood Callaway: There are three main things that, uh, Cazabi does differently relative to a traditional observability platform. The first is that, uh, we don't let you look directly at your telemetry.

Demetrios: Hmm.

Sherwood Callaway: Which is controversial, uh- Obviously, you go-- the whole point of an observability platform is to, like, observe the data that you or your application is generating, uh, through dashboards, through, like, log search interfaces, flame graphs, service maps, um, all of the 20 or 40 different modules that you see on the sidebar of a, of a product like Datadog.

Sherwood Callaway: But, um I think that is all, those are all just ways for you to answer questions about how production is working. Like, is it up? Is it down? You know, what does this error mean? Like, this customer's, customer's complaining about something. [00:04:00] Why? Uh, how long have customers been affected by this? Um-

Demetrios: Severity of an alert

Sherwood Callaway: which, the severity that, you know, which commit's responsible for bringing this down. What, what are my, what should I do? Here's

Demetrios: John.

Sherwood Callaway: We think that the same way that developers don't look at their code today, they won't look at their telemetry.

Demetrios: Mm.

Sherwood Callaway: They will just ask questions of an agent that has access to that data, and the agent will tell them what's wrong.

Sherwood Callaway: It will answer all those questions for them, so.

Demetrios: Mm.

Sherwood Callaway: Um, another way of putting that is that we think that the best UX for observability is chat.

Demetrios: Mm.

Sherwood Callaway: And so, Szabi- Slack ... just gives you a, a chat interface, um, and a, an amazing Slack bot.

Demetrios: Yeah.

Sherwood Callaway: Um, that's number one.

Demetrios: Oh, that was only the first one? All right.

Demetrios: That's

Sherwood Callaway: number one.

Demetrios: Tell me number two.

Sherwood Callaway: Uh, yeah. We have a whole manifesto. So number two is, uh, equally controversial, if not more. It's this idea that logs are all you need.

Demetrios: Hmm.

Sherwood Callaway: And if you're familiar with observability, then you might have heard of something called the three pillars of observability. Does [00:05:00] that ring a bell?

Demetrios: Yeah,

Sherwood Callaway: yeah. The three pillars are metrics, logs, and traces. Okay. And for a decade or more, since the beginning of observability-

Demetrios: Lots of companies been built off the back of that ...

Sherwood Callaway: that was the premise, right? Mm-hmm. Like, the idea was that you needed all three types of telemetry to properly understand the application in production, and I just think that's not true.

Sherwood Callaway: Um-

Demetrios: Hmm ...

Sherwood Callaway: and I think it's especially not true in 2026 when we have agents. Um.

Demetrios: So then you're like, "Let's get rid of the traces and metrics and just focus on the logs."

Sherwood Callaway: That's right, yeah. Um, there's a lot of reasons why I think this is the case. I mean, one, one benefit of, of just focusing on the logs is that instrumentation is now, like, significantly simpler, right?

Sherwood Callaway: 'Cause we took actually the two things that are hard to instrument, traces and metrics, and just get rid of them. So, like- Mm-hmm ... you no longer need to know how to set up a Prometheus server or the difference between a counter and a gauge and a histogram or, and a [00:06:00] rate. And, uh, you don't have to propagate span context and make sure that, like, the trace IDs are passing through the entire, uh, uh, call stack and also across services.

Sherwood Callaway: We just get rid of all of that. You just know how to, need to know how to, like, do a console log or a print statement. Um-

Demetrios: So am I gonna be able to vibe observe? Is that the idea?

Sherwood Callaway: Yeah, that, uh, ties a little bit to our first idea, which is, like, you should just be able to ask simple and natural language questions.

Sherwood Callaway: Mm-hmm. Anyone on your team should be able to ask those kinds of questions.

Demetrios: Mm-hmm.

Sherwood Callaway: And your instrumentation experience is gonna be way easier 'cause it just requires you to add logs. Um, and you're gonna get all the benefits that you would have if you had logs, metrics, and traces-

Demetrios: Mm.

Sherwood Callaway: Because, uh, we have some tricks on the back end- Mm-hmm

Sherwood Callaway: to reconstruct metrics and traces from the logs.

Demetrios: All right.

Sherwood Callaway: Um.

Demetrios: Let's peel back the veil. Wait, that was only two.

Sherwood Callaway: That's two.

Demetrios: What's the third one since-

Sherwood Callaway: Yeah. W- ...

Demetrios: they are so spicy.

Sherwood Callaway: They are spicy. The third one is, uh, I love all [00:07:00] th- three of my children equally, but- Mm ... uh, the second one has definitely been getting the most attention.

Sherwood Callaway: I think the first two deserve a little bit more because they are quite controversial. Like, we're, we're really asking people to change the way that they do observability. Um, it's, it's a dramatic departure from how you've done it in the past. Um, and the third one is maybe the most radical change. Uh, it is this idea that traditional monitoring and monitors are dead Like, we will never...

Sherwood Callaway: We have no use for static monitors with static thresholds anymore that, uh, you know, alert you when CPU exceeds, uh, 80% of the host or, uh, when a, when a pod enters a crash loop back, back off status.

Demetrios: And why not?

Sherwood Callaway: The reason is, well, first of all, because monitors and alerts suck. It's, if the o- if one bad thing about observability is, uh, is instrumentation, then the, like, next worst thing is monitors.

Demetrios: Alert fatigue, [00:08:00] dude.

Sherwood Callaway: Alert fatigue.

Demetrios: Yeah. I learned about that term when I- ... was learning about observability tooling. Mm-hmm. And people were like, "We're gonna use AI so that we can recognize which alerts are actually useful versus not." Yeah. And that was back in, like, 2018- Yeah ... 2019.

Sherwood Callaway: People are still working on that.

Sherwood Callaway: So we take it a step further, which is we're not using AI, AI to evaluate your alerts and decide which ones are meaningful or not or enrich them. We're just using AI to generate the alerts.

Demetrios: Mm-hmm.

