Phone companies start letting AI software fix problems on their own

Phone companies are starting to give AI software more of the work. Nokia, Nvidia and the Dutch carrier KPN are each exploring AI "agents" that can fix network problems and handle customer calls. An agent is software that can take steps toward a goal on its own, not just answer questions. For customers, that could mean talking to an AI first and getting faster fixes for dropped calls.

How AI could fix a dropped call

Nokia describes a system where engineers ask for help in everyday written or spoken language. It pulls together information and tools that engineers would otherwise check one by one. Nokia says it already uses this approach in its network design and improvement services.

Take dropped calls. An engineer could ask an agent to watch call failures and investigate when they pass an acceptable level. The agent could find likely causes, recommend a fix and carry it out with human approval. Then it would check the results and summarize what changed. Nokia has not shared measured results, so the speed gains are unproven.

What Nvidia says it takes

Nvidia says phone companies need shared data, tools and clear operating rules before agents can take on more. Many carriers still use automation only for set tasks. Nvidia pictures agents that look into unfamiliar problems, compare possible fixes and track results over time. Operators would review higher-risk changes before they go live.

In a simulated network problem, one agent finds the cause and ranks fixes. Another applies a plan, checks whether service improves and tries something else if needed. In a separate experiment, Nvidia says AI improved the use of wireless airwaves by more than 3% over an industry-standard method. These are early examples, not proof from live networks.

KPN puts AI on the phone

KPN, working with McKinsey and its AI business QuantumBlack, built voice agents for routine calls. They verify customers, check orders, manage technician appointments and troubleshoot internet connections. Human support stays available.

• KPN handles roughly 5 million customer calls a year.
• It aims for AI to handle 10% to 20% of service calls by 2027.
• The AI calls scored 83 on customer satisfaction, comparable to calls handled by people.
• McKinsey reports shorter handling times but gives no figures.

What it took to make it work

The team designed conversations to respond in under two seconds and let callers interrupt mid-sentence. Staff reviewed transcripts from up to 100 calls each day to improve the agents. Frontline employees helped build and test the service. KPN is also training its own teams to create more uses.

What is not known yet

The reports do not show how well this works across large, live networks. Nokia and McKinsey both leave their speed gains unmeasured. The KPN satisfaction score is the firmest number so far, and it comes from one company in one country.