Hi everyone,

Apologies for this being a later than normal post. There is one thing that I had been working on over the last month that I’ll talk about in a bit, but I also wanted to capture the events for this week as I traveled back to the states to attend Okta’s annual trade show, Oktane.

The first is a webinar I had alluded to last month looking at my approach to getting an initial handle on the tech stack from a systemic perspective. You can catch the recording here. Don’t worry, it is capped to about 30 minutes, and I will throw in a little extra something for those who watch it.

This ties in to the first full day at Okta’s trade-show: Oktane. They are one of the two major identity access management platforms and not one I have directly worked with in the past. In fact, this event was to see how much had changed from what I have heard about over the past few years.

While very high level, the entire focus is where you would expect: AI. Given Kong’s version of this conference is next week, expect more of the same.

Cutting through most of the marketing and AI fluff, the conference focused on a common theme for handling the agentic AIs floating through the ecosystems. Their first point has perhaps the most direct impact for this community: identification of the “shadow agents.” Yes with the scare quotes under the guise that they can be anywhere and everything. Almost like The Matrix.

Without defining what they mean by an agent, it could be as simple as someone using Claude Code or Cowork on the company laptop unmanaged and accessing a company’s specific bridge to their common service, or an unknown agent running through AWS Bedrock. Regardless, of what they are, the challenge becomes how to identify them so that you can place it under management, apply security policies, and then enforce the policies by taking it down when necessary.

And that is the magic: how do you identify the rogue “agent?” From the due diligence side to even post-acquisition, this is one of the hardest challenges facing the IT groups. Identifying everything that is running. Someone’s app that they bought using their personal card to make their life a lot easier when it comes to the day job, the script that another employee wrote to run the P&L statements, etc. Now to include some one-off or recurring tasks using Claude Cowork.

This is the long-winded background on how it all ties together: a common gateway that acts as an authorization enforcement point as well as an inspection point. You can’t monitor or enforce what you don’t know about. Likewise, you can’t migrate or assess post-acquisition if you don’t know about it.

Okta’s take (and suspect Kong as well) is to act as a gateway in front of your MCP instances, whether on-prem or cloud based. Any corporate app (or agent) will need to interface through that endpoint if they want to access company resources. Ergo, that is when it gets detected and management begins.

When you look at your other custom apps or scripts, what else accesses company resources, and what could you put in-front of it to see what else needs to come out into the light?

cab