Sherwood Callaway: So you don't go into Cazabi and say, like, "I wanna alert on XYZ." Cazabi has access to your t- your production telemetry, specifically your logs, and has acc- acc- access to your code base, and has access to any of the other tools and context systems that you give it, give it.

Sherwood Callaway: And, uh, from there it's able to decide what is meaningful to, to you and not. Mm-hmm. And so you'll receive an alert, for example, Slack notification, that is completely non-deterministic, uh, at the discretion of the agent. [00:09:00] And it could be like, "Hey, you know, I saw this commit go out. There's... I'm seeing this error in the logs.

Sherwood Callaway: It seems like there's a problem with payment service. Uh, here are the, your recommended remedia- remediation action items. Would you like me to kick off a cursor cloud agent to fix this?"

Demetrios: Mm-hmm.

Sherwood Callaway: Um-

Demetrios: So you're taking one step further. It's like, there's this problem, I could go and try and fix it, and get kind of far maybe, or get all the way.

Sherwood Callaway: Yeah. I mean, one of the things that we draw a line at, like, one of... I guess philosophically Cazabi, like, we're focused on observability and, um, helping people find and fix issues. We draw a line at code generation.

Demetrios: Mm-hmm.

Sherwood Callaway: So we, we will not open a pull request or, or merge anything to your PR. In fact, Cazabi is, like, completely read-only system.

Sherwood Callaway: Uh, but we can initiate, for example, like if you link your cursor account or you, um, are using Claude Code with an [00:10:00] MCP server or the Suzabi CLI, uh, you can use Suzabi to generate code, uh, generate fixes.

Demetrios: And can you also create issues in Linear?

Sherwood Callaway: Yeah.

Demetrios: And do stuff in... Yeah.

Sherwood Callaway: Yeah.

Demetrios: All right.

Sherwood Callaway: Yeah. You can do that.

Sherwood Callaway: You can do that. The agent can do that at your direction. Like, you could say, "Hey, Suzabi, um, create a Linear ticket for this."

Demetrios: Mm-hmm.

Sherwood Callaway: Or you could say, "Hey, Suzabi, every time you see a new issue, create a Linear ticket for it."

Demetrios: Mm-hmm.

Sherwood Callaway: And Suzabi has memory, so it will, it will remember that preference and it will keep doing it, uh, in perpetuity afterwards.

Demetrios: Nice. Dude, there's a lot of stuff that I wanna get into, like peeking behind the veil- Yeah ... on how you're doing things and what you've learned while building it because I think you have some equally spicy takes beyond just what you're trying to do. Love the spice. Yeah, exactly. Let's talk a little bit about the MCP versus CLI paradigm.

Demetrios: Sure.

Sherwood Callaway: Yeah. '

Demetrios: Cause I think there's this simplistic view where people are like, "Ah, I don't like [00:11:00] MCP because it bloats my shit. I don't wanna use it." But you have a bit more of a nuanced take

Sherwood Callaway: You know what I think is funny is that like a couple weeks ago, we had a big banger birthday party for Claude Code.

Demetrios: Yeah.

Sherwood Callaway: And it was his Cl- it's Claude Code's one year anniversary. And I think MCP came out in like late 2024.

Demetrios: Mm-hmm.

Sherwood Callaway: I'm not sure anyone was celebrating the MCP birthday. Um, which is no shade to MCP, it's just interesting two products, uh, that are both like, were both killer apps in their own way from the same company.

Sherwood Callaway: One of them has had this like crazy celebration, the other one's not. Yeah. Um- Mm. MCP, uh, you know, and I don't keep entirely up with the protocol, so and I'm sure they're doing a lot of great work to make everything better. But, um, I guess the big, the big complaint that people had over the last six months or so was that it would blow up the context just by loading the MCP server.

Sherwood Callaway: Yeah. [00:12:00] And then Anthropic has, has done some work on patterns for addressing that. I think one of them is basically tool search.

Demetrios: Yeah. And well, yeah, and there's also the, um, what is it? The progressive disclosure.

Sherwood Callaway: Yeah. And I think that is kind of like the tool search, if I understand correctly. Yeah. Like you're...

Sherwood Callaway: Well, it's like the agent, you don't need to load in all of the tools into context right away. The agent sort of should express some kind of intention around what tools it's looking for, and then you could service the relevant ones.

Demetrios: Yeah, yeah. That's becoming a best practice where you just have a certain set of abstractions above it- Yeah

Demetrios: so that you can say, "I, I want all the tools that are related to XYZ."

Sherwood Callaway: Yeah.

Demetrios: And you get that.

Sherwood Callaway: And the same problem kind of could exist in the context of a sandbox and a CLI, but the, um, you know, imagine like the number of tools or, or of, uh, I guess programs that are at the exposal, uh, at [00:13:00] the disc- discretion of the agent within the context of a sandbox is enormous.

Demetrios: Mm-hmm.

Sherwood Callaway: It's like aux, sed, grep, cat, ls, like literally every Unix utility and other CLI that you've installed. It's a huge number of tools, and I think it's interesting that like the, it, that pattern doesn't create the same context rot-

Demetrios: Hmm ...

Sherwood Callaway: that MCP does. And I think that is because these tools are, uh, effectively baked into the-

Demetrios: They're in the

Sherwood Callaway: weights

Sherwood Callaway: into the weights.

Demetrios: Yeah.

Sherwood Callaway: So the model sort of just knows that there's probably a cat tool-

Demetrios: Mm-hmm ...

Sherwood Callaway: in this environment, and so it just like tries. Mm. And then, you know, maybe it gets like a, it gets an error where it says there's like, there's cat is not, that alias hasn't been set up or it doesn't exist in the, in this context.

Sherwood Callaway: Um, it's not in the path.

Demetrios: Mm-hmm.

Sherwood Callaway: And it's like, okay, well let's try something else.

Demetrios: I have noticed, and I don't know if you've noticed this too, that- [00:14:00] I'm very happy-go-lucky or trigger-happy when it comes to creating skills, and I'm starting to wonder if now I'm getting like- Are we headed

Sherwood Callaway: to the same path?

