PagerDuty’s Leap into Generative AI and Operational Automation with Damon Edwards | AWS re:Invent 2023

This is Textron tv. Hey, everyone. We're back here on day three of our EWS Reinvent video streaming from, uh, our suite here at the Win.

I'm really joined. I'm happy to be joined by a, a good friend of ours. com, you've, you've heard of Damon Edwards, uh, him and our good friend John Willis.

I've known, well, they did that first DevOps days in the us Yeah, yeah. Silicon Valley, uh, com or Cam Cams. Cams, cams, yes.

To going back, going back to our DevOps world Yeah. Cams. And, uh, just heavily involved in, you know, it's really the idea of how do you get organizations moving quicker, moving the right direction, moving quicker, execution, it's operations, execution.

Well, but what I always love from you, Damon, is the two outta three. I want all three. It's not only moving quicker, right?

So we're moving faster, but with higher quality, better security. Yeah. I mean, it was having your, it was the ultimate having your cake and eating too.

It was Breaking, breaking the iron triangle or the whatever. That was the choice. Exactly.

Yes. Exactly. Yeah.

Um, so, but of course, Damon also was the founder of Run or Co-founder of rundeck. Yeah. Which was acquired by PagerDuty, what, about three years?

Three years ago ago now. Exactly. Yeah.

And, um, and since then Damon's been with PagerDuty. Yeah. And I don't even know what your title of PagerDuty is, man.

Yeah. Oh, uh, senior director product. Uh, basically I do product portfolio strategy.

So we have a lot of pieces. A lot of parts. Right.

And a lot of people don't know we have these parts, right? It's, uh, you know, think we're this, this on-call tool. You send an alert in and it notify somebody that we have all these AI ops capabilities, all these automation capabilities.

A lot of 'em are built on the Rundeck acquisition. And then we've just acquired Jelly, which is like, sure. Probably the best, you know, uh, really incident management going to, you know, the, the learning from incidents Mm-Hmm.

Side of things. And, you know, kind of world class team there. We acquired another company called Catalytic.

It's more human in the middle kind of workflow. So we've either built or, um, uh, acquired all of these parts. And, uh, so one of my kind of areas to oversee on the product side is looking, how do we stitch all that together to better, to make it more obvious to our customers and to the world.

Like, you know, what does handling unplanned work incidents being the most common thing in 20 23, 24, what's the look like? What's the world class way to do it? And it's not just, you know, letting somebody know in the middle of the night there's something to be done.

It's how do you deflect as much as possible through AI and automation? How do you drive not only the notification, but the decision making process and the resolution process? And then how do you keep all that synced up with your, you know, systems of record, right?

Your, uh, mm-Hmm. Your, uh, your service nows and, you know, JIRA, ITSM and those, and those things. Um, so, you know, we're the kind of like hub for this ecosystem of operational tools from the monitoring the automation to the ITSM.

Um, so yeah. So that, that's, that's part of it. And the other big, big part of that is reason a tremendous amount of, you know, or emphasis on ai.

Right. And, um, I think You're the only ones. Yeah.

So if you want to called language models and, uh, no. Yeah. So it's, uh, it's pretty, um, it's pretty cool.

And I think, you know, the way we see it is, you know, the LLM gives you this, or just general AI in general. It's LMS today, probably something tomorrow. This linguistic capability, right.

Okay. That it allows you to have fuzziness on the front end, which is, you know, human talk, right. Input, Input, Input.

But also, you know, the, the, these tools and infrastructure gives off language as well in terms of telemetry and alerts and, you know, things like that. So, and that's all often fuzzy. You need to have semantic understanding of that to know what to go and do, right?

Mm-Hmm. And then what, what to go and do is all the automation underneath it, right? So we're really, you know, leaning in on AI to not only kind of democratize access to all that information and, and automation, but also kind of really help people help drive through that, right?

Mm-Hmm. To, to to, um, not only help deflect things, but also to bring the, the, uh, the costs of your operations down dramatically. So, You know, I think one of the things that I hear very often when people talk about generative ai Yeah.

Is that the, the output of generative AI is in plain English, if you will. Mm-Hmm. Plain easy to understand language.

Yeah. 'cause we've always had the ability in it to input data and then output data and, and, you know, most recently, and let's say the last five, 10 years and output that data and, and make it actionable. Yeah.

Whether we're talking about an API kind of thing or, or whatever. Mm-Hmm. Make it actionable.

You called it fuzzy. I call it easy to understand language. I, I think that's really one of the big it, it, it Is.

And Advances here. The problem is, in the past though, you had to use like it's, you know, programming languages, right? Mm-Hmm.

Even if you call YAML a programming language, right? That's very specific, very finicky, very formatted tightly. He said that I didn't Into into a API that must get it a certain way.

Uhhuh, you, you have to know the magical incantations. Yeah. No, But it was very, and so it was very Structured.

And so human language or natural language, we'll call it Right. Better is, is is a fuzzy thing. Right?

And so if, if I just say to you, Hey, that webpage is slow, or you say, Hey, someone else says, Hey, the, the response time is taking too long, right? Like classic kinda machine learning. It's really hard to kind of pull those two things and say we're talking about the same thing.

