1. AI as support (alarms, dashboards, prediction)

2. AI integrated into flows (optimisation, healing, ticketing)

3. AI as the basis of the operating model (autonomous agents)

Although Latin America is currently mainly in stages 1 and 2 — with only a few isolated cases in stage 3 — I believe progress will be much faster than we think. Most operators, and even service providers, will adopt strategies where AI as at the basis of their operations. It is time for AI to take on a tangible role, leaving aside romanticism and traditional practices to focus on the productivity and speed that modern telecommunications demand.

In areas and domains such as PM, CM, NOC, and optimisation primarly, it is important to consider the role that new AI agents should play:

• Engineering co-pilots, not subsitutes

• Executors under supervision

• Analysts within the multi-vendor context

Today, we have countless KPIs that reflect performance level on various fronts. However, it is essential to have automated analysis of KPI degradations following every change. We must be aware that an action is not always the definitive solution. Occasionally, it will be necessary to retrace our steps but with the right tools, so that process becomes much more manageable.

In networks where two, three, or more vendors coexist, alarm correlation in the NOC must be clear to all engineers involved, so that communication between areas is truly efficient. Likewise, certain automated tasks will require prior validation before executing massive changes. This is where specialised engineers come in: to evaluate impact, timing, costs, and even working environment.

In activities such as optimisation, where parameter adjustments are made daily, the concept of controlled rollback becomes critical. Automating does not mean relinquishing all control but, quite the opposite, gaining traceability, repeatability, and speed in decision-making.

The big challenge: operational prompts, context and control

This point is crucial for our operations. If we get it right, everything flows; but a poorly orchestrated agent does not just fail occasionally — it fails on a large scale. Poorly designed prompts lead to inconsistent changes, daily rollbacks, and, ultimately, a loss of confidence in automation.

Therefore, organisations must focus on train their teams to build versioned prompts, training models with the real context of the network (3G, 4G, 5G, private networks, etc.), define strict security policies, and establish clear execution windows. Agents must operate under user schemes with well-defined limits of autonomy, in accordance with each team’s responsibilities. In live production networks, a poorly defined prompt can be as risky as an untested script.

Why Latin America needs agents more than futuristic speeches

The regional market is not immune to the shortage of senior engineers (a situation similar to that in Europe and North America), but it also faces highly reactive operations, intense pressure on OPEX, irregular traffic growth and poorly planned mass events among other factors.

The reality is that, as a region, we cannot afford to indefinitely expand our workforces or wait for the market to mature on its own. We need to rely on more efficient operating models starting now.

There is definitively a hybrid model that works today:

• Humans define the strategy

• Agents execute routine tasks, both basic and complex

• AI proposes but the engineer decides

• Automation + agent + human = True, secure closed loop

It is not just about more autonomous networks to reduce people’s workloads, but also about smarter operations, with technology use aligned to the business. The solution will never be to replace engineers, but rather to raise the level of operations, by assigning better priorities, eliminating low value repetitive tasks, and preparing for the new technological challenges ahead, whether it be 6G or any future disruption.