What did we learn at Google Cloud Next 2026?

Agents are starting to act, not just assist

One message that came through loud and clear at Google Cloud Next was that chatbots and retrieval-augmented generation (RAG) assistants are no longer enough. Enterprises are expected to treat agents as part of everyday operations.

In a telecom context, that might mean agents diagnosing network issues, helping resolve customer billing problems, triggering provisioning flows, or supporting contact center agents in real time. These are powerful use cases, and many operators are already experimenting with them.

But once agents move beyond recommendations and start taking action, things change quickly. They are no longer sitting on the edges. They are touching BSS and OSS systems, interacting with customer data, and influencing outcomes tied to revenue and compliance. At that point, any misstep becomes operationally serious, not just technically inconvenient.

That is where many telecom leaders are feeling tension.

Security needs to follow the agent, not the perimeter

Security was a major theme across sessions and discussions at Google Cloud Next, especially as AI systems become more autonomous. Traditional security models that focus on system boundaries start to break down when agents can dynamically call APIs, chain tools together, and operate across workflows.

For telecom operators, the questions are very practical. Which agents can access customer accounts. Which ones can make changes to services. How actions are logged, monitored, and stopped if something goes wrong.

What stood out is how much the conversation has shifted toward runtime controls. Agent identity, fine‑grained permissions, continuous monitoring, and clear audit trails are becoming core requirements. In regulated industries like telecom, these are not optional add‑ons. They are table stakes for deploying agents in production.

Governance is now the bottleneck

Another strong signal from the event was how quickly the problem has shifted from building agents to governing them. Many enterprises now know how to create agents. Far fewer are comfortable running large numbers of them across live systems.

That challenge is especially sharp in telecoms. Agents may span customer care, network operations, IT, partner ecosystems, and multiple regions, all with different regulatory constraints. Operators need clear visibility into which agents exist, what they can access, what decisions they can make, and how that behavior changes over time.

Governance frameworks built for people or traditional software were not designed for autonomous systems that run continuously and operate at scale. Updating those frameworks is becoming one of the most critical pieces of the puzzle.

Automation still needs people in the loop

One of the more encouraging aspects of the conversations at Google Cloud Next was the focus on balance. Full autonomy can deliver efficiency, but telecom leaders are clear that not every decision should be hands‑off.

Many operators are leaning toward tiered autonomy. Agents handle routine tasks within strict guardrails, while higher risk decisions involving billing, service disruption, or compliance are escalated to humans. That approach allows organizations to move faster while staying in control.

Designing for that balance early matters. Retrofitting it later is much harder.

Readiness will decide who wins

For telecom leaders, the lesson from Google Cloud Next is not to move slower, but to move smarter. The agentic era is real, and the upside is clear. Success will come from thoughtful deployment, not simply by moving fast.

The operators that get the most value will be the ones that invest in the foundations. Security built in from day one. Governance that scales across teams and systems. Platforms that give real visibility and control as agents multiply.

The gap facing the industry is not a lack of ambition or vision. It is the hard work of making AI agents safe, predictable, and trusted inside some of the most complex environments in the world.

Closing that gap is what makes the difference between experimentation and sustainable deployment.

Jessie-Lee Fry is vice president of growth and go-to-market at Amdocs' Data and GenAI Studio.