I’ve written seven posts in this series since last October. The first four walked the road from reactive support to responsive service, proactive assurance and autonomous optimization. The three after that asked what happens to the people, how the software gets built, and why none of it holds up unless the ecosystem is open.

Those posts laid the framework. Principles are how companies that don’t report to each other end up pointed the same direction, and I think we needed that agreement before any of this could get built on top of it. This one is the next step. We announced Mosaic One Fabric last month, so here’s what we actually built and what you’d do with it.

So what is it?

Mosaic One Fabric is an expansion of the Mosaic One platform that makes an operator’s network usable by AI agents. Your agents, built by your people, around your priorities.

Here’s how I’d put it to my VP if I were the network engineer. We want to build our own agents around the way we actually run this network. The models aren’t really the problem anymore. The problem is that a model knows nothing about our network. That gap is where the integration work goes. Fabric closes it. It presents our network, our operational data and Adtran’s telecom expertise as tools an agent can call, over MCP and REST. So our agents look at real state, reason with real telecom logic, and act, inside limits we set.

The tools arrive agent-ready rather than raw, low-level APIs somebody still has to wrap. That’s what turns a workflow into days of work instead of quarters. They reach across domains, because an agent is only as good as the context it sees. Governance is enforced at the tool – protecting the network from the agent and the agent from the network. And the token spend is yours.

What’s actually in it?

A new engineer isn’t much use on day one. They need access to see what’s happening. They need domain knowledge to understand it. And they need permission to touch things, handed over as they earn it. An agent needs those same three things, and that’s what Fabric provides.

The seeing part is the Digital Network Shadow, a cross-domain model of your estate covering the home, the access network, the aggregation and even into optical transport. Multi-vendor, so third-party hardware is in there too.

The understanding part is our intelligence functions. Deterministic root cause analysis. Cross-domain correlation. Service impact. Optical budgeting. These aren’t a general model guessing at telecom. These are Fabric’s intelligence functions, the same ones Clarity runs on, exposed as agent-ready tools, so the answers are repeatable and explainable. I’ve found that handing an agent an RCA tool makes it a lot smarter in a hurry.

The knowledge bases sit alongside them. Release notes, troubleshooting guides, product documentation, so an agent can look up what a card does or which release fixed a fault.

The permission part is tool governance, enforced at the tool itself rather than in a prompt somebody hopes the model respects. You decide what each agent can read and change, and open that up as you build confidence. That protection runs in both directions, an agent can’t overreach into the network, and the network can’t swamp the agent, because an event storm buries an agent as fast as it buries a NOC.

Mosaic One Fabric lets you build agents around your network, your workflows and your priorities.

We’ve been building on it ourselves. For example, one agent troubleshoots and tunes Wi-Fi in the home. It pulls live data through the MCP tools, mesh topology, interference, channel utilization, then calls root cause analysis to find where the problem sits. Anybody who has worked a Wi-Fi complaint knows the fix is rarely the hard part. Knowing what to fix is.

Another plans optical transport. It works over live optical data and calls our optical intelligence functions to check whether a planned service fits, what the margin looks like, where you’d need regeneration.

A third manages upgrades. It reads live firmware state across the fleet, then works the release note knowledge base for what each candidate release fixes and what it breaks. From there it designs the scenarios, which groups move, in what order, onto which release. Then it executes through governed tools, staged and bounded, with a person approving the plan first. Designing a change is one thing. Being trusted to execute it is another.

Those three are ours, running on the same tool plane you’d build on. Clarity runs there too, inside Mosaic One, asking the network questions through the same tools your agent would call. That’s the test I’d put to anybody selling you an open platform, us included. Does their own software come through the front door, or is there a side door?

And yes, we’re building more of these agents, for the frontline rep, the NOC engineer, the tech in the truck, the marketer running campaigns, the executive steering the business. More on those later this year. Because our agents run on the same tools you have, you can swap one for your own agent or ignore it and build something else. A sealed catalog offers neither.

The rest of it isn’t ours to build. Those three stop at the edge of what Adtran can see. An outage workflow that warns subscribers before they call and puts a work order in the field needs the billing platform and the mapping system agent-ready on the same terms. That invitation stands.

Where does that leave us?

Fabric is launched. It’s a foundation. I think what gets built on it is the part worth writing about. So that’s what comes next.

What workflow do you know you need and can’t buy? The one crossing your billing system, your access network and the ticketing queue that no vendor can see end to end? Or the workflow spanning the GIS and weather? Bring them to us.

And hold us to a few things. The tools we build our own agents on are the tools you get, nothing held back. You can build your own agents on your own models, so that you’re not beholden to our agents. Your own agents can work with your data, and the APIs stay open whether you use an agent or not. And you decide what any agent can touch.