Continuous Delivery Becomes AI’s Final Bottleneck
Continuous delivery is quickly becoming the pressure point in AI-driven software development. AI has moved through the industry like a wave. First it strained Git hosting, as agents pushed far more changes. Then CI ran short on minutes to test all that new code. Now CD is the final constraint between those changes and real business value.
Octopus Deploy VP of Open Source Dan Garfield examines that shift with Alan Shimel at KubeCon in Salt Lake City. Garfield co-founded Codefresh, maintains Argo and helped create the OpenGitOps principles.
Agents Need Guardrails Too
Garfield describes agents as people, only faster. People make mistakes, and so do agents. Excellent validation is what lets teams deploy more often without losing reliability. That is why policy engines and approvals matter more than ever.
Octopus Deploy ties together deployment environments, Argo CD instances and approvals from tools like ServiceNow and Jira. Its MCP server and configuration as code let agents spin up pipelines quickly. Permission guardrails keep agents from taking risky shortcuts.
OctoDoctor Catches Argo CD Anti-Patterns
Octopus Deploy also released two free open source projects this month. The first is OctoDoctor, a tool that scans Argo CD clusters and manifests for misconfigurations. Garfield explains that it packages years of hard-won lessons into deterministic rules. It is not AI-powered, but it can run as an agent skill or a CI step.
He offers one example of a dangerous combination in application sets. Paired the wrong way, two settings can delete applications after a simple network hiccup. Users quickly found settings they never knew were causing trouble.
Bringing Costs Closer to Developers
The second project is Cost Intelligence for Argo CD. It pulls OpenCost data into the Argo CD interface at the application level. Developers can see how a change affects the cost of their app. That feedback can also go to an agent for investigation and action.
Both projects live on GitHub, and Garfield invites pull requests and feedback. Better tooling around continuous delivery, he argues, helps teams stay productive and cost-conscious.
Transcript
Hey everyone, we're back here. We're live at KubeCon. I think for as long as I've been doing KubeCon, and even before that, I remember doing Jenkins Days and CloudBees Days.
Oh my gosh. This guy right here has been sitting next to me doing interviews. We were a little younger then, both of us.
You know, I figured out this is maybe my 20th KubeCon. Is it? Between Europe and Japan and all the North Americas.
I've been to every one except for the very first one in San Francisco. It was like 40 people. Yeah.
And ironically, it was like 20 miles from my house, so I could've been- There you go. You should've. Yeah.
But as I said, I think I met you before KubeCons, in CloudBees Days. Yeah, that's right. Yeah.
Well, we did DockerCon stuff. Yeah, I saw you at DockerCon and all that. Yeah.
Absolutely. If you don't know, this is Dan Garfield. Dan, I don't even know what your title is these days.
It's lost on me, too. I run open source for Octopus Deploy. Is it still Chief Open Source Officer or something like that?
Same kind of role. Okay. Except, co-founder of Codefresh.
Right. Sold Codefresh to Octopus Deploy about three years ago. Yeah.
So I've been kind of expanding in that capacity, focused on open source. I'm an Argo senior maintainer. Yes.
Helped create Open GitOps. And you and I partnered up to do the very first ArgoCon a couple years ago. 6,000 participants.
I think it was more than-- I don't want to bum you out, but it was more than a couple years ago. It was more than a couple years ago. Yeah.
Yeah. It was probably, what, about seven years ago? Maybe seven.
I don't know. Who knows? Seven or eight.
Yeah, something like that. Yeah. Time flies when you're having fun.
It was... No, it was 2020. So we're on 2027, my friend.
Yeah, 2020, seven years ago. Okay. All right.
Yeah. We got a problem. We're right there.
Okay. Anyway. All right.
You're trying to tell me I'm old. I'm old, and you were there with everything, too. But anyway, as Dan said, he's had his hand in so many different things.
The Argo, which is still the third largest- As CNCF projects goes- CNCF, yeah ... third largest. It's the number one way people deployed Argo to Kubernetes.
Yes. Probably most of the people that are walking around the floor today learned Kubernetes because they were staring at an Argo dashboard to do it. Absolutely.
And you were an evangelist evangelizing in the wilderness back then, right? Yeah. I remember those days.
Yeah, we built the GitOps principles, partnered up with AWS, Microsoft, GitHub, Weaveworks back in the day- Yep ... to come up with those. Worked with over 100 different companies, roughly, to come up with those principles.
And they have become the dominant way that people think about software delivery today, and I'm so proud to have been the scribe to write those down when working with all those people and helped juggle that all together. You were more than the scribe. I think you're being modest.
You were really a force there, Dan, and good on you. Listen, but you're with Octopus Deploy now. Yep.
com still, and cloud native now. They're familiar with Octopus. But for those who aren't, and there might be some folks here who aren't, give them the Octopus Deploy story.
So Octopus Deploy has made this big transformation into Kubernetes and cloud native, coming on with bringing in all of the Argo maintainers from Codefresh back in the day. And so we are a continuous delivery platform focused on how to get changes from the developer into production, validated along the way, handle approvals, meet policy. And the tweak that's happened over the last few years is that all of those policy engines have suddenly became way more important with agentic agents.
Because agents are like people, but faster. People make mistakes, agents make mistakes. What you really need to make it work is excellent validation, and so that's how you get that resiliency and reliability that allows you to deploy more frequently.
So we've seen AI is moving like a wave through the software industry. First Git got hit, and GitHub's going down every 15 minutes. They're getting demolished.
Then CI was getting hit. Oh my gosh, we don't have enough CI minutes to run all the tests we need to do to run all these changes, do the iteration. Now CD is the final bottleneck here, where if I actually want to get value from these changes, I got to have a CD that's resilient to get me there.