It's like the website's slow. Yeah. Right?

So there's a fuzziness there. There is. And, and so that's kind of what I mean, that that unlocks a lot.

And in the past you could only go after problems 'cause it took that Programming. So you could Define, well, you define well, and you could spend enough time and money to, to lick that problem. So it means you really have to only pick certain things.

There's a lot of stuff you just didn't do. Yeah. Right.

And because that fuzziness at the, at the top end, at the, the interface layer and also the responses back from all these tools, that's kind of fuzzy as well as well too, right? So being able to apply, um, you know, uh, generative AI to that linguistic capability, right? To those different, to those different layers, really unlocks a lot of things that would've been too hard to do.

Too painful, too brittle, right? Yeah. Um, and so that's really kinda the excitement if you don't have to look from an inside out perspective, the excitement for us, we can do so many things that before we just wouldn't get to because, you know, it just was just too much.

Right? And now it's like we're found these ways to unlock that, but it's all based on the data and automation that we have access to already have. Already Have.

And I think that's the, what we're seeing is that the, the linguistic, the language model part of it is all gonna trend to parody very quickly. Mm-Hmm. Um, and you know, so we use the best of breed across that, but the hard part is getting the right knowledge and the right, um, action capabilities together.

Married, married, and expose that to, to that, you know, to that linguistic capability. Um, and you know, today we're doing it for, we call it like an iron man suit, right? So, you know, when you're going into an incident, you know, it can pull this and pull that and present it to you and, and hey, call your attention to things that are, that are abnormal, um, understands where to get the sources of data, presents you with options of what you, what you can, automation you can run to do about or help you Mm-Hmm.

Or what's what you've done recently, right. Or who's the right person to talk to. So it's really like the Iron Man suit.

But we're eventually you wanna get to is, you know, where you're basically building these personas that help you in terms of like incident commanders and you know, a deep analyst in these different areas like responders, right. Um, or you know, like a, a post-mortem investigator, right. Or a, you know, uh, after actions, the words always change for what you're supposed to call the, the after.

I understand the learning from the incident later. And so eventually you get to where, you know, it's, as the agent side of this technology develops, matures, it matures, then I can see, you know, where that's where the really the focus is gonna be. 'cause one is saving percentage of time, the other is like exponential increase in right.

Capacity To, to work. And, and that I think is the, the push and pull of this, right? Yeah.

Or the yin and the yang. It's, I want to 10 x things, I want to go faster, right? To that classic DevOps, I want to 10 x this stuff, but at the same time, I wanna save money.

I want to be more efficient. And, and at some level they do meet, right? But, but they're two separate things.

Warren did to quickly mention, you guys made some announcements I think today around, Yeah, I mean, I think they're soft announcements, but I mean, it's no secret. We're keeping, we're keeping in, which is, um, you know, we've had these different generative AI capabilities that are already in the product today. They're already helping people out.

And it's funny 'cause it's things that if you live in operations, you're like, oh, that's cool. If you're outside of operations, you're like, what's that? But things like, uh, status updates.

So in the middle of an incident, you got this team, they're thinking really hard, and then a little buzzer goes off and goes, it's been 20 minutes. We gotta give an update to our internal stakeholders. And then you're like, okay, well let's clear our heads.

What can we write that the boss's boss is gonna understand what we're doing? So it's like the ability to auto generate that, or auto generate the first draft of postmortem or morph on our runbook automation side, generate automation jobs. Mm-Hmm.

So if you say, Hey, I wanna create this, this, uh, health check that's gonna know how to check x, y, Z technology. Gimme the top 10, you know, things you'd wanna check, you know, someone's gotta go and figure that out and put it together. Us you type it in, it builds the scripts, it builds the rundeck job, you know, or the, excuse me, the runbook automation job for, uh, for you.

So we've got these different pieces, right? And I think, you know, what we've done now is we're pulling it together in what we call PagerDuty co-pilot, and it's an operational co-pilot, right? So it's an idea of that is our, the intelligence behind the scenes.

And you see it some ways in, uh, features in the product, like behind a button. And then what we've been, um, kind of showing off, you know, it's very, that that's in the product now. And then the preview we're showing now is the ability to actually chat with it, right?

So through teams Slack, that kind of stuff. Excuse me, to be able to, you know, uh, ask it to help you with things or that p type things. So that To me is, so that's the beauty of the chat bot, right?

Is you could put that interface into a Slack or a teams or whatever you're Yeah. Not picking any side steps. Sure.

Um, it it, it's a, you don't think, but it also, you know what, from a knowing PagerDuty as I know it, it's, it's almost organic for you guys to be able to have that now. Yeah. Yeah.

And you know, it, it's interesting 'cause what PagerDuty's always been this unique position, which is like, it's about the human to tool interaction. Mm-Hmm. Like the human automation, so to speak.

Getting people to do things know things is as important as the machine automation, like actually going out and affecting change in the world. And when you can bring those two things together And use like a chatbot interface for it, that's pretty damn cool. Yeah.

