In the last 18 months, the idea of AI voice agents managing real interactions for businesses has gone from science fiction to reality. Thousands of companies, from SMBs to enterprises, are using voice AI to schedule appointments, complete bookings, run surveys, do intakes, and much more. These agents save costs for businesses, generate additional revenue, and free up human employees to do higher leverage—and more enjoyable—tasks.
But we're still in the earliest innings. Most companies deploying voice AI today are in what an Andreessen Horowitz executive recently termed the "voice-as-a-wedge" phase—using it to automate one or two narrow call types as a point solution. The real opportunity lies ahead: voice agents that manage entire workflows, operate across multiple modalities, and eventually own full customer relationship cycles from first touch to long-term retention.
Why Voice, Why Now
The sudden viability of voice AI stems from a confluence of factors that all matured at roughly the same time.
First, the underlying models got dramatically better. The jump from GPT-3.5 to GPT-4—and the subsequent improvements across foundation model providers—meant that AI could finally handle the ambiguity, context-switching, and nuance that real conversations demand. A customer calling to reschedule an appointment might also mention a billing question, express frustration about a previous experience, and ask about a new service. Modern LLMs can navigate these shifts gracefully in ways that older NLU systems simply couldn't.
Second, latency dropped to acceptable levels. Voice is uniquely intolerant of delay. Even a 500-millisecond pause feels unnatural. But optimizations in inference, streaming responses, and text-to-speech synthesis have brought round-trip latency down to the point where conversations feel genuinely fluid. This wasn't true even two years ago.
Third, tool use and function calling became reliable. An agent that can only talk is of limited value. The real unlock comes when voice agents can take actions: check a calendar, update a CRM record, process a payment, send a follow-up email. The ability to call APIs and operate across systems transforms voice from a novelty into infrastructure.
Finally, businesses are ready. Labor costs continue to rise, hiring remains challenging in many sectors, and customer expectations for responsiveness have never been higher. The economic case for voice AI has become obvious to operators across industries. One example is the popular voice assistant Sloane which had explosive adoption by businesses in 2025.
Voice Solutions Today
Today, most voice AI deployments follow a familiar pattern. A company identifies a high-volume, relatively straightforward call type—appointment scheduling, order status inquiries, lead qualification, after-hours answering—and deploys an AI agent to handle it. This works remarkably well. Businesses report handling rates of 70-90% for these targeted use cases, with significant cost savings and improved customer satisfaction scores.
But this is just the entry point. The strategy makes sense for a few reasons: it's easier to sell a specific solution to a specific pain point, it's easier to build when the scope is constrained, and it's easier to measure ROI when you're replacing a clearly defined function.
The limitation is that point solutions capture only a fraction of the value that voice AI can ultimately deliver. A scheduling agent that can book appointments but can't answer questions about services, handle complaints, or identify upsell opportunities is leaving enormous value on the table. It's also creating friction—customers who call with complex needs still get routed to humans, or worse, get told the agent can't help them.
The Evolution: From Calls to Workflows to Relationships
The next phase of voice AI will be defined by expansion along three dimensions: workflow depth, modal breadth, and relationship scope.
Workflow depth means handling not just the call itself, but the entire process around it. Consider a home services company. Today, a voice agent might schedule the appointment. But the full workflow includes confirming the appointment via text the day before, handling rescheduling requests, dispatching the technician, collecting payment, requesting a review, and following up on any issues. Each of these touchpoints is an opportunity for voice (or voice-adjacent) AI to add value. The companies that will win are those building agents that can orchestrate the entire workflow, not just answer the phone.
Modal breadth recognizes that voice is often just one channel in a broader interaction. A customer might start with a text message, escalate to a phone call, receive a follow-up email, and later chat through a web interface. Agents that can maintain context across these modalities—understanding that the person texting about their order is the same person who called yesterday—will provide dramatically better experiences than those siloed into a single channel.
Relationship scope is the most ambitious expansion. Rather than handling individual transactions, voice agents can begin to manage entire customer relationships. This means understanding a customer's history, preferences, and patterns; proactively reaching out when appropriate; recognizing when a customer is at risk of churning; and building genuine rapport over repeated interactions. The agent becomes less like a call center rep handling tickets and more like an account manager nurturing relationships.
The Integration Imperative
None of this is possible without deep integration into business systems. An agent that can only access limited context will provide limited value. But an agent with access to the CRM, billing system, inventory database, scheduling platform, and communications tools can do genuinely useful things.
This is where the current landscape gets interesting. The first wave of voice AI companies largely built around proprietary models and closed systems. The next wave will be defined by integration breadth—how easily can an agent plug into Salesforce, HubSpot, Square, Toast, ServiceTitan, or whatever vertical-specific systems a business runs on?
The companies that build the most robust integration layers will have a significant moat. Not because integrations are technically difficult (though they can be), but because they require understanding the specific workflows and data models of each business system. This is unsexy, grind-it-out work that compounds over time.
Vertical vs. Horizontal
A natural question is whether voice AI will be won by vertical specialists or horizontal platforms. I suspect the answer is both, but with vertical players having an advantage in the near term.
The reason is that voice interactions are deeply domain-specific. The way a dental office handles appointment calls is different from how a property management company handles maintenance requests, which is different from how an e-commerce company handles order inquiries. The scripts, the integrations, the edge cases, the compliance requirements—all of these vary by vertical.
Vertical specialists can build these domain-specific capabilities from day one. They can speak the language of their customers, understand their workflows, and build integrations with the systems those customers actually use. This creates faster time-to-value and higher customer satisfaction.
Over time, horizontal platforms may emerge that are flexible enough to serve multiple verticals well. But I expect this will happen through a "vertical then horizontal" playbook rather than trying to be everything to everyone from the start.
What To Look For
Given this thesis, here's what excites me in the space:
Companies moving beyond the today’s configurations. I want to see founders who are thinking about the full workflow and the full relationship, not just the initial call type that gets them in the door. This doesn't mean they have to build it all at once—in fact, they shouldn't—but they should have a clear vision for expansion and a product architecture that supports it.
Deep vertical expertise. The best founders in this space have either operated in their target vertical or have spent significant time understanding it. They know the specific pain points, the existing systems, and the buying process. They're not building generic voice AI and hoping to find a market.
Integration-first architecture. The technical approach matters. Companies that treat integrations as an afterthought will struggle to deliver the contextual, workflow-spanning experiences that create real value. I look for teams that are building integration as a core competency from the beginning.
Thoughtful approaches to trust. Voice AI is inherently more intimate than text-based AI. Customers expect that when they speak to a business, they're being heard and understood. Agents that feel robotic, that make obvious errors, or that can't handle reasonable edge cases will damage brand relationships. The bar for quality is higher in voice than in any other AI modality.
The Opportunity Ahead
We're at a fascinating moment. The technology has crossed the threshold of viability, early adopters have proven the model works, and the broader market is just beginning to understand what's possible. As the underlying models continue to improve—and agents can now call tools and operate across systems—there's no reason why every company shouldn't have voice-first AI products running and optimizing critical parts of their business.
The companies that will define this space are being built right now. Some of them will start as point solutions and expand into platforms. Others will go deep in a specific vertical and become the default infrastructure for that industry. A few will figure out how to build horizontal capabilities that genuinely work across contexts.
What unites them is a recognition that voice AI isn't just about answering phones more cheaply. It's about reimagining how businesses interact with their customers—making those interactions more responsive, more personalized, and more valuable for everyone involved.
