Enterprise Connect 2026 opens today in Las Vegas — and the headline isn't about unified communications or collaboration platforms. It's about voice.
Amazon Connect, RingCentral, Zoom, and Dialpad are all showcasing agentic voice AI — autonomous agents that don't just answer questions on a phone call but reason through problems, take action across enterprise systems, and resolve issues without human intervention.
Voice has been the hardest enterprise channel to automate. Text-based AI can take its time, check its work, format a response. Voice is real-time, emotionally complex, and unforgiving. Customers expect immediate, natural responses. They expect the agent to know their history. They expect resolution, not transfer.
This week, the industry is demonstrating that voice AI has crossed the threshold from scripted interaction to autonomous action. And the numbers behind these deployments suggest the shift is already happening at a scale most enterprises haven't appreciated.
Amazon Connect: 20 Million Interactions Daily
Amazon Connect is the quiet giant of enterprise voice AI. Handling over 20 million customer interactions daily across voice, chat, and messaging channels, it operates at a scale that makes it one of the largest customer experience platforms in the world.
At Enterprise Connect, Amazon is showcasing its agentic self-service capabilities — AI agents that understand complex requests with multiple intents, reason through solutions, and take autonomous action on behalf of customers. A customer calls about an order issue and the AI agent greets them by name, asks clarifying questions, looks up order status, processes a refund, and adapts its tone to match the customer's sentiment — all without involving a human.
The architecture behind this is significant for enterprise AI more broadly. Amazon Connect now supports the Model Context Protocol — the same MCP standard that has become the universal connector for enterprise AI systems. This means voice AI agents can access customer profiles, case histories, knowledge bases, and third-party business systems through the same standardised protocol that connects every other AI agent in the enterprise.
Amazon's approach blends deterministic and agentic processing. Predictable, compliance-critical steps run through defined workflows. Complex, open-ended customer interactions use AI reasoning. The system can seamlessly escalate to human agents when the AI encounters situations beyond its confidence threshold — and when it does, the human agent receives full context from the AI interaction so the customer never repeats themselves.
Pasquale DeMaio, VP of Amazon Connect, described the philosophy: “Most will blend AI and human, and meet somewhere in the middle.” The AI doesn't replace the contact centre. It handles the routine at machine scale so human agents focus on the interactions that genuinely need human judgment and empathy.
Zoom Virtual Agent 3.0: 98% Resolution Without Humans
Zoom's announcement of Virtual Agent 3.0 may be the most striking proof point at Enterprise Connect. Built on Zoom AI Companion 3.0, ZVA can now automate multi-step workflows across CRM, billing, order management, and other enterprise systems — across both chat and voice channels.
The results from Zoom's own deployment are remarkable. Deployed on Zoom's support site, the Virtual Agent resolved 98% of customer interactions without escalating to a live agent. When voice capabilities were added, Zoom achieved a 76% containment rate and reduced the abandonment rate from 23% to just 1% within months.
Three architectural elements make this possible. First, ZVA integrates with enterprise systems including Salesforce, Zendesk, ServiceNow, Microsoft Dynamics, Intercom, and Kustomer — meaning the voice agent has access to the full customer context, not just a scripted FAQ. Second, Zoom built what it calls a “glass-box” approach to decision logic: administrators can see exactly how the AI reached each decision, which data sources it used, and which workflow paths it followed. Third, the system uses a no-code builder (Zoom AI Studio) where the agent's goals and available data sources are defined in natural language.
Metrigy's senior research analyst noted that this “glass-box” approach directly addresses the trust barrier that prevents enterprises from deploying autonomous voice AI. When leadership can audit every AI decision — see why a refund was processed, why a case was escalated, why a particular response was chosen — they gain the confidence to expand deployment.
Additional capabilities coming in Spring 2026 include multimodal intelligence — the voice agent will be able to interpret customer-submitted documents, images, and structured identifiers like serial numbers during a conversation. A customer can photograph a damaged product, send it during a voice call, and the AI agent processes the image and the conversation together to resolve the issue.
RingCentral: Agentic Voice AI in Production Workflows
RingCentral is taking a different angle at Enterprise Connect — showcasing its agentic voice AI portfolio through real customer stories rather than technology demos. The keynote, delivered by President and COO Kira Makagon, will focus on how agentic voice AI shows up in everyday workflows, driving intelligence before, during, and after conversations.
The “before, during, and after” framework is significant. Traditional voice AI focused on the call itself — what happens while the customer is talking. Agentic voice AI extends to preparing for the conversation (pre-populating customer context, predicting likely issues, suggesting resolution paths), managing the conversation (real-time sentiment analysis, dynamic response adaptation, live system access), and following up after the conversation (automated documentation, triggered workflows, proactive outreach based on conversation outcomes).
