AI agents are moving from pilots into production at companies, automating logins and routine tasks, but there's no centralized way to manage them. GitGuardian found hard-coded credentials exposed on GitHub, and agentic tools risk leaking tokens. Without proper governance and maintained playbooks, automation can drift or become orphaned.
Chapters
Welcome to What the Fudge for April 12th, 2026, the two-day-late edition. I'm late. Yes, that's right, I'm late. That's okay. This doesn't have to be something where it's immediately on time, right? That is a podcast and it's mine, so I can sort of make the rules. But just catching up on where we were, uh, the week, uh, in, uh, rearview mirror, I, I think this is maybe the week where all things AI agents and secrets really started to get into the zeitgeist. And so what we've, what we found is that security came up again, not, not the same one where it was everyone deciding that the security industry would be rendered moot by something related to AI. But I think it's more around people are just not being as careful as they should be. They're, they're doing a lot. And so in terms of the governance side of AI, you're looking at now kind of an AI sprawl. And that means that people are sharing secrets and not secrets meaning like corporate secrets, but sometimes it's usernames, passwords, identity, things that you really shouldn't be doing, just generally speaking, in sharing source code. But I think part of it is that you're going to see more enterprises out there that are going to be using some form of an AI agent. They're going from what would have been a pilot to more of an actual execution exercise. That also means that when you listen to the analyst community, it effectively means that if you look at the sprawl that has been enterprise applications, that will probably mean that some part of those will be very task-specific. And so imagine the thing you don't like logging into at your enterprise day job, if you have one, or are fortunate enough to be in an enterprise day job, you know, about, less than half of those will probably be ones you won't have to log into because perhaps something agentic will be doing that on your behalf. And you can perhaps stand back a little bit and work in your favorite harness and have a conversation with it as opposed to logging into it. Conversationally speaking, meaning you're typing something that an LLM is going to interpret as a command set or assuming what it thinks or intuits would be the next most logical thing that you might want to see back. Anywho, if you look at the leadership side of this, of course it means that you probably have some number of leaders that are going headlong because they are maybe C-suite and have outsized ambitions about what AI is going to do. And then you have the more, maybe more pragmatic kind of in the no, no, no, no category, if you're familiar with Simon Wardley's no, no, no, no, no, and then the oh fudge moment, that tipping point. So this is where there probably isn't a centralized platform in place to manage the growth of AI agents, let alone AI itself. And then if you look at, there was in the news a story around GitGuardian does a state of the secrets report and they found a bunch of hardcoded secrets being leaked, quote unquote, publicly in GitHub. That was from, say, you know, last year. Now it's kind of a question of if it's AI-related secrets, does that mean that someone is going to burn your tokens up using what you've already paid for because they're using your platform? So it'll be interesting to see what happens because more people are using so-called agentic coding tools. And so those coding tools, if they're not properly instructed, will do the absolute least to protect those said secrets. Maybe, maybe the agent doesn't know, as you say, it hasn't provided with the knowledge base to know that you might share it eventually on GitHub and in a public repository and so on and so on and so on. So again, sometimes computers will do exactly what you want them to do. And now we're dealing with computers that will do things, statistically speaking, maybe might be close to what you wanted to do, but they'll definitely do what you ask them to do if they don't have the information that's relevant, or as they say, the context. But, uh, switching, switching out from that, um, you know, there's, there's all the things that should be in your, you know, your CI/CD pipeline, and if you skip a step, don't be surprised if bad things happen. So just, uh, just, just put a, put a fork in that one. I think that topic's done for today. Then we go back into the world of, uh, the enterprise and the strategy stories that were coming out. So everything from Amazon defending what they're doing in their AI investments and what they tell their shareholders, or other people describing what they think is going to happen in terms of project viability, meaning going from the pilot phase to that production phase that we talked about earlier. And then the next element of this might be in terms of how we are looking at where data can maximize or predict its own value by figuring out whether or not the AI is relevant for the organization in terms of whether that would be a more profitable outcome or it draws cost out to increase, you know, bottom line savings. It'll be interesting to see how AI is predicted to help decide whether or not you should do AI. And then I think the next piece of that would be is the analyst community has started researching to understand what happens when an organization decides to cut staff, frankly, and embrace automation, but without necessarily the concept of a playbook. So what happens if the person that created the playbook doesn't understand the playbook because it drifts over time, or the person that created the playbook leaves, and then the person that inherits the playbook suddenly realizes that drift has occurred? And so how survivable is automation without an attention to the need for that lifecycle? So nothing new under the sun, just new sort of widgets being enabled. And then just rounding out, because I am of course interested in the space value chain, absolutely amazing. You know, all of the stories that came out of Artemis II, all the great memes, some of the amazing kind of sound bites we're getting from that. It just, it is just delightful, you know, to just be alive and on this little blue sphere you know, making its way through the universe. And, and then suddenly these folks, you know, tell stories that kind of put that into a greater context because they have literally been on the high ground. And it's just, it's, it's, it's really amazing. So enough about that. We're going a little bit long, but I just wanted to make sure I did check in, even if it is 48 to 72 hours late. And so thank you again for checking out this edition of What the Fudge for April 12th, 2026. Just a few days late. Thanks.