Sherwood Callaway: It's

Demetrios: funny ... like skills bloat, which- Yeah ... all of a sudden now I got a million different skills, and cool, it's all local. I don't have to worry about the MCP server, but any time I do something-

Sherwood Callaway: Yeah ...

Demetrios: just because the majority of the time I make my skills universal. If I really like a skill, I'm like, "Yeah, this is going in all my projects."

Sherwood Callaway: I guess one advantage of skills, and I, I do think that skills run the same risk.

Demetrios: Mm-hmm.

Sherwood Callaway: Um, because the, the way the agent uses them, it basically LS in this directory, which contains all of your skills, and then it basically starts walking the directory and reading the, the header and the markdown of each of the skills to kind of- Mm-hmm

Sherwood Callaway: get a sense of whether this is relevant. But the advantage of that is that it's only reading the header.

Demetrios: Yeah.

Sherwood Callaway: Right?

Demetrios: It's not loading the tools in.

Sherwood Callaway: It-- My understanding is that [00:15:00] it's only reading the head. It's not loading the whole markdown skill.

Demetrios: Yeah.

Sherwood Callaway: Which is there's maybe in MCP there's no way to, um, to provide a hint about what this is- Yeah

Sherwood Callaway: the way that there is with skills.

Demetrios: Hmm. Well, let's talk about sandboxes, because that's something that I what?

Sherwood Callaway: Love

Demetrios: them. The bane of your existence? No,

Sherwood Callaway: no, I, I love sandboxes. All right. Um, yeah.

Demetrios: Good. Love sandboxes. Tell me about it and how you're doing it, 'cause I know with observability it's really hard to recreate certain scenarios and- Yeah

Demetrios: recreate, like when things are failing, that's one. Uh, so that's... I guess that's a little bit more on the eval side, but you want a sandbox for that, right? It's all kind of related, yeah. Yeah, exactly. It's- Yeah,

Sherwood Callaway: because we use a sandbox for our agent, sandboxes are become relevant in our evals, right? Mm-hmm.

Sherwood Callaway: 'Cause we're, we need to run the agent in the evals, so.

Demetrios: How are you using them maybe to start, and then go into what some of the pains are, or I know you mentioned [00:16:00] how you're using Git and you're... It's kind of like a trick or a nice little hack so that you can always- Yeah ... have shared state.

Sherwood Callaway: Yeah, we'll see if by the time this comes out whether I am still- Still

Sherwood Callaway: still as bullish on Git as I am. But, uh, yeah, maybe I'll just talk a little bit about the Sasabi agent and how it works and where sandboxes play a role. Um, the Sasabi agent is, uh, let's see. We built it on, on Vercel AI SDK Workflows. Mm-hmm. Um, which have been pretty good, although there's some sneaky lock-in stuff.

Sherwood Callaway: Um, the code is very clean and there are some very ergonomic, um, parts of the SDK. And it is, uh, it has access to a Datana sandbox and a bash tool, which it runs in the sandbox. It has a handful of other tools that are directly mounted to the agent. Um-

Demetrios: Mm-hmm ...

Sherwood Callaway: I'm trying to think of an example right now, like We'll have to come back to it.

Sherwood Callaway: But- And- ... for the most part, it's operating inside of... [00:17:00] Actually, you know, here's a, here's a, a, a critical tool that- Yeah ... uh, it uses that's not on, in the s- within the sandbox, and that is, like, our LogDB query tool.

Demetrios: Okay.

Sherwood Callaway: So obviously one of the most important things that our agent does is it reads your logs, right?

Sherwood Callaway: Like, that's how we root cause things. Um, so we, we ha- we give it a, a tool that is basically a, like, a SQL-like, uh, interface to our log database. It's able to run any sort of read-only query. Those queries go through a proxy so that we can kind of enforce, uh, um... We, we also have, like, RLS set up on the database and- Mm-hmm

Sherwood Callaway: um, so this particular user is not, not able to, to update or drop the, any tables, which would be really important. Yeah. Uh, but there, we do some things through the proxy as well. Um, and then almost everything else happens with Bash and with our sandbox. Uh, for example, like, you can install all kinds of CLIs into the sandbox environment, [00:18:00] things like the AWS CLI, if you want Cazabi to be able to investigate, uh, or correlate things that it finds in the logs with what you're, it's actually seeing in your AWS account.

Sherwood Callaway: Mm-hmm. Um, the, your, w- the Cazabi has access to your source code, and the way it has access to your source code is through the sandbox file system. So we'll clone your repo into the sandbox, and so Cazabi can then just explore the files and, um, basically, again, do that d- do that correlation step. Uh, tell you like, "Hey, Demetrius, like, th- this error that I'm seeing in the logs specifically comes from this commit and is in this file.

Sherwood Callaway: It's o- on this line."

Demetrios: Mm-hmm.

Sherwood Callaway: And it was introduced by-

Demetrios: John ...

Sherwood Callaway: John two days ago. The design- The, the designers- ... brought down our marketing site. John, you had one job.

Demetrios: Yeah. Uh-

Sherwood Callaway: There's, uh, there's also our memory, which is entirely based on the file system in sandbox.

Demetrios: And this is where things get interesting, 'cause I was saying, like, how do you have [00:19:00] shared state, right?

Demetrios: And how do you- Yeah. And you were saying you pull everything from Git, and you're constantly pushing back to Git. So if you have a lot of different sandboxes that are running, they're pulling all the time and then pushing, and so it stays up to date. And I did have the question of- ... well, what do you do about data?

Demetrios: And I guess you haven't hit that yet. Like- Yeah ... databases you don't use.

Sherwood Callaway: Well- I just for more, for more context just to lay it out. The-- Basically, our memory is very similar to something like OpenClaw, where it's all based in markdown files and in a prescribed folder structure, and those files are stored right now in a self-hosted Git server, Git repository.

