• Home
• News

Can an AI Agent Replace the Call Center? Telnyx Brings the Question to Dubai

A customer calls a bank, airline or telecom operator.

There is no traditional menu asking the caller to press one for sales or two for support. There is no conventional chatbot waiting behind the phone line.

Instead, an AI agent answers the call, understands what the customer says, searches for the information it needs, responds in real time and potentially completes the requested task.

For Telnyx, this is becoming far more than another application of generative AI.

The communications technology company is heading to GITEX GLOBAL 2026 as a Bronze Sponsor with an increasingly ambitious proposition: the network carrying an AI conversation, the speech technology understanding it and the computing infrastructure generating the response should operate much closer together.

Telnyx is therefore expanding beyond programmable communications into AI inference, speech models, voice agents and the infrastructure needed to connect autonomous systems directly with customers.

That puts the company in an interesting position as enterprises across the Middle East explore one of the most practical applications of AI agents yet: answering the phone.

Voice AI Faces a Latency Challenge

Human conversation is highly sensitive to timing.

When one person stops speaking, the other usually responds almost immediately. A long pause can make the interaction feel unnatural.

That creates a significant technical challenge for voice AI.

A customer's voice must travel across a communications network. Speech-to-text technology needs to interpret it, an AI model needs to determine a response, and text-to-speech technology must convert that response back into audio before it travels across the network again.

Every stage can add latency.

A voice AI agent assembled from multiple providers may involve one company providing telephony, another handling speech recognition, another hosting the language model and another generating the voice.

Telnyx is betting on a different architecture.

The company wants more of those layers to operate within the same infrastructure.

Telnyx Wants to Control the Full AI Conversation

This is one of the most important elements of Telnyx's current strategy.

The company already operates communications infrastructure for programmable voice, SIP trunking, phone numbers and messaging.

It has now added AI inference running on its own GPU infrastructure, along with speech recognition, text-to-speech and AI-agent orchestration.

This allows developers to build voice agents without having to connect as many separate infrastructure providers.

The argument is partly about simplicity, but it is also about speed.

By bringing communications and AI computing closer together geographically and technically, Telnyx aims to reduce the delays that can make automated conversations feel artificial.

For enterprises, that could determine whether voice AI remains useful mainly for simple scripted interactions or becomes capable of handling more complex customer conversations.

Arabic-Speaking AI Agents Bring Voice AI Closer to the Middle East

One development makes Telnyx particularly relevant to GITEX's Middle Eastern audience.

During 2026, the company introduced Arabic speech models for its Voice AI platform.

The stack includes Basira for Arabic speech-to-text and Telnyx.Bayan for text-to-speech.

This enables businesses to configure AI assistants that can listen to Arabic-speaking customers, convert their speech into information the AI can process and generate spoken Arabic responses.

The potential applications are broad.

A telecom operator could use an AI agent to answer account-related questions.

A bank could handle routine service requests.

A healthcare provider could automate appointment interactions.

A retailer could provide order information.

An airline or travel company could use voice agents for reservations and customer support.

Because the system is connected to programmable telephony, customers can interact with the AI through an ordinary phone call rather than having to install a new application.

Multilingual Voice AI Could Transform Customer Service

Telnyx has continued expanding its speech capabilities beyond individual languages.

In September, the company added Soniox text-to-speech to Voice AI, bringing around 200 voices capable of speaking 60 languages through a single model.

One particularly notable capability is allowing the same voice to follow a caller when they change languages during a conversation.

This could be highly relevant in markets such as the UAE.

A customer might start a conversation in English and switch to Arabic, while another could move between English and Hindi.

Traditional automated phone systems are generally structured around selecting a language before entering the service.

Multilingual AI has the potential to change that model.

The conversation itself can determine which language is used.

GPT-Live Pushes Voice AI Beyond the Traditional Pipeline

Telnyx has also expanded the range of AI models available behind its voice agents.

Its platform supports a growing selection of proprietary and open-weight models, allowing developers to choose different combinations of intelligence, latency and cost.

During 2026, the company added models including GPT-5.6 Luna and GPT-5.6 Sol to its Voice AI Assistants.

In September, Telnyx also added OpenAI's GPT-Live, bringing a full-duplex speech-to-speech model to the platform.

This represents an important architectural shift.

The conventional voice AI pipeline can be summarized as:

Speech → transcription → language model → text → synthesized speech

Native speech-to-speech models can reduce some of those intermediate steps.

Instead of treating voice simply as text entering and leaving an AI system, the model can work more directly with audio.

That could eventually make conversations faster and more natural while preserving more information contained in the way people speak.

AI Agents Are Learning to Understand More Than Words

Voice contains information that text cannot fully capture.

A customer may sound confused, frustrated or hesitant. They may interrupt, change direction midway through a sentence or pronounce a name differently from what a transcription system expects.

Real-world conversations are unpredictable.

Telnyx has therefore been adding technologies around turn detection, voice activity detection and conversational orchestration rather than treating speech recognition as a simple transcription task.

Its Voice AI Assistants can manage turn-taking and interruptions, while direct WebSocket conversations introduced in August allow developers to stream audio to an assistant from web and mobile applications without requiring a conventional phone call.

This expands the technology beyond the traditional contact center.

The same voice agent could potentially operate inside an app, website, connected device or other digital interface.

The Phone Call Is Becoming a Software Interface

This transition could prove more significant than it initially appears.

Software traditionally requires users to learn an interface. They click buttons, navigate menus, complete forms or learn commands.

