How to suss out the fake agents at MWC25

• Share via Facebook
• Share via Linkedin
• Share via Twitter
• Copy URL

Gartner recently predicted that by 2028, a third of enterprise software applications will include artificial intelligence (AI) agents. The AI agent revolution isn’t coming—it’s here. and your vendors are in full-on panic mode trying to catch up. 

Remember MWC24? Overnight, every vendor suddenly “had AI.” This year, they’ll all claim to have AI agents. It’s the predictable evolution of tech hype, and I get why it happens. Vendors desperately want to appear cutting-edge without actually investing in the hard, expensive work of building something genuinely useful.

As you wander through the Fira this year, how will you separate the innovators from the imitators? Stick with me—I’ve got your back.

What actually makes an AI Agent?

AI agents aren’t just chatbots with fancy marketing. They’re autonomous programs that do real work while you focus on more strategic priorities. True AI agents:

• Gather information independently across multiple systems
• Analyze complex data sets to extract meaningful insights
• Make decisions based on both historical patterns and current context
• Take concrete actions without constant human supervision
• Adapt to changing conditions and learn from experience
• Integrate seamlessly with existing workflows
• Collaborate effectively with other agents as part of a system

For reference, check out OpenAI’s Operator and deep research agents. Operator handles web browser tasks like filling out forms and finding hotels, while deep research conducts multi-step internet research and compiles comprehensive reports in a fraction of the time it would take even skilled humans.

Here’s the uncomfortable truth your vendors won’t admit: building legitimate AI agents is extraordinarily difficult. It requires fundamentally rethinking how software works. You can’t just bolt an agent onto a legacy system, slap on a new UI, and declare “mission accomplished.”

I’ve been deep in the AI agent trenches while working with the Totogi team on our revolutionary BSS Magic product, and I’ve learned enough to help you cut through the inevitable smoke and mirrors at MWC.  Let me arm you with the tough questions that will expose who’s really building game-changing tools and who’s just riding the hype wave.

Step 1: Test vendors’ LLM sophistication

Imagine this scenario: You approach a vendor’s stand, and a sales rep eagerly asks if you’ve heard about their “revolutionary new AI agent.” Your first move is to assess their fundamental understanding of large language models (LLMs)—the technology powering most modern AI applications.

Use this handy interrogation guide to quickly filter out the pretenders:

As vendors answer, watch their body language and comfort level. Those with real capabilities will eagerly demonstrate them and openly discuss current limitations. Those selling vapor will retreat to buzzwords, marketing claims, and vague promises about future releases.

Step 2: Make the agent pass the tests

Now that you’ve separated the LLM novices from the experts, let’s move on to assessing the core of what makes a true AI agent valuable in telecom: Find out how they work, integrate with existing systems, learn, and resolve issues. These questions will help you look “under the hood,” and tell you if the agent is doing real work, or is mostly smoke and mirrors cobbled together for MWC25.

My guess is you’ll find over 80% of ‘AI agents’ in telecom today are just glorified automation scripts with basic natural language interfaces. True agentic systems with reasoning capabilities will represent less than 5% of what you’ll see at MWC.

Question 1: How does it make decisions? 

AI agents can do real work because they’re trained to “think” through situations via a reasoning process. That process may include gathering information and comparing it to a knowledge base or decision tree before making a decision. Vendors with legitimate AI agents will be able to show you the reasoning behind the actions. A good answer will show clear decision logic. Red flag: black box responses. (They could be pre-programmed.)

Follow-up questions:

• What tasks or processes can your AI agent handle 100% on its own?
• If it gets stuck, what happens? Does it require human intervention to proceed?
• What level of autonomy does it have in high-stakes or regulated scenarios?
• What happens if the agent behaves unpredictably or erroneously?
• How do you retrain your agents as new regulations emerge or industry standards evolve?

Basic: Can handle simple, repetitive tasks, such as data lookups, and operate on its own with little supervision and oversight.

Intermediate: Can perform straightforward predictive or classification tasks within a single domain or department.  

Advanced: Capable of multi-step reasoning, error detection and self-correction within its domain.

Question 2: Can you show me this agent learning from its mistakes?

Agents should be able to learn from data and experiences, adjusting and improving performance over time. Ask vendors how this process works for their agents. Can they demonstrate a real knowledge update in action? A good answer: “Here’s a mishandled case from last week, and here’s how the agent handles similar cases differently.” A red flag: “It’s constantly learning everything automatically.”

Follow-up questions: 

• How does it gather performance feedback?
• How do you measure performance improvements?
• Can your agent give itself a grade?

Basic: Little to no learning capability, requiring manual updates when rules or requirements change (this is programming, not AI).

Intermediate: Can adapt to a basic level of changing data patterns, though retraining might require human intervention.

