You Can't License an AI Agent

I had a conversation this week that reorganized how I think about the next few years of our industry. I was talking with Nicole Reineke, the Chief AI Officer at N-able. I asked her something to the effect of "Will Microsoft eventually sell us an 'agent' license, the way they sell us a Business Premium seat today?"

Her answer was no.

Now, before I unpack this further, let's settle on the key thing we were talking about here: AI Agents. An AI agent is software you hand a goal, not a script. A normal program does exactly the steps you write for it. An agent takes the goal, works out the steps on its own, uses the tools it has and has access to, and comes back with a result.

Nicole shared that an AI agent is not a product, but rather 3 things:

1. A set of instructions
2. A connection to a model
3. An output

That's it. You cannot license a set of instructions any more than you can license a recipe or a checklist. Anyone can generate one, and the agent itself has no obvious moat around it.

This raises the next question: if the AI Agent is not the thing you can sell, what is? The answer lies within in the Control Plane.

What a Control Plane is

Since my background at Net Friends is strongly rooted in networking, I have to create a metaphor from that world. Networks already separate these 2 jobs. The data plane, sometimes called the forwarding plane, is what actually moves the packets. The control plane is what decides where those packets should go and enforces the rules. The switch or router forwards the traffic, but the control plane builds the map that tells it how. Software-defined networking took this further and pulled the control plane out on its own, so one place could direct many devices.

AI Agents work the same way. The Agents are the data plane, doing the work. But the Control Plane sits above them, deciding which Agent runs, what data it is allowed to interact with, how many tokens it is allowed to spend, and keeping track of everything that happened.

The Agent is something cheap and copyable, while the Control Plane is where the real value sits. This is where you contain cost. This is where you orchestrate outcomes. This is where you allow or deny which AI Agents can run, or grant or restrict what they can access while they run. Whoever operates the Control Plane manages the overall outcomes, not whoever writes the AI Agent.

Why Per-Seat Pricing is Going to be Problematic

Most business software is priced per seat. Nearly everything we purchase or manage for our customers aligns with a per-user, per-month model. That model assumes that there is a human sitting in a seat somewhere doing the work.

But how do you put an AI Agent in that same seat? The Agent performs a task (like running a report) but since the Agent is just a set of instructions, there is not a clear place to apply the license to. Yes, you could associate a traditional User with the AI Agent, but this scales poorly since a single User likely will utilize multiple AI Agents in a variety of ways that are impossible to establish some sort of 1:1 relationship with.

You can see the problem, right? When AI Agents do the work that seats used to do, then the per-seat pricing has nothing to cleanly attach to. The industry has to find some new unit of value. And complicating this picture, nearly all businesses have become accustomed to predictable subscriptions.

That last point is the whole ballgame.

Predictability is not just a nice-to-have. It's the main thing that makes an AI service adoptable.

Most businesses over time will increasingly incorporate AI into their operations and infrastructure. However, they cannot do this if there's no way to forecast or predict costs. Any AI service offering must be simple to understand and have a clear formula that the client can reliably use to budget and forecast with.

What this Means for Managing AI in your Business

As AI Agents and AI Skills become more and more prevalent as business infrastructure, it's clear that they need to both be hooked into and managed by a Control Plane. I think the major software vendors who will win the next round of AI will be the ones that can figure out how to standardize the Control Plane and provide a straightforward path for businesses to utilize it and Managed Services Providers (MSP) like Net Friends to operate it on behalf of businesses.

Someone has to establish the AI Control Plane and the associated protocols that standardize the way the plane works. And someone also has to operate the Control Plane, deciding what the AI Agents can do, watch what they actually do, keep the spending from running away, and produce the report and audit trail on a regular basis. There's real work and value here, and I believe this is the direction where a notable portion of the MSP's value proposition is headed.

AI Agents are going to be commodities and likely given away freely. AI Skills might be more closely guarded as they are more akin to the specific way a particular business operates. But both will not likely be sold like traditional software was, since the real money is made through the AI Agents and Skills being put to use and consuming tokens. And all businesses will need someone orchestrating all this through the Control Plane.

At Net Friends, we do not see this change as a threat. We see it as the next version of the job we're the experts at. For our clients, we have always been the trusted operator responsible for technology. We're already responsible for keeping the lights on, securing the data and identities, and responding to problems. While the AI Control Plane is going to be a new thing to operate, operating it well takes the same discipline and skills that good award-winning MSPs already have.

I believe that all businesses that understand these concepts early, and the Managed Services Providers that build for this now, will both be very well positioned over the next few years. I very much intend for Net Friends to be one of them on both counts.