du and Comarch bring AI agents to the network: the value is in clean data, not the agent

By The Fast Mode · November 5, 2026.

Telecom operator du and Comarch Communications, a provider of connectivity software solutions, have expanded their partnership. Comarch will enhance du's OSS (Operational Support Systems, the systems that manage network inventory, configuration and faults) platform with autonomous AI agents. According to the release, the collaboration is now in its seventh year.

The centerpiece is a unified repository of network data, regardless of source, that is kept up to date automatically and gives du real-time visibility into its resources. On that foundation, the agents monitor network health, analyze logs from multiple vendors and resolve routine workflows.

The announcement describes four capabilities: detecting, synchronizing and restoring missing configuration data across systems from different vendors; diagnosing discrepancies and distinguishing vendor faults from configuration gaps; generating tickets and automatic fixes, with resolution times said to drop from days to minutes; and cataloging hardware by reading vendor manuals and updating libraries via API.

There is an important caveat in the fine print. According to the release, the initial deployment runs in sandbox environments (isolated, without touching the live network) with human verification in the loop. The architecture, however, is designed for future closed-loop autonomous execution. In other words, full autonomy is a stated aspiration, not a capability already deployed. Moreover, the "days to minutes" claim comes from the companies, with no independent figures or metrics in the material.

**Our take.** This kind of announcement illustrates a thesis we hold: the bottleneck for agentic AI is not the model, but data and control. An agent that corrects network inventories is only useful if the inventory is reliable; that is why the announcement devotes so much space to unifying data before talking about autonomy. Those with their data house in order will be able to automate; those without will automate the mess.

It is also an example of the sensible adoption pattern: sandbox, human verification and governance guardrails before letting go of the reins. Overall, in a sector where a configuration error can knock out service for millions of users, that gradualism is a virtue, not a brake. It ties into another recurring lesson: the unit of analysis for agent safety is the sequence of actions, and layered least-privilege permissions matter more than the agent's intelligence.

On the labor front, the short-term effect is as expected: technical back-office tasks, such as reconciling inventories or opening tickets, are the most exposed. The roles that gain are those that design, audit and govern these agents. The transition requires reskilling operations teams, and operators would do well to plan for it rather than discover it after the fact.

In the long run, self-healing networks mean fewer outages and more reliable services, a small piece of the infrastructure of abundance that underpins everything else. What to watch is whether du publishes measurable results once it leaves the sandbox: that is when it will become clear how much was reality and how much press release.