Organizations have built the inbound side at speed. They are racing to deploy AI voice bots that handle inbound calls, route inquiries, and resolve issues before a human agent ever picks up. The supply side is crowded, well-funded, and moving fast. Five9, PolyAI, Bland AI, Retell, Amazon Connect, and dozens more are competing for enterprise contact center budgets. All enterprise contact center platforms are deploying conversational voice AI at scale for inbound customer service.
And there is a major shift happening within that supply side too: from AI-assisted customer service to agentic AI. Unlike AI tools that support human agents in contact centers, agentic AI operates autonomously, coordinating tasks across systems, adapting in real time, and resolving complex cases from start to finish on its own. Most of the top players now have their agentic AI offering.
But here’s what almost nobody is building: the consumer side.
The consumer-side gap
Imagine delegating your next customer service call to an AI agent. You tell it: “Call my broadband provider, file a complaint about the outage last week, and get me a credit.” The agent calls, navigates the IVR, speaks to a human or bot, and reports back. You never wait on hold. You never repeat your account number three times.
The capability exists today. AI agents that file complaints, cancel subscriptions, and make inquiries are live and in use. But the deployment gap is stark: organizations have built the inbound side, and barely anyone has built the consumer-outward side for the agent that acts as your delegate.
That gap has a name. Gartner calls them machine customers, and, according to a Forbes article I was interviewed for alongside Gartner’s Don Scheibenreif, they are on track to control trillions in purchases by 2030.
The data is already there
In a recent survey of nearly 5,000 consumers in early 2025, Gartner found that 51% would willingly use an AI assistant to handle customer service interactions on their behalf.
“By 2029, Gartner predicts agentic AI will resolve 80% of common service issues without human intervention, reducing operational costs by 30%. By 2030, it predicts agentic AI will initiate 50% of all service requests.”
Read that last number again. Half of all service requests will be initiated by AI, not humans. Organizations are building for human callers. The caller of 2030 may not be human at all.
Who is doing this today?
On the consumer-delegation side, the field is thin. DoNotPay pioneered it: their AI calls customer service lines, navigates phone trees, and advocates for users in real-time voice conversations.
Pine AI (19Pine) followed, positioning itself as a per-task consumer delegate for bill negotiations and complaints. Newer entrants like Autocalls and Hiatus are moving into the space, but most players remain focused on outbound sales automation rather than consumer advocacy.
Partu AI operates at the intersection of both worlds, acting as a consumer delegate for service requests and inquiries while also supporting agentic commerce workflows. The distinction matters: most agentic platforms start from the organization’s side, optimizing for their own processes. Partu starts from the consumer’s intent, which is a fundamentally different design philosophy and a much harder problem to solve.
For agentic commerce, i.e., AI buying on your behalf, there is more activity. But for agentic service delegation—AI complaining, inquiring, and negotiating on your behalf—the market is still almost empty.
Apart from DoNotPay, Pine AI, and Partu AI, we are not aware of anyone else building (machine) customer journeys that account for the consumer’s viewpoint and intent at the level the market is heading toward.
Most organizations active in the agentic space are still starting from agentic commerce or AI visibility. Consumer service delegation, especially with voice, remains largely unaddressed.
The landscape, however, is shifting faster than any article can track. Visa just launched Intelligent Commerce