At CASA25, the Showcase Challenge wasn’t about polished slides or perfect demos. It was about ambition, direction, and showing where this industry is heading. Telnyx’s session delivered exactly that—even if the demo gods decided to test everyone’s patience along the way.

What looked like a simple appointment reminder call quickly revealed something far more interesting: a glimpse of how CPaaS is evolving from APIs and flow builders toward natural-language-driven platforms built for Voice AI at scale.

A Voice-First CPaaS, Built Before Voice AI Was Cool

Pete, leading the Telnyx showcase, has seen the CPaaS industry evolve from the inside. From witnessing Jeff Lawson demo early Twilio at an SF New Tech meetup, to becoming the first product manager for voice at MessageBird, to now helping build Telnyx’s European presence—he’s lived through multiple CPaaS eras.

That context matters, because Telnyx is not an SMS company adding voice as an afterthought. It started as a voice and number provider, building its own network over the past decade with a focus on high-definition audio, low latency, and global reliability.

That foundation turns out to be critical for Voice AI.

While many AI voice solutions stitch together third-party telephony, cloud AI, and infrastructure layers, Telnyx brings voice, messaging, numbers, storage, GPUs, and AI orchestration together in one platform. The result is not just convenience—it’s control over latency, quality, and reliability, which are non-negotiable for real-time conversational AI.

The Demo: When Voice AI Meets the Real World

The live demo aimed to show a simple but powerful use case: an automated voice assistant calling to confirm or reschedule a healthcare appointment.

The assistant did call.

It did understand intent.

And it also stumbled—mishearing a response and handling an edge case imperfectly.

That moment, instead of undermining the story, actually reinforced it.

Voice AI doesn’t fail because the idea is wrong. It fails when transcription, prompts, or latency aren’t yet tuned for messy, real-world human behavior: accents, hesitation, nerves, background noise. And as Pete openly explained, this wasn’t a network failure—it was a prompt depth and transcription issue. In other words: solvable problems.

The key takeaway wasn’t the hiccup. It was what was happening behind the scenes.

Voice AI Without Developers (Yes, Really)

One of the most compelling parts of the Telnyx story is who can now build voice applications.

The demo was created by a product marketing manager—not a developer—using tools like n8n and Supabase, connected to Telnyx’s Voice AI capabilities. No custom telephony code. No SIP gymnastics. No AI plumbing nightmares.

Instead, the logic is defined in natural language:

• What should the assistant do?
• What outcomes are allowed?
• How should edge cases be handled?

This marks a real shift in CPaaS evolution:

• First era: code-only APIs (powerful, but developer-exclusive)
• Second era: no-code / low-code flow builders
• Next era: natural language as the interface to CPaaS

Telnyx isn’t abandoning APIs or low-code tools. They’re adding a new layer on top—one where “speaking English to CPaaS” becomes a legitimate way to build.

Model Choice, Voice Choice, and Real Flexibility

Another important signal from the Telnyx platform: it doesn’t lock customers into a single AI stack.

Users can choose:

• LLMs (OpenAI, Google, Groq, others)
• Voice engines (Telnyx-native voices, Azure, AWS, ElevenLabs)
• Integration patterns (webhooks, Zapier, MCP servers)

That flexibility matters in a market where AI quality, cost, compliance, and sovereignty requirements vary widely by use case and region—especially in Europe.

Telnyx positions itself less as “the AI model provider” and more as the orchestration layer where Voice AI actually meets the network.

Why Voice AI Needs a Telco-Grade Backbone

One theme came up repeatedly during the Q&A: call quality.

Many customers experimenting with Voice AI start with tools like VAPI or standalone voice-generation platforms. Then reality hits—latency issues, dropped audio, unreliable call handling. That’s when they come looking for a provider that understands telephony deeply.

Voice AI only works when:

• Latency is consistently low
• Audio quality is predictable
• Fraud, abuse, and traffic anomalies are monitored in real time

Telnyx has been solving those problems for over a decade. Voice AI just raises the stakes.

From Enterprises to the Long Tail

So who is this for?

Not everyone at once.

Telnyx sees Voice AI adoption starting internally—tools for teams, operations, scheduling, data collection—and then moving outward to SMBs and mid-market platforms that want to embed voice intelligence directly into their products.

The real scaling lever isn’t selling to 50,000 restaurants one by one. It’s enabling platforms, MSPs, and SaaS providers to resell and embed Voice AI as a native capability. That’s why Telnyx is investing in channel partners—especially those already selling SIP trunks and looking for their next growth engine.