That's Octopus Deploy. Some people would call that the theory of constraints. Yeah.
That's right. The theory of constraints is exactly right. There's a book about that.
There's a book. A couple of books. A couple of books on that, yeah.
So how exactly is Octopus Deploy helping us with the CD bottleneck? So there's a couple of things. Ephemeral environments, feature flagging, being able to tie together all of your deployment environments, your Argo CD instances, make your changes through Git.
But then layer in, hey, I need an approval from ServiceNow. I need an approval from Jira or whatever your ticketing system is. We have a built-in approval engine, too.
So tying into that, removing those bottlenecks so that you get better validation, better pullback. And the thing that we've really had move the needle on a lot of adoption here is we've got, of course, an MCP server for Octopus Deploy. So, hey, I want to spin up additional deployment pipelines, validation.
That works directly with your agent. You can do it very quickly. It's all as code, config as code, so it works really well in the agent interface.
We've got a recovery agent to automatically help you diagnose what went wrong with the deployment, and a custom agent step that you can use if you want to trigger like, "Hey, go do a remediation. Write a pull request. " So all of those things.
But then to have the permissions guardrails around, that ends up being a key structure here. Agents love to take shortcuts when they can, just like people do, but they do it faster than people. So having those guardrails in place to say, hey, these are the policies that are applied to different environments, that really opens up the sandbox to be more productive.
But we're doing a lot of work on the open source side, too. Always. So this week, we just announced two new open source projects.
The first is Octo Doctor. Okay. And Octo Doctor is free.
It's open source. It's designed to help you with your Argo CD instance. We've been helping people with Argo CD since the beginning.
You were there. You saw it. Yep.
And we found over time, we've been writing internally all of these anti-patterns that we see that lead to problems. A good example is if you're using an application set, which is how you generate applications programmatically, and you have two settings, Allow Empty true and Prune true. There can be a condition where maybe you generate, there's a network hiccup, suddenly it deletes all the applications.
They should never be used together. It's a very dangerous setting to use. So we've taken all these rules that we've learned over the years.
We've built them into a CLI tool. You can use it as an agent skill as well. It's deterministic, so it's not an AI powered tool, but it powers an AI interaction.
So you can use it as a cloud skill. You can use it as a CI step. And it scans your Argo CD manifest and looks for, hey, are you following the best practices?
What are the things we should flag? Here's how to fix them. And since we launched that on Monday, we've had hundreds of people run these reports on their Argo CD instances and tell us, "I didn't realize how many bad settings I had.
" Go figure. Suddenly they can figure it. So that's really cool.
And we've had a huge response to that. And then we also announced Argo CD Cost Intelligence, or Cost Intelligence for Argo CD. Before we jump into that, let me tie the bow on this first one.
Please. It's available on GitHub? Yep.
It's available from the Octopus Deploy site? Yep. Probably not from the Argo site?
Or from the Argo site, too? Not from the Argo site. That's CNCF.
Yeah, we have kind of a shared CNCF group. There's a Chinese wall there, right? Yeah.
I don't know. I used to call that. A little bit, yeah.
So yeah, you can find it- Not a good thing to use this week. Go ahead. I'll firewall it though.
Yeah. com/octopusdeploy/opensource, you get links to all these projects and a summary to all of them. Great.
com as well. Okay. Now we can go to the second.
Okay, so Cost Intelligence. This is another thing that we're seeing from Argo users, where if they're using a tool like OpenCost, that gives them cost data for nodes or for an entire cluster, but it doesn't get more granular than that. " is so foreign because the whole point of Kubernetes is I don't have to think about nodes.
I think about pods or my application. So what we've done is we've integrated Cost Intelligence into Argo CD. So you install this as a plug-in in Argo CD, and it will give you cost data at the application level, and you can assign it to cost centers.
You can look at cost at the namespace level. So suddenly it's exposed to the developers who normally are so... I know the cloud providers love it to have us disconnected from our costs because we spend more.
But to have like, "Hey, I just made a change," Cost Intelligence for Argo CD can tell you the change that you just made makes your app 20% more expensive. You're going to spend an extra $800 this month on this app because of this change. I don't know if that's a good idea or not, but you may want to double check.
Right. Hey, is this the way that we want to run this? It's computationally expensive or whatever.
So putting that power in your hands. So this means that's AI powered? It's not AI powered.
No. No, it uses OpenCost data to get everything. But like these other tools, it's AI empowering, where you can have it give you back this feedback about, hey, this change, and I can pass that to my agent and have it do an investigation and take action.
So it's one of those things where we do really want to enable that agentic development experience, and we think that these are the kinds of tools that help people do that, especially with Argo CD, be cost conscious and be more productive ultimately. I love it. That's the idea.
And it's available same place as we just went over. Yep, same place we just mentioned. Free, open source.
There's no premium version for those things. It's just, hey, it's there. And you know what?
I'd love to have pull requests are open. It's not a CNCF project right now. Maybe in the future.
But right now, it is an open source project, and we'd love to have your pull requests and your issues and feedback. As you always do. Look, this is a man who knows how to build open source on open source communities and projects.
I got to embarrass you a little bit. I don't know if it's going to show up. These are 4K cameras, but I don't know if we transmit in 4K.
But show folks your little octopus. I got a little octopus sport coat going on here, yeah. Yeah, I don't know if that pops up on stream, but it's a pretty cool one.
Dan, hey man, it's always a pleasure seeing you. Thanks a lot. I appreciate you.
I appreciate you, all you do. Thank you for all you do. Dan Garfield, Octopus Deploy, Argo, so many other things, here on Techstrong TV.
We're going to take a break. I think Mike Broussard's going to run the anchor lap here for Techstrong TV. So stay tuned for that coming at you in just a minute or two.
But for now, thanks. This is Techstrong TV.