Agentic AI Is An Organizational Virus

I’ll admit up front — this comparison isn’t fully baked. But I think it’s worth sharing anyway, half-formed, because the shape of it keeps bothering me, in a useful way.

Here’s my thought: Agentic AI is moving through organizations right now the way a virus moves through a body. It doesn’t ask permission, and it doesn’t care about your org chart. It travels through whatever channels are open. Someone in finance connects an agent to pull numbers and draft a summary, shares their output with their team, and suddenly four more agents pop up as others try to automate. Maybe there was an official pilot of Microsoft Copilot somewhere in the business, but most people are just using Copilot Chat at this point…as well as their personal installations of Claude, ChatGPT, etc. Then someone in marketing wires an agent or two into the CRM. None of it shows up on IT’s radar, because there’s no fever yet, no visible symptom, just quiet replication in places nobody’s looking.

Does that sound like your organization?

The more I think about it, we’ve been through this viral issue before. Fifteen years ago, it was Yammer. Someone signed up with a work email, invited a few people, and within a quarter, half the building was on it while IT had no idea it existed. It spread through the exact same paths of least resistance, invisible until it wasn’t, until eventually it wasn’t a rogue tool anymore; it was load-bearing. Microsoft eventually bought it, less as a planned expansion of the collaboration roadmap and more as a way to stop competing with something they couldn’t dislodge.

Agentic AI is doing the same thing, except faster and with sharper teeth. Microsoft is adding some observability capability, but it lacks the centralized control that organizations are quickly realizing that they need. Yammer, at worst, meant duplicate conversations and shadow IT annoyance. But an ungoverned agent with access to your CRM or your financials is a completely different category of exposure.

And when governance finally shows up, what most companies are doing amounts to a policy document and a Copilot usage guideline. That’s not nothing. The other metaphor running through my brain is that this is like purchasing a giant bottle of chewable Airborne — 400% of your daily dose of Vitamin C and Zinc that you pop like candy. You take it because doing something feels better than doing nothing, and it might genuinely shorten the cold. But nobody taking it thinks it prevents the cold, or cures it once it’s caught. Most AI governance content doesn’t admit that part.

The harder question underneath the metaphor is visibility…which is why Microsoft is investing into the observability space. Some parts of an org you can see into. Other parts you can’t, until something breaks. Governance that only covers what’s visible isn’t governance. It’s a dashboard with blind spots.

—

If you’re looking for a conclusion to this story….sorry. As I said at the start of this post, these are just ideas rolling about in my head.

Maybe the honest ending is that we don’t get to skip straight to the vaccine this time. We’re still at the Airborne stage, arguing about whether we even need it. The Yammer story eventually resolved itself, mostly by acquisition and time. I’m not sure agentic AI gives us that same runway.

For now, take your zinc. Wash your hands. And go find out what’s already running in your org that nobody’s told you about.