Guardrails Matter

Voice AI is powerful—and dangerous in the wrong hands.

Here again, Telnyx’s telco DNA shows. Fraud detection, traffic anomaly alerts, call-rate monitoring, and abuse prevention aren’t new features bolted on for AI—they’re already part of the platform.

That’s a meaningful differentiator as Voice AI moves from experiments to production.

The Bigger Shift: From Developers to Assistants

Pete closed with a line that stuck with many in the room:

We’re moving from “Ask your developer” to “Ask your assistant.”

That’s not just a tagline. It’s a reframing of who CPaaS is for, how products get built, and how value gets unlocked—especially in the long tail of businesses that never had access to this level of communication intelligence before.

The Telnyx demo may not have been perfect. But the direction was clear.

Voice is back.

Voice is intelligent.

And the platforms that truly understand voice—from the network up—are the ones best positioned to lead this next phase of CPaaS.

A phone call. That’s the point.

At first glance, what the audience saw during this CASA25 Showcase Challenge didn’t look very exciting. A phone. A dialer. A live call.

No app demo. No futuristic UI. No flashy slides.

But that was exactly the point.

What Adnan Saleem from Radisys showed was a real phone call—using the native dialer—backed by an agentic AI assistant that doesn’t live in an app at all. The AI lives inside the telco network itself. And once you realize that, everything about the demo changes.

This wasn’t voice calling plus AI. It was AI voice as a native network capability, available on any device, on any operator network, without downloads or friction.

For readers who want to go deeper into the architecture behind this approach, Adnan covered that earlier at CASA25 in his keynote: “Voice Is Back (and Smarter Than Ever): How Radisys Is Unlocking New AI Revenue Streams for Telcos.”

👉 https://casa26.amsterdam/2025/10/11/voice-is-back-and-smarter-than-ever-how-radisys-is-unlocking-new-ai-revenue-streams-for-telcos/

A personal AI assistant, reached through the dialer

Instead of opening with architecture diagrams, Adnan went straight into the experience. Using his native phone dialer, he called his personal virtual assistant.

The assistant already had access to private information Adnan had explicitly shared—documents, calendar entries, reminders—and could also draw on public data when needed. He asked about upcoming travel, reminders for purchases, and itinerary details, all through a simple voice interaction.

Turning a phone call into a shared AI experience

The pivotal moment came when Adnan asked the assistant to place an outbound call to his colleague Ralph.

The call connected like any normal voice call. No applications on either side. Once Ralph joined, Adnan invited the virtual assistant into the live call, turning it into a three-party conversation. The AI actively participated—suggesting restaurants near Amsterdam Central Station, answering follow-up questions, and staying fully aware of the context of the discussion.

Automatic summaries, actions, and memory

When the call ended, the system automatically generated a call summary, action items, and a full transcript. These were delivered instantly via SMS with a secure link.

Meetings were identified, dinner plans captured, and next steps outlined—without anyone taking notes or switching tools. Adnan then showed how the call transcript could be imported into the assistant’s private knowledge base.

To close the loop, he called the assistant again and asked about his upcoming meetings and dinner plans. The AI responded instantly and accurately, using the context from the call that had just taken place.

Why “native” matters

What made this demo stand out was not just the AI, but where it lived.

This wasn’t an app layered on top of voice. It was native to the network itself. There was nothing to install, no interoperability issues, and no reliance on high-end smartphones. The service worked on any device, on any operator network, wherever voice works.

That makes it always available—and inherently inclusive.

The operator opportunity

During the Q&A, the conversation naturally shifted to monetization. From an operator perspective, how does this turn into real revenue?

Adnan outlined two clear paths. At the consumer level, this becomes a value-added service—an incremental monthly charge for a powerful personal productivity assistant. At the SMB and enterprise level, the value increases significantly: automatic call summaries, embedded business intelligence, action tracking, and AI-assisted workflows tied directly to everyday voice conversations.

Questions around AI accuracy, data privacy, and enterprise readiness were addressed directly. These are real concerns, but also expected steps on the path to large-scale commercial deployment—not fundamental blockers.

The bigger takeaway

The broader takeaway from this CASA25 Showcase Challenge was simple but important.

Voice doesn’t need to be reinvented. It needs to be elevated.

When AI voice is built directly into the operator network, the network itself becomes the platform—unlocking a new class of services that are personal, contextual, and ready to scale.