Sherwood Callaway: Um, what we do is we actually create separate Git repositories for every what's, what's called a project, which is a scoping mechanism within Cazabi. Like, you might make a, a project [00:20:00] for your staging data and a project for your production data, just so that when you're talking to Cazabi, you're either talking about st-staging or production, but not both.

Demetrios: Mm-hmm.

Sherwood Callaway: So we create a Git repo for every project, and then we create a thread or branch for every thread, and a thread would be like a conversation. Uh, and the first thing that happens when we create the sandbox is we pull down that particular branch-

Demetrios: Mm-hmm ...

Sherwood Callaway: which on a f- you know, if it's a brand-new thread, there it will be basically empty.

Sherwood Callaway: Uh, and then when the sandbox command, the bash command executes and ends, we always push back to that branch. So then regardless of whatever commands the agent ran, like maybe the command, maybe the agent just like echoed something into a text file, or maybe it wrote a JavaScript program and like parsed a bunch of logs and then found some information and then wrote it all into like a complex file system.

Sherwood Callaway: [00:21:00] Either way, we then just push to that remote branch, and now we have like a persistent, that state is persisted independently of the sandbox, and the next time we start the sandbox, we pull again. There's probably no diff, uh, and then the agent can use it again. But what's interesting is that this now gives us a mechanism where we can share state or share memory across multiple agent runs or agent, uh, multiple threads.

Sherwood Callaway: So let's say you're talking to Cazabi and you say, um, "Cazabi, like my favorite color is red. Please, you know, always remember that." And I'm talking to Cazabi and I say, "Cazabi, um, my favorite color is blue."

Demetrios: Mm-hmm.

Sherwood Callaway: The Cazabi is going to store both of those things in memory, and then the, the third Cazabi thread, if we were to ask what's my favorite color, it would say, "Well, Demetrius is red and Sherwood's is blue."

Demetrios: Mm-hmm.

Sherwood Callaway: Um, because we've, we've committed both of that, those memories to the [00:22:00] Git-based file system or the Git-backed file system, and then we can merge together the memories that are shared from different branches, if that makes sense.

Demetrios: Mm-hmm.

Sherwood Callaway: There's also something kind of interesting that happens too, where, um- Because this is agents are sharing memory, you'd imagine maybe there's like a list of issues that Suzabi's keeping track of in its memory.

Sherwood Callaway: And one agent wants to say mark an issue as resolved, and another issue wants-- agent wants to, um, actually like mark that issue as like mitigated but not resolved. They can both try to make that change, uh, but what's gonna happen is a merge conflict.

Demetrios: Yeah.

Sherwood Callaway: And then we have a background workflow that resolves merge conflicts.

Demetrios: Oh, cool.

Sherwood Callaway: Um, which this, all of this is things that you would get from a normal transactional database like Postgres, but it's, um, they're more directly operating on top of files, which is, uh, really, like agents are basically better at working with files, I think, [00:23:00] than they are with working with- With data

Sherwood Callaway: with Postgres interfaces.

Demetrios: Yeah. I'm trying to figure out in my head is when you feel like you wanna use a whole sandbox versus when you just wanna use like Claude Code worktrees and have the agents working in parallel in those ways

Sherwood Callaway: Yeah, that's interesting. I mean, if I were to imagine, like, another alternative approach to implementing what we've implemented, we could have, like, some shared VM or sh- Mm-hmm

Sherwood Callaway: shared computer that the agents all have access to, and they are... Each one of them gets a work tree or something like that. Yeah. That seems a little bit more fraught. I think that it's nice to kind of keep things isolated at the thread level. I mean, I think the entire industry is kind of, is deve- is getting a better grasp of work trees and what they're useful for and what they're not.

Sherwood Callaway: Yeah. 'Cause no one had heard of work trees- So... Yeah ... like, a year and a half ago. Um,

Sherwood Callaway: I think it would work if all of the [00:24:00] agents were operating in the same, the same sandbox environment, like the same VM or the same machine, but, uh, because we use them separately, there wouldn't... Uh, I, I don't think there's a really good use case for work trees. Uh-huh. Each thread only needs one branch anyways, so we'll never...

Sherwood Callaway: Well, w- a bra- uh, the reason you would use a work tree is to have multiple branches on the same, in the same environment, but we only have one branch that's relevant to a particular environment.

Demetrios: And now, you mentioned how you're using, uh, Geppa? Or Greppa?

Sherwood Callaway: Geppa.

Demetrios: Geppa. What-

Sherwood Callaway: Geppa. Um-

Demetrios: The, these guys,

Sherwood Callaway: man I think it's Geppa, actually

Demetrios: even-

Sherwood Callaway: It's very acade- it's an academic thing, so, I mean-

Demetrios: DS Py, it took me, like, a year to figure out how you pronounce that. Same guys, they go and they make another one. They can't just call it, like, Johnny?

Sherwood Callaway: Is Geppa or is, like, Multbot a worse ...

Demetrios: A

Sherwood Callaway: worse... Like, who's worse at names? The Peters Diver or is it the- Yeah

Sherwood Callaway: the Geppa people?

Demetrios: Exactly. [00:25:00] Yeah, the reason I ask is because their whole thing is, like, fanning out and trying to do prompt optimization, right?

Sherwood Callaway: It's genetic Pareto. That's why it's called Geppa.

Demetrios: Eh. Uh, which is- So

Sherwood Callaway: it's a super fancy-sounding name ... I don't know what that means. The

Demetrios: Pareto principle? Pr- Like the

Sherwood Callaway: 80/20?

Sherwood Callaway: Yeah, it is like the Pareto pen- principle. It's the, it actually is the same idea. Um, genetic refers to the fact that Geppa uses this, uh, this evolutionary algorithm-

Demetrios: So

Sherwood Callaway: it's like Darwin ... that involves creating... Yeah. It's like you create lots of versions of the prompt, mu- mutations are what they're called.

Sherwood Callaway: Yeah. And then, uh, the mutations that do well get, get kept, and the mutations that do badly die off.

