Tech Talks Daily
00:16:05 · Valerie Henderson
[00:16:05] S1: I think what we're seeing is a lot of our customers are already doing read only workloads. And I think that is an area that there is a little bit more comfort in how agents are operating there. I think outside of that, in terms of, like, true autonomous operation, so you can think about self healing infrastructure. That's only happening at the most advanced customers. And, frankly, within Caylent, our own managed service practice leverages that, but it requires very strong expertise. And to your point, security, governance, and the right platform that is essential. I also feel like there is a just a trade off of autonomy versus risk, and I think that that should always be reevaluated. So thinking about the accuracy, the reliability, and the capability to improve, like, can agents test things without side effects? And I feel like that removes a huge amount of the inherent risk. And the other piece is agents can iterate and do nothing useful and run up a bill, and that's super wasteful. But if you have the right workflows, the right guardrails, the right traceability, the right limits, which that's something that our Amazon Bedrock AgentCore does in a really purpose built way, you can be so successful here. And, again, that is the stuff that is the hardest. It is not sexy. It is not clashy. It is really breaking down these workflows and the access, the platforms, and how you can make sure that you're doing this responsibly.
00:17:42 · Neil C. Hughes
[00:17:42] S0: Again, for people listening here, many companies will be moving from AI that recommends an action to agents that will go and take one. What changes are you seeing when software is given authority inside a live business process? What kind of things are you seeing? Any surprises, delight? What kind of stories are you hearing?
00:18:00 · Valerie Henderson
[00:18:00] S1: Yeah. I feel like I am truly inspired by our customers because, like I said, this is not easy work, but I've been super impressed. We have a customer, Smarsh, that basically rebuilt their entire software delivery process with Agentic SDLC on Amazon Bedrock and Claude Code. And millions of dollars of engineering capacity were freed up, and they were able to reinvest that back into the business to continue to innovate. And to me, that's really amazing being able to break things down and then build it back up. And they were really clear on what they were trying to achieve. And at the end of the day, with that return, they said, okay. We have an opportunity to jump ahead of the competition to continue to push ourselves to be better. And at Caylent, we're disciples of only the paranoid survive, and getting to work with customers that are thinking about things from that perspective too is amazing. And so I would say I go more into the delight rather than kind of the fear, uncertainty, and doubt side of where we are right now. I think in any given time with a shift like this, you have dystopian and utopian thoughts. Yeah. I believe things are often somewhere in the middle. My mom used to say things are never as good as you think. They're never as bad as you think either. And I think right now that that's something that you have to really ground yourself in. And the cool thing is we're getting to work with customers that are seeing ROI and especially in that software development space. And we... You can see that in a lot of the data. You have a ton of economists doing work on where this is actually showing up. And right now, we work with a ton of technology customers, and that's a massive opportunity. And how you make that put it back into your business to go even further or faster.
00:19:49 · Neil C. Hughes
[00:19:49] S0: And, also, when I was reading up on you guys, I I think I read that there's something was a big obsession with model accuracy over the last couple of years, but I think I was reading that you were saying that model accuracy is no longer enough to judge whether an agent is actually ready to go into production. Can you just expand on that for for me for anyone that's not seen that?
00:20:06 · Valerie Henderson
[00:20:06] S1: Well, I think model accuracy is just one small part of it. I think you have to think about the... There's so many different models, agents, and there has to be, in my opinion, a sense of experimentation, but you also want the right expertise, which goes back to why companies like Caylent and other organizations are so important right now. I feel like what we do has never been more critical than it is today. And number one, I think that there is a sense of, of course, excitement about that. But like I said, a deep level of responsibility is needed right now. And you have to think about things, like I said, in terms of the security, the guardrails, the spend, the access. And access, by the way, is one of the most challenging things. And you think about how simple it should be, but every customer that has worked with different platforms for single sign in and then... We all know that that is not easy. And if it's not easy for humans, think about what that looks like for agents. So just thinking about those building blocks is really critical. So regardless of how good the agent is, if you are not, again, breaking these things down and being very smart about what and where and when and how, it doesn't matter.
00:21:24 · Neil C. Hughes
[00:21:24] S0: So Yeah. Completely agree. And I think we will have to get you back on early next year to talk about identity access management. It's a whole podcast episode on its own. And I was saying a moment ago, 2026 has been the year of Agentic. I don't know if you can share what you've been up to this year, but most importantly, any teasers you can leave us with on what you're looking towards, what you're working towards next year? Anything you can share there.