Voice AI reverses that relationship.

The user simply explains what they want, while the software is responsible for understanding the request.

For companies, this could make the telephone — one of the oldest digital communication channels — unexpectedly relevant in the AI era.

Instead of disappearing, the phone number could become an access point for an intelligent software agent.

And unlike a new application, almost every customer already knows how to make a phone call.

Voice AI Agents Need to Do More Than Talk

The real transformation begins when an AI agent can take action after understanding the caller.

A useful customer-service agent may need to search a knowledge base, check an order, retrieve an account record, schedule an appointment or trigger another business process.

Telnyx's agent platform therefore supports tool calling and knowledge retrieval rather than limiting assistants to conversation.

The company took this concept further in October with Agent Browser.

The technology provides an AI agent with an isolated browser session through an API and allows developers to restrict the websites it can access.

This addresses a practical challenge.

Not every business system has a modern API. Some tasks still require opening a website, navigating through pages and interacting with a conventional web interface.

Controlled browser access could allow an agent to perform tasks on those systems without granting unrestricted access to the wider web.

Agent Browser Could Turn Voice Conversations Into Actions

Consider a customer calling to change an appointment.

The AI agent understands the request.

If the scheduling system provides an API, the agent can interact with it directly.

If it does not, a controlled browser agent could potentially navigate the relevant authorized web interface and complete the task.

From the customer's perspective, the experience is simply a conversation.

Behind that conversation, however, multiple AI systems may be listening, reasoning, retrieving information and interacting with software.

This is where voice AI begins moving beyond automated customer support.

It becomes an interface between people and software.

Telnyx Builds Its Own AI Inference Infrastructure

Another part of Telnyx's strategy may receive less attention than Voice AI but is strategically important.

The company now provides AI inference directly.

Developers can run supported language models through an OpenAI-compatible interface, while Telnyx operates GPU infrastructure across different regions.

In September, the company introduced different inference service tiers that allow customers to trade cost against latency depending on the workload.

The Priority tier is designed for interactive applications where response time is particularly important.

Importantly for GITEX, Telnyx has made Priority inference available in the UAE.

This gives the company's Dubai presence a regional infrastructure dimension.

For real-time AI applications, physical distance matters.

Running inference closer to users can reduce latency, while regional deployment can also become important for organizations concerned about where their AI workloads are processed.

AI Gateway Brings Governance and Cost Controls

As organizations adopt more AI models, controlling access to them becomes another challenge.

Telnyx introduced AI Gateway in October to provide companies with a centralized layer for managing inference access.

Organizations can create separate gateways for different teams, determine which models they can use and establish spending limits.

Guardrails can also identify or block sensitive information, including card numbers, IBANs, email addresses, phone numbers and other data in prompts or responses.

This becomes particularly relevant as AI moves from experimentation into widespread employee use.

Giving every developer unrestricted access to every model through shared API credentials can become difficult to manage.

Telnyx is attempting to bring the same type of controls enterprises already apply to communications and cloud infrastructure into AI consumption.

The Cost of Voice AI Could Determine Its Scale

The economics of voice agents will be just as important as their intelligence.

Funding a demonstration involving ten calls is relatively straightforward.

A bank, telecom operator or airline handling millions of customer-service minutes faces a very different calculation.

Every conversation can involve telephony charges, speech recognition, text-to-speech, model inference and orchestration.

Telnyx is attempting to reduce those costs by controlling more of the technology stack itself.

Its current Voice AI pricing starts at $0.05 per minute for the voice engine, including orchestration, speech-to-text and text-to-speech, while language-model inference and telephony are priced separately.

The company estimates that some production configurations can operate at around $0.06 per minute, although actual costs vary depending on the model, call profile, token consumption and additional services.

That pricing question matters.

If voice agents cost almost as much as human-assisted customer service, deployment is likely to remain selective.

If the economics improve significantly, AI could begin handling a much larger share of routine telephone interactions.

Why Dubai Is an Important Test Market for Voice AI

The Gulf provides an unusually interesting environment for voice AI.

Banks, airlines, telecom operators, government agencies, healthcare providers and retailers operate large customer-service organizations.

At the same time, the region has highly multilingual populations.

An AI agent in Dubai may need to interact with customers in Arabic, English, Hindi and other languages, sometimes within the same service environment.

That makes multilingual speech, low latency and regional inference more than technical specifications.

They become practical requirements.

Telnyx's combination of Arabic speech models, UAE inference infrastructure, global communications and multilingual voice technology therefore gives the company a clear regional story heading into GITEX.

What to Expect From Telnyx at GITEX GLOBAL 2026

Telnyx has not published a complete list of GITEX-specific product launches, so any unannounced technology should remain unconfirmed until the company officially announces it.

However, the timing of its recent releases provides a strong indication of what is likely to matter in Dubai.

Voice AI should be central to the company's presence.

Arabic speech capabilities give the platform clear relevance to regional enterprises, while GPT-Live and other new model options demonstrate how quickly the underlying AI technology is evolving.

Regional AI inference provides another important angle, particularly for organizations concerned about latency and where workloads are processed.

AI Gateway brings governance and cost control into the discussion.

Agent Browser, released only weeks before GITEX, also demonstrates where Telnyx ultimately wants its AI agents to go.

The first generation of automated phone systems asked customers to press buttons.

The next generation learned to understand what customers said.

Telnyx is betting that the next step will be much bigger.

The AI answering the phone will not simply talk to the customer — it will increasingly be expected to finish the job.