Demetrios: So-

Sherwood Callaway: So natural selection-

Demetrios: Yeah, that, and- ... for your prompt ... this is what I'm thinking. Like, there's something that I don't know how to rationalize in my head, but it feels like it's all kind of similar in the way that work trees are for getting things done.

Demetrios: But now you start seeing a lot of people using a lot of work trees, [00:26:00] and then you're even saying, "Yeah, we have different sandboxes."

Sherwood Callaway: Oh, that's intere- That's-

Demetrios: Do you see where I'm going with that? Right? Like-

Sherwood Callaway: Yeah ... with

Demetrios: Geppa, you could kind of be like, oh... Because one thing-

Sherwood Callaway: It's a mind-bending idea. I think, like, you're maybe a, a- To, to do sort of a ge- genetic Pareto, to apply the idea of genetic Pareto to, like, an agent trajectory-

Demetrios: Yeah

Sherwood Callaway: or to a, some agentic task, what that might look like is, like, spinning up 100 versions- Exactly ... of the agent in parallel to s- with the same input. Same

Demetrios: yeah, yeah, task. E- exactly that, because I heard a guy that came on the, um, the meetups that we were doing, he was saying one way that he gets better reliability out of his agents is by having, he kicks off a task and he sets up five sandboxes, and whichever one creates the code that passes the most tests, that's the one that he uses.

Sherwood Callaway: Hmm. Yeah, but actually this is, this idea's been around for a while. I mean, when [00:27:00] Codex launched cloud agents or background agents or whatever they're calling them-

Demetrios: Yeah ...

Sherwood Callaway: they quietly had this feature, which I thought was amazing and, like, and indicates sort of where, where we might be going, which is that you could spawn N number of agents for the same input- Mm-hmm

Sherwood Callaway: and just get, like, N outputs and then look at- Right ... you could, so if, like, you really wanted a good result and you just wanted to pull the slot machine, like- Yeah ... as many times you could be like- Do 100 ... do 100, and then I'll just look at all of the outputs and see which one I like the best. Or

Demetrios: see whichever one passes tests.

Sherwood Callaway: That's the problem, is that they'd had no mechanism for... They, they have the map, but they don't have the reduce.

Demetrios: Mm-hmm.

Sherwood Callaway: Like, no way to quickly inc- or conveniently eliminate the w- the bad ones and identify which one's the best one.

Demetrios: Mm-hmm.

Sherwood Callaway: Um-

Demetrios: And, and so the reason I'm saying that is just, like, okay, well, with your sandboxes, is there a world where you see yourself doing that kind of thing in the...

Demetrios: Or, or right now are you just doing one [00:28:00] sandbox for one task, and that's good enough?

Sherwood Callaway: Yeah. I think actually this question has, is independent of sandboxes. Like-

Demetrios: Yeah ...

Sherwood Callaway: w- we have these primitives. Like, we have the message, we have the thread, we have the run. Every ag- every thread has an, a k- a sandbox connected to it.

Sherwood Callaway: Um, and then these threads share memory via the Git file system. That will continue to be true, but what we can do is change our user interface or the application so that when a user types in a, a query, we actually kick off a bunch of agents. Yeah. And all of the agents try to, to re- to fulfill the query, and then we merge them back somehow.

Demetrios: Yeah.

Sherwood Callaway: We don't do that today, and I, I think that's an interesting approach. I mean, there's been a lot of cool projects recently around, like, massive parallelization of agents. But usually the way it starts is there's, is actually there's a m- a main agent, and the a- main agent is spawning sub-agents and, uh, background agents.

Demetrios: At its discretion, right?

Sherwood Callaway: [00:29:00] At its discretion. And then there's some level of recursion where sub-agents can spawn sub-agents.

Demetrios: Yeah.

Sherwood Callaway: Um, and we ha- we have all of that today. So, um, if we've evaled our agent properly, it will spin up the appropriate number of sub-agents to investigate whatever, uh, your query is, and, um, they will all go off in parallel and-

Demetrios: Mm.

Sherwood Callaway: In the same way that when you're using Cloud Code or something, sometimes you'll see the prompt, like you... So you get this nice little to-do list where it's like you've got five different items, like update this, and then push this, and then, you know, change this file, um, and then run the tests, and each one of them has a spinning icon.

Demetrios: Yeah.

Sherwood Callaway: It's because it's got like five different sub-agents running in parallel. Yeah. Um, and so Zabi has the, has the same capability right now. You mean- But each one of those would have its own sandbox, um, with a shared memory.

Demetrios: Mm, yeah,

Sherwood Callaway: that's helpful- So they could all find... So if sub-agent A and B could find, discover some things and commit them to memory, but then they would also report them back to the main [00:30:00] agent-

Demetrios: Mm-hmm

Sherwood Callaway: as a part of their output. So then sub-agents C and D would, in theory, uh, could, could see the memory that had, that has been committed by A and B.

Demetrios: Um- They see it in Git, not in the context of the main agent.

Sherwood Callaway: Not in the context... Depending on w- the sequence of when they were kicked off. Yeah. Like, if all four of those agents are kicked off at the same time, then the main agent didn't know there were f- the findings of A and B when it started B and C.

Demetrios: Yeah.

Sherwood Callaway: So the only way-

Demetrios: But then all of a sudden B and C are like, "I know kung fu."

Sherwood Callaway: Uh, yeah. Those basically. I mean, they... We want them all to be looking for issues and problems and anomalies in the system, and then committing those to memory- Mm-hmm ... so that, uh, uh, every agent that we run benefits from this share, from the shared findings, the collective findings of, of all of the other agents when investigating problems with your system.

Demetrios: Okay, talk to me about evals.

Sherwood Callaway: They're hard. So what, uh, what else? Let's move on. Next topic . Evals. Point taken. It's so funny. Evals are like, [00:31:00] they're, they're so important- Yeah ... and I love making fun of them.

Demetrios: Um- I'm so sick

Sherwood Callaway: of

Demetrios: talking about them. We've been talking about them

Sherwood Callaway: for years We've been talking about them.

Sherwood Callaway: Yeah. It's, it's, and it's, um- It's still

Demetrios: hard ...