This full-lifecycle approach transforms voice from an isolated channel into an integrated enterprise workflow. The voice interaction becomes a trigger point that connects to CRM updates, case management, fulfilment systems, and customer success workflows — all orchestrated through AI agents that operate across the entire chain.
What This Means for Enterprise Operations
The convergence at Enterprise Connect tells enterprise leaders something important: voice AI is no longer a future capability. It's production infrastructure operating at scale today.
The Economics Have Shifted
When Zoom achieves 98% resolution without human agents on its own support operations, the economics of voice-based customer service change permanently. The traditional model — human agents handling every call, with technology supporting them — inverts. AI agents handle the volume. Human agents handle the exceptions. The cost structure shifts from linear (more calls = more agents) to fixed (AI infrastructure handles scale, human agents handle complexity).
For enterprises processing thousands of customer calls daily, this shift is measured in millions of dollars annually. Not in some future state — in deployments running today.
MCP Makes Voice Part of the Enterprise AI Fabric
Amazon Connect's adoption of MCP is particularly significant. It means voice AI agents now connect to enterprise systems through the same universal protocol as every other AI agent. A voice agent can access the same customer data, trigger the same workflows, and operate under the same governance frameworks as a document processing agent, a sales automation agent, or an internal operations agent.
This is exactly the architecture our Dewply platform is built on. Voice AI that operates as part of the enterprise AI ecosystem — not as a separate, siloed channel. Every voice interaction accesses the full enterprise context through standardised connections. When Dewply handles a customer call, it knows the customer's history, their open cases, their recent transactions, their preferred language, and their emotional state — because it connects to the same enterprise data layer that every other AI agent uses.
Transparency Becomes the Differentiator
Zoom's “glass-box” approach to AI decision-making is becoming the standard enterprise buyers demand. When an AI agent processes a refund, resolves a billing inquiry, or escalates a complaint, leadership needs to see why. Not as an afterthought audit — as a real-time capability built into the platform.
Our Dewply platform generates complete interaction records for every voice conversation — including AI confidence scores at each decision point, the data sources accessed, the escalation triggers evaluated, and the resolution path followed. This transparency isn't overhead. It's the feature that enables regulated industries — financial services, healthcare, government — to deploy voice AI with confidence.
Hybrid Reasoning Is the Architecture That Works
Both Amazon Connect and Zoom use hybrid approaches: deterministic logic for compliance-critical steps combined with AI reasoning for open-ended conversation. This isn't a compromise — it's the architecture that production voice AI requires.
A customer calling to update their address needs a deterministic workflow that validates identity, confirms the change, and updates records in sequence. A customer calling to complain about a billing error needs AI reasoning that understands the emotional context, investigates the issue across multiple systems, and proposes a resolution that accounts for the customer's history and the company's policies.
Our Minnato orchestration platform manages this hybrid routing across all enterprise AI operations — including voice. Each incoming interaction gets evaluated: does it require deterministic processing, AI reasoning, or a combination? The orchestration layer routes accordingly, applies the appropriate governance framework, and ensures human oversight where required.
Three Actions This Week
Benchmark your voice AI against the new standard. Zoom achieves 98% resolution and 76% voice containment. Amazon Connect handles 20 million daily interactions with agentic AI. These are the benchmarks your customers will compare you against. Audit your current voice operations: what percentage of calls could be handled by AI agents with access to your enterprise data?
Connect voice to your enterprise AI fabric. If your voice AI operates as a separate system from the rest of your enterprise AI, you're leaving value on the table. MCP-compatible voice infrastructure means every call benefits from the same customer data, workflow automation, and governance frameworks as your other AI operations.
Deploy transparency before you deploy autonomy. Leadership will approve autonomous voice AI when they can see how every decision was made. Build the audit and observability infrastructure first — then expand the autonomy.
“Voice was the last enterprise channel to resist AI automation. This week in Las Vegas, every major platform demonstrated that resistance is over. The question for enterprises isn't whether voice goes agentic — it's whether your voice AI can see what the rest of your AI knows.”
The Voice Channel Has Changed
For decades, the phone call has been the most expensive, most labour-intensive channel in enterprise operations. It required humans for every interaction. It was difficult to scale. It was resistant to automation because customers expected — and deserved — natural, empathetic, contextually aware responses.
This week, the technology caught up with the expectation. AI agents that understand not just what customers say but how they say it. Agents that access enterprise systems in real time to take action, not just provide information. Agents that know when to handle something autonomously and when to bring in a human — and that transfer the full context when they do.
The enterprises that deploy this capability first don't just reduce costs. They deliver better customer experiences — faster resolution, more personalised service, 24/7 availability — while freeing human agents to focus on the interactions that genuinely benefit from human connection.
Voice AI just went autonomous. Enterprise Connect 2026 is the proof point. The question is how quickly your operations catch up.