.svg)
There is nothing wrong with “agentic AI” except it doesn’t explain anything new.
Nothing will compete with peak AI hype but coming out of MWC Barcelona 2025, agentic AI is certainly having a moment. An industry well aware of the pitfalls of tech hyperbole now sees AI-powered agents as the key to unlocking automation, efficiency and intelligence across telecom networks.
Here we go again. Another buzzword and unicorn-like promise of transformation.
An honest question as we ramp up for another buzz cycle: Do we actually know what agentic AI is?
Because it is sounding a lot like automation rebranded.
Let’s explore how agentic AI is being framed today, what is truly different this time and the deeper shift in how software is evolving that could make agentic AI more than just a marketing refresh.
It's likely the term “agentic” is resonating because it feels human. Imagine, AI programs that can act independently, make decisions and even work together like a trained workforce!
That’s a compelling narrative for an industry constantly on the hunt for ways to reduce operational costs, improve service delivery and automate complex network functions. But it is not a new storyline for telecom, which has been chasing automation and efficiency for years.
Many of the big software players are strategically embracing the agentic AI wave with sights set on reducing reliance on human labor, selling more software and positioning at the heart of digital transformation.
For many telco influencers, agentic AI is seen as the next leap in network automation, a nice big serving of self-learning, self-healing and self-optimization that can reduce manual intervention, improve customer support and drive infrastructure management efficiency.
Here are some of the primary use cases being discussed:
These are all interesting. But each of these areas were previously known targets for improvement.
Is agentic AI truly revolutionary? That’s a tough case to make when its distinction from the automation strategies telcos have been using for years remains unclear.
We’ve already seen AI-driven self-optimizing networks (SON), closed-loop automation and AI-based customer service chatbots. American Airlines has been making us talk to machines for years. So is today’s agentic AI just a better version of the same thing?
Let’s look under the hood to better understand. I asked xAI’s GenAI assistant Grok how it would code for agentic AI.
Grok chose to implement its beginner-level example using a rules-based algorithm.
In this answer lies the true paradigm shift and the transformational journey we have increasingly followed since AI started to deliver game-changing results.
And that highlights the real shift underway: not what agentic AI does, but how it’s built.
As we consider AI’s evolutionary path, it is helpful to understand the underlying shift taking place from software 1.0 where code is written line-by-line, describing logic rules and conditions to software 2.0 where goals and architecture are defined and then models are trained on data. In the latter, logic is learned, not hand-coded.
The transition from software 1.0 (i.e., algorithmic, rule-based programming) to software 2.0 (i.e., data-driven development) marks a significant shift in how complex systems are built.
Instead of manually coding logic for every scenario, software 2.0 leverages machine learning and vast datasets. Neural networks are trained to recognize patterns and make predictions, enabling systems to operate without explicit instructions. This approach delivers adaptability and generalization, allowing software to handle novel situations, improve over time and mimic human-like decision-making. In place of static, brittle code, we get scalable, learning-based systems.
Tesla’s Full Self-Driving (FSD) Version 12 exemplifies this shift using billions of video clips from its vehicle fleet to train neural networks for perception and planning, a big departure from earlier algorithmic versions.
Software 2.0 enables systems to navigate complexity not by hardcoding every edge case, but by learning from data. This pattern-based understanding of reality opens the door to more flexible, autonomous behavior.
The true promise of agentic AI is to build mental models and frameworks that make developing these kinds of intelligent, adaptable systems faster and easier. It’s about turning advanced AI into a repeatable, scalable approach rather than a bespoke engineering challenge.
From this perspective, agentic AI shares similarities with web 2.0, which democratized software development by introducing toolkits and frameworks that made it easier to build interactive applications. That shift enabled the app-driven economy.
Agentic AI could do the same for the AI-driven economy upon us.
As we seriously consider anointing agentic AI as the next savior of telecom there are a few important questions to ask:
Agentic AI may ultimately define the next generation of telco automation. But not as a silver bullet. It’s only helpful as a framework for designing systems that learn, adapt and interact.
Remember, while AI capabilities evolve, the telecom’s complexity, scale and efficiency challenges remain unchanged.
The next significant shift in network management won’t come from new unicorn terms but from software that’s built differently, on the principles of software 2.0 and in collaboration with communities that accelerate innovation through shared toolkits, frameworks and open source. That’s what made Web 2.0 transformative and it could be what makes agentic AI more than just another buzzword.
Uber didn’t change the taxi industry because of web 2.0. It recognized the failings of the status quo and used web 2.0 to change it. The future AI leaders in all industries will do the same, either to strengthen existing domination or displace the untouchables of before.