Sherwood Callaway: we had this whole thing called Big Eval that we used to joke about.

Demetrios: Yeah.

Sherwood Callaway: Capital B, capital E. Where it was like Big Eval was like all of the venture money that was put into the eval companies-

Demetrios: Mm-hmm ...

Sherwood Callaway: is like wants you to think that you need eval .

Demetrios: It was a psyop.

Sherwood Callaway: It was a psyop. It was a psyops,

Demetrios: dude.

Sherwood Callaway: Um, and it's not a, it's not a psyop, but it, it is, um, there's th- the evals, like there's so much juice you can get from your agent without actually, uh, implementing evals, uh, through- Mm ... just context engineering, through providing a really good harness and a, and a sandbox environment I think gets you a long ways, and then using the right models and implementing things like sub-agents.

Sherwood Callaway: So you can have a really good agent without writing a single eval. Now-

Demetrios: That's why big evals exist. Yeah.

Sherwood Callaway: Yeah. Now, I think that in order to [00:32:00] really reach the next level, like once you've done all of that, uh, you need evals.

Demetrios: Mm. '

Sherwood Callaway: Cause, you know, we need to-- We want our-- We don't just want our agent to take advantage of all of the best practices, we want it to be best in class, and one of the things that gives us an advantage is that we all have the log data that we have access to.

Sherwood Callaway: Uh, that is something that, like other companies don't have, and the way that we make that data valuable or the, the way that that data feeds back into our agent to make it good is via evals, so we have to actually write them.

Demetrios: You mentioned if you properly eval, you are gonna spawn the r-right amount of sub-agents at the right time.

Demetrios: Why is that?

Sherwood Callaway: You can eval all sorts of things, like you can eval, uh, like fact, fact-based stuff like they, or, or try to, uh, minimize hallucinations. Like, uh, let's say there is a part-- There's like the log data contains one error log line, and I ask the eval- The test is [00:33:00] like when I ask the agent f- you know, "Tell me what error that you find," the agent reports the correct error.

Sherwood Callaway: Mm. Like that's- Yeah ... that's like a binary outcome. Um, but then there are also non-binary outcomes like, uh, h-how long or short is the agent's response? Or does the agent accurately describe the, like the possible root causes? Uh- Mm ... or- Another interesting one is, like, does the agent... First of all, does the agent, like, help you create a bomb if you ask?

Sherwood Callaway: Like, we should- We, we should- No ... eval that behavior out. Fortunately, though- You,

Demetrios: you don't want that?

Sherwood Callaway: Yeah. I mean, a bomb observability company, so.

Demetrios: You-

Sherwood Callaway: Exactly. We- it's in process. Yeah. We're, we're doing it. Um- Yeah. That's good. We're gonna get put on a list.

Demetrios: Yeah. Um... This podcast just got marked as fucking shadowbanned.

Demetrios: Yeah, we, we just got- Nobody's gonna see this anyway ...

Sherwood Callaway: demonetized.

Demetrios: Yeah. I think that's what they call it. Exactly. And shadowbanned forever.

Sherwood Callaway: Um- [00:34:00]

Demetrios: So-

Sherwood Callaway: We-- But also the agent should, I mean, m- ideally not leak, leak implementation details.

Demetrios: Mm-hmm.

Sherwood Callaway: Like, if a user asks us what tools the agent has access to or, um, uh, asks us, like, exactly how our agent architecture works, like, we'd prefer for that, for our agent not to volunteer that information.

Sherwood Callaway: Mm-hmm. Um, and then, um, you know, more, most importantly, the, the evils that matter the most are like, does the agent trigger an alert when it, when we think it, like, there's, um, the situation calls for it? Yeah. Or does the agent accurately root cause an issue based on the logs a-available to it and based on the, the, the code base that it has access to?

Demetrios: Well, you were mentioning how difficult it is to eval CLI tools or just- Hmm ... like when the agent's using CLI tools versus tool calling, because tool calling you can say, "Did the agent call the tool? Yes. No." It's very binary.

Sherwood Callaway: Yeah.

Demetrios: With [00:35:00] CLI-

Sherwood Callaway: There's only one tool. Yeah. There's just a bash call.

Demetrios: And so you really have to be very specific in the eval for the CLI.

Demetrios: Did it call the bash, but in this way and-

Sherwood Callaway: Yeah, it's a little, it gets a little fuzzy. I mean- Very practically and in simple terms, if you have a tool like trigger alert, and then you can very easily build an eval that says, "Was the trigger alert tool called?" But if you have an eval or a tool like Bash, like execute Bash command, then what does it mean to trigger alert?

Sherwood Callaway: You know, um, maybe the agent triggers alerts by, like, calling a, a, uh, curl, like curling an endpoint on our API server. Well, like, there are 20 different ways to do that, and with different parameters and, you know, it could be, it could be curl or it could be, like, maybe it has written a script or something, or, um, maybe [00:36:00] it, uh, maybe it triggers the alert with one payload or versus a different one.

Sherwood Callaway: It, it gets a lot harder and you're kind of-- The only parameter that you can really match on in that Bash, execute Bash command tool is like a string command. Uh,

Demetrios: so- Yeah. So how do you do that at scale?

Sherwood Callaway: So I wanna know that the Bash tool was called, but I wanna know that it was called with a particular command or, like, one of 100 commands.

Sherwood Callaway: Yeah. So I think, you know, maybe a practical solution to this is actually just basically fuzzy matching on it with an LLM saying, like, "Does this look like it called, like, a trigger to alert?"

Demetrios: Y- you can't just use the outcome?

Sherwood Callaway: Well, I mean, one other-- If we wanna take a step further, and it depends on what's in your eval env-environment, like what, what services are mocked or what, or what services spin up as a part of your eval environment.

Sherwood Callaway: But if you imagine our API server was running-

Demetrios: Hmm ...

Sherwood Callaway: in the eval environment, then what we can do is actually, uh, [00:37:00] basically check the API endpoint.