Advanced: Continuously learns from new data and user feedback in real time.

Question 3: How long does it take to train your agent on a new telco system?

AI agents can operate across separate systems, gathering info and using it to complete tasks, just like a human would. The more systems they integrate with, the more useful they become. The best will pull information from sources beyond the vendor’s own ecosystem. Beware of AI agents that are “single siloed,” like Salesforce’s Agentforce. Unlike these walled garden approaches, true telecom transformation requires agents that work across your entire stack, from your Amdocs billing system to your Ericsson network elements.

Structural design reveals how tasks, data, and decision-making flow through agents or systems. Real-world deployment requires seamless integration so agents can act on relevant data in a business environment. You want to ensure that adding more agents or tasks won’t degrade the system or lead to bottlenecks.

Follow-up questions:

• Describe the architecture that supports your autonomous agents. Are they centralized (orchestrator-based), decentralized, or a hybrid? Why?
• How do your AI agents integrate with existing business systems (OSS, BSS, CRMs, network platforms) to gather data and trigger actions?
• What sort of domain knowledge or knowledge graphs do your AI agents use to inform decisions on telco-specific tasks?
• What simulation or “digital twin” environments do you use to stress-test your AI agents’ behavior before deploying them in production?

Basic: Agent integration with business systems is minimal or done via simple APIs or file exchanges like CSV uploads.

Intermediate: Agent communicates with various internal systems through well-defined APIs or event-driven architectures in a scalable deployment on a cloud platform.

Advanced: Agent works in a highly distributed, cloud-native architecture with sophisticated orchestration and real-time data exchange across multiple domains and platforms.

Question 4: Can you show me two of your agents disagreeing on something, and how they resolve it?

Not only can AI agents do their own work, but they can work together in teams with other AI agents, collaborating via messaging protocols or shared memory. The most advanced agents can coordinate in real time, in parallel or in serial flows.

When you ask this question, listen for a description of real coordination protocols vs. marketing fluff. A good answer: “Here’s a case where the billing agent and retention agent had different recommendations…” Red flags: “Our agents never disagree,” or “The AI automatically resolves everything.”

Follow-ups:

• How do your AI agents coordinate with each other to complete tasks?
• How do you monitor multi-agent systems to operate efficiently and remain synchronized?
• Are your multi-agent systems scalable? What happens as the number of agents or tasks grows?
• Do you employ reinforcement learning, evolutionary algorithms, or other techniques to optimize agents’ strategies?

Basic: Multiple agents operate serially, but with limited direct interaction or coordination outside of simple data handoffs.

Intermediate: Agents actively communicate, coordinate, or share information to accomplish tasks as a team.

Advanced: Agents engage in a highly cooperative ecosystem using sophisticated coordination and negotiation to achieve shared or competing goals. Agents can learn from each other’s experiences to get better collectively.

Question 5: Can I type in my own query?    

The real test of all demos—does the vendor put you in the driver’s seat? If it’s a real agent, they should step aside and let you take over the keyboard (or microphone for some “vibe coding”!). Watch for a look of fear in the vendor’s eyes when you take the reins. A vendor with a real solution will show comfort with unscripted scenarios. 

Step 3: See the real deal with Totogi at MWC25!

Don’t just survive MWC25—come armed with questions that expose the difference between real innovation and vendor theater. Stop by the Totogi stand in Hall 2, Booth 2G51, and ask me personally to demonstrate what a truly agentic BSS looks like. I’ll show you the future of telecom software without the marketing fluff. 

We’ve been using BSS Magic in our own business: on our CloudSense customer deployments. We’re seeing an easy cost reduction for our projects of 50%, and we’re delivering them 2x faster. With BSS Magic, our customers are getting an AI platform that makes their entire telco tech stack work as one system—enabled, in part, by a telco ontology and data model that acts as a universal translator. We built it on open standards, like TM Forum’s Open Digital Architecture and APIs. It provides one intelligent layer—BSS Magic—that sits on top of your existing solutions, understands your business, and makes everything work together. 

And don’t miss my must-see talk at MWC25’s GenAI Summit. Join me in Hall 6 on Monday, March 3, at 10:30 AM CET to hear how AI is about to turn telco software on its head—and why Amdocs should be very, very worried.

Now you’re ready for MWC. See you there!

Recent Posts

1. The AI-first telco won’t wait 5 years for Amdocs
2. Amdocs is ruining your business to keep their empire. AI to the rescue!
3. Totogi made a money machine
4. Your single-model AI strategy is costing you millions
5. Signing a massive AI contract could be your biggest mistake of 2024

---

---

Get my FREE insider newsletter, delivered every two weeks, with curated content to help telco execs across the globe move to the public cloud.