I think it's really cool. And, and you know what, what effectively happens is, you know, you have the call it like the, the natural language understanding part of it, and then you sort of have this like semantic router, right? Which understands what you're TA asking and then says, gee, I think that is related to these things.

And you know, if you say like, what's wrong with that service? Right? Well, someone's gotta know that.

Like maybe I'm asking you for, uh, you know, to run some automated diagnostics. Maybe I'm asking you just to show me the monitoring events that came in. Maybe I'm asking you a little further out here, and meaning maybe I'm asking you to show me what recent changes have happened.

So being able to understand that, but then really to, to deliver those tasks then. So that first part, the natural language chat is like, that's, that's a solved problem now, right? Mm-Hmm.

I mean, it's, it's gonna get better, but it does that, the, the semantic routing part of it is, you know, I think very interesting kind of emerging part of this, but then the real thing is all of the automation and the data that you can bring to it. So it's like the hardest part is not the emerging cool part. The hardest part is the good old fashioned.

How do you get the automation and the data to that semantic router, so the natural language, you know, can, can access it. And, um, yeah. So that system that we're building, um, you know, with all those skills is, uh, what we're gonna be calling, you know, PagerDuty copilot.

And that gives us the, and it's, you know, it, it's an operational copilot now, the name, you know, TBT might might change, but that's the idea is you might have a think of a code co-pilot, which is, you know, for generating code for you. Um, this is, you know, an operations co-pilot, which is riding shotgun with you in your, uh, operational work. So the lawyer in me though says we've got Microsoft Co-pilot.

Yeah. With all due respect, I think you're the fourth company we've had here over the last three years. Yeah.

Over the last three days. Yeah. Using the word co-pilot is co-pilot.

Such a generic word that we, we think it could be co-opted by everyone. I, I think it is. I I think, I think it's actually more of an industry architectural term than it is a, uh, um, uh, Remarkable name.

I don't know. That's, that's way outta my depth. But we have, we have people that do the Well won't work for a public company, your people, right?

There's A public company like that. Let Them deal with it. I Don't have to worry about those things, but I, I know It's not co-pilot, it'll be something else, but it's similar.

Well, I think one of the things we didn't want to do is over anthrop, anthropomorph anthropomorphize it. Like, or they make it like a, like, Don't even, you know what I'm talking about. Right?

I know what you mean, but I'm not, I don't, we don't Wanna call it like, you know, the shimmy, like shows open, you know, Tells you, but that is gonna happen. Believe me. You know, we have barred, someone's gonna come up with Shimmy or Damon, or that's, we'll name our ai, someone else, another company we interviewed here had to name Davis.

Yeah. And I think, so what I look at it, and this is just our kind of working theory right now, which is our thesis, because all this stuff is just brand new, is that I feel like that's the company's responsibility, that's the customer's. I dunno if it's responsibility or their joy to bring personas to their agents, right?

Yeah. So, so it's like, I don't wanna like force you to use, you know, shimmy or you know, Clippy, right? Yeah.

Like, I wanna be like, I want your people internally to be like, what is the, what is this thing we're building and these personas? And we, we saw this even with Rundeck, like people would personify it because it's a self-service operations thing. So people would be like, they call it their, like they'd give an acronym, like it's the lead, it's the whatever.

They have these names. And uh, one guy, as they called it, the virtual Rafael and Rob, that's what they little, they have acronym for it because they, because they were, 'cause it was a stuff you had to go bug Rafael and Rob for now they're gonna give you a button to go do it. So, but That's a human, that's a human factor statement.

It's, but I think let the end user decide what's the culture of their organization and what do they want to do. I think for us to put that on them is a little bit like, it's limiting, right? It's it like, we Don't call me late for dinner.

Yes, We're gonna build this, we're gonna build this in intelligence service that's gonna help your people. And down the road when it turns into agents that are like per personas in your organization, well then you figure out what you want. It's like, you know, it's like naming your servers, right?

What do you, what do you used to be? I used To do, used to be Star Wars, then it was like Tolkien and things, and then it got so big. Now it's just, you know, numbers.

But, but remember that. But you know, let, I do remember that. Put my star, my, my son machines.

That names Yeah. Great names. C3 P is broken, you know?

Yeah, Exactly. No, we have Star Wars names, but I digress. Yes, dam, we're about outta time, man.

I want to thank you for popping up and saying hello. It's always awesome us a PagerDuty update. Yeah.

com if you want to know more. Definitely. Um, and that, you know, let me just say nothing we said here was, uh, uh, embargoed or anything.

You could get all this information, I believe on the PagerDuty website today. Yeah, Yeah. It's all emerging.

Uh, and check it out. Um, hey man, I hope to see you soon. Talk more.

I know you're doing a ton of stuff on operationalizing AI as well. I know that's a passion. Yeah.

Yeah. I, We can talk to you about that soon. Yeah, that's, that would be great to make, Talk about that.

Actually heads up, we're gonna be, uh, releasing some video around that, that Damon was part of and hopefully in the next couple weeks. So stay tuned on that and text on. But for now, we're gonna wrap up this, uh, segment here from the win at, uh, where are we again?

AWS Reinvent.