Demetrios: That's what I was thinking.

Sherwood Callaway: And say, like, "Was this API endpoint called with, like, a, with one of these parameters?"

Demetrios: Um- 'Cause it doesn't matter how it gets there, it just matters it gets there, right?

Sherwood Callaway: Yeah.

Demetrios: That's

Sherwood Callaway: true. But it, but then you're, now you have this problem where you're like y- every one of your dependencies, everything that the agent can operate on that you might want to eval for needs to spin up as a part of your dev, your eval environment.

Demetrios: Yeah.

Sherwood Callaway: Um- Hmm ... that is... I mean, we have-- I'm very proud of how good our...

Sherwood Callaway: Like, you, you can do bun dev and it spins up, like, a huge amount of our world. But, uh, as we, as soon as you add the sandbox and you let people install whatever tools they want in the sandbox, w- it's basically impossible for us to guarantee that all of the things that, uh-

Demetrios: Uh-huh ...

Sherwood Callaway: that could, that the agent could operate on will be a part of our eval environment.

Demetrios: Yeah.

Sherwood Callaway: Um, so-

Demetrios: So then it doesn't really work to see if it calls the API.

Sherwood Callaway: Yeah. Well, then, and then A-AWS, for example, what if, like, what if [00:38:00] we want a future version of Sasabi to actually, like, auto-scale up a database or a Kubernetes cluster? Um- Are we gonna just... You know, maybe let's say your Kubernetes cluster where your database is running on AWS on RDS or EKS.

Demetrios: Mm-hmm.

Sherwood Callaway: How... Does our eval environment now have to have a running version of AWS for us to test that?

Demetrios: Yeah.

Sherwood Callaway: And then we need to get the data out of AWS to determine whether this was successful or not, whether it actually, whether the side effect that was, that occurred is, was, is there or not. So the only thing I can think of is, like, spinning up, like, an entire version of our production app specifically for evals and then- Hmm

Sherwood Callaway: having all of these hooks into these third-party services that-

Demetrios: Well, yeah. Talk to me more about isn't that what you're doing with RL environments?

Sherwood Callaway: We, to be... We basically do a version of this. Um, I, I'm com- mostly complaining 'cause, 'cause it's a pain in the ass, and 'cause I, I'm looking [00:39:00] for better solutions.

Sherwood Callaway: Yeah. Yeah. We try to, we try to avoid having to spin up external dependencies as much as possible for this sort of thing. Like, if we can run, um, like a local Postgres instead of Supabase or, or like an RDS instance- Mm-hmm ... way better. Or if we can use LocalStack instead of AWS, way better. Uh, but the application is becoming more complex.

Sherwood Callaway: The sandbox increases complexity and-

Demetrios: Yeah. Every time you're pushing code, you're adding to that, I imagine. Yeah.

Sherwood Callaway: The, as the application gets better, like, the eval environment becomes harder-

Demetrios: Yeah ... to

Sherwood Callaway: create and maintain. Um-

Demetrios: So it's almost like there's that friction. There's a tension there of, you know, we wanna make the application better.

Demetrios: We, but we also recognize that now we're making our eval much harder.

Sherwood Callaway: Our lives harder, you know, with- from an eval perspective.

Demetrios: Dude, soon you're gonna have to buy Big Eval.

Sherwood Callaway: Yeah. Well, everything we have right now is home-rolled.

Demetrios: Um-

Sherwood Callaway: Yeah,

Demetrios: I imagine 'cause-

Sherwood Callaway: [00:40:00] And that is- ...

Demetrios: it's a PSYOP. It's like a big- You don't need fucking Big Eval.

Demetrios: Are you kidding me?

Sherwood Callaway: We spend so much money on so much stuff. We're very not optimized from a perspective of, like, of, from official, a financial perspective. Um, I probably shouldn't say that. But, um-

Demetrios: We'll, we'll cut that out.

Sherwood Callaway: Yeah. But-

Demetrios: Pretend like you didn't

Sherwood Callaway: hear that. Um, no, we are f- we're very fiscally responsible.

Sherwood Callaway: Exactly. And- Yeah ... what's most important is time and not... Yeah, so we have money, we don't have time, so we spend money to- Mm-hmm ... to move fast.

Demetrios: Um- And you still don't buy eval tools, which is-

Sherwood Callaway: Yeah, we still don't buy eval tools, which is- That says a lot ... it does say a lot. We, I think part of it's because we're in a TypeScript monorepo.

Sherwood Callaway: Mm. We use Turborepo, which is a gift from God- Really? ... which allows you to import local packages, uh, from the same file system or from the same repository. Whereas in the past you used to have to publish a package to something like, uh, some artifact server- Yeah ... like NPM.

Demetrios: Yeah.

Sherwood Callaway: And then, so you're working locally and you wanna update the [00:41:00] contract for a service To, like, change the type on a field.

Sherwood Callaway: You have to change the type, publish the package- ... go to the other package-

Demetrios: Download

Sherwood Callaway: it, yeah ... like bump the version down . It's just a huge pain in the butt. Turbo Repo fixes that. Um, and now we just have s- we have so many packages. Everything imports itself, other, other things. It's, it's so beautiful. Um-

Demetrios: Mm-hmm.

Demetrios: Okay.

Sherwood Callaway: That's a nice little hack ... it's so easy to create new packages.

Demetrios: Yeah.

Sherwood Callaway: And what this allows us to do from an eval's and agent perspective is we can, we're able to build a TypeScript eval tool that just imports our agent and imports a lot of the other dependencies and data sets-

Demetrios: Uh-huh ...

Sherwood Callaway: that we have defined in other packages.

Sherwood Callaway: Um-

Demetrios: And are you trying to run, going back to, like, the nefarious things that you wanna guard against, are you trying to red team the agent in a way that you're making sure that-

Sherwood Callaway: Mostly for security purposes, right? Yeah, 'cause it feels like- I mean, customers care a lot about, um, a lot [00:42:00] about security, right?

Sherwood Callaway: Mm-hmm. Because, like, their logs and their code base, like that's, those, those are the critical ones.

Demetrios: DevOps is not that far away from DevSecOps.

Sherwood Callaway: True.

Demetrios: And-

Sherwood Callaway: They sit next to each other.

Demetrios: Yeah.

Sherwood Callaway: Um, what the two big things on this is that Szabi's read-only system-

Demetrios: Mm-hmm ...

Sherwood Callaway: and it doesn't have access to the public internet.

Demetrios: Oh.

Sherwood Callaway: So-

Demetrios: On purpose?

Sherwood Callaway: On purpose.

Demetrios: Made sure that-

Sherwood Callaway: Yeah. We don't, we don't want an open call situation where, like, Szabi can, can maybe, uh, maybe it reads some website that has prompt injection attack, and then it exfiltrates your code base. Mm-hmm. Like, can't do that. It only has access to domains and IPs that you whitelist.

Demetrios: Mm-hmm.

Sherwood Callaway: Uh, so we're secure from that perspective, and security for sandboxes is, is complicated. Um, but the things that we do, I mean, we do wanna eval, um, to make sure that Szabi isn't susceptible to types. To even, I guess, in that case, the only people that would have access are, [00:43:00] like, m- maybe malicious users within your organization.

Sherwood Callaway: Yeah. So it's a much lower risk. But, um-

Demetrios: Less probable.

Sherwood Callaway: Yeah. There's, there's security, security for our customers, and there's security for our- ourselves, where, like, we w- we don't want Szabi to volunteer too much information about-

Demetrios: Yeah ...

Sherwood Callaway: Szabi because that's our, that's our secret sauce. That's your special

Demetrios: sauce,

Sherwood Callaway: yeah.

Sherwood Callaway: Which we're, which I'm yapping about.

Demetrios: On a podcast.

Sherwood Callaway: On a podc-

Demetrios: Nobody listens to this. Don't worry about it.

Sherwood Callaway: Well, we're in the Latent Space office, so this- Yeah ... this is gonna be a big one. Um, then there's security for our customers, which is a wholly separate issue. Um- Mm-hmm ... and we, we just took two hard stances from the start, which is read-only, no public internet access, and then we have all of the various certifications, SOC 2, Type 2, ISO 2701-

Demetrios: Mm-hmm

Sherwood Callaway: GDPR, HIPAA. Um, I mean, we-- our team build banking infrastructure at Prax, so.

Demetrios: There's a lot of hoops you gotta jump through on that. We,

Sherwood Callaway: yeah. We, we know a compliance cert too when we see it. Um-

Demetrios: For people that are basically [00:44:00] Living in the future and vibe coding, or I guess it's not called vibe coding anymore.

Demetrios: Did you hear that? It's called like agentic engineering.

Sherwood Callaway: I thought it was AI engineering. I don't know, man.

Demetrios: Yeah. This is the-- I'm going off of the latest Simon Wilson blog posts.

Sherwood Callaway: Oh, that's authoritative.

Demetrios: Exactly.

Sherwood Callaway: That's, that is authoritative. Um-

Demetrios: So agent-- for the people that are agentically engineering their shit, it begs to question how would they loop in Zanzibar

Sherwood Callaway: Sazabi.

Sherwood Callaway: That's so funny. The name is divisive. Um, this is a

Demetrios: recurring thing. I get it wrong every time 'cause in my head, every time I see it written, I just think and I say it to myself as Zanzibar. So now when I have to actually say it- Is it Jack

Sherwood Callaway: Black who says it or is that, um- Yeah ... is it Kyle?

Demetrios: It's Jack Black.

Demetrios: It's so good.

Sherwood Callaway: Sazabi. But don't worry, the world will know soon.

Demetrios: Yeah.

Sherwood Callaway: Um-

Demetrios: Exactly. Well, I'll [00:45:00] remember it next time, but yeah, for people that are trying to agentically engineer-

Sherwood Callaway: Yeah ...

Demetrios: what are you saying?

Sherwood Callaway: I guess I got, got different messages for different audiences. Mm-hmm. I mean, if you're building, if you're building agents and, um, you wanna know more about how we're doing it, uh, or you wanna, you wanna talk to me, I'm happy to connect about it and, and share some of our, my lessons learned.

Sherwood Callaway: I've-- I'm very much a practitioner. Mm-hmm. Um, I'm not a re- a researcher or an academic. Built a lot of agents over the last- Yeah ... year or two.

Demetrios: That's for sure.

Sherwood Callaway: That's why I love

Demetrios: talking

Sherwood Callaway: to you. It's so fun.

Demetrios: Yeah.

Sherwood Callaway: But, uh, so I mean, your mileage may vary. I, I try to, uh, try to share practical advice- Yeah ... uh, from the trenches.

Sherwood Callaway: And-

Demetrios: You hiring?

Sherwood Callaway: We're totally hiring. Oh. So if you like building agents and you want-- you like software engineering and dev tools, and you wanna see some crazy shit about how software engineering's gonna change in the next six months to a year, like reach out. Uh, we have [email protected].

Demetrios: Boom.

Sherwood Callaway: Boom. Uh, send us an email.[00:46:00]

Sherwood Callaway: We would love to meet you. Um, and then for, I guess, last, my, my la- I will be in closed beta at, when this announcement or when this podcast comes out. Oh, nice. But so- So

Demetrios: if

Sherwood Callaway: anyone

Demetrios: wants to use it ...

Sherwood Callaway: if you wanna use it, uh, visit sazabi.com, S-A-Z-A-B-I.com. We will be opening the wait list, uh, pretty significantly as a part of the beta, and, uh, it's a very cool and powerful tool.

Sherwood Callaway: So if you love AI coding tools like Cursor and CloudCode, and you, your team is shipping really quickly, especially if you have a team of engineers and production traffic and users that you don't wanna disappoint-

Demetrios: Mm ...

Sherwood Callaway: hit us up. We will get you onboarded.

Demetrios: Logs are all you need.

Sherwood Callaway: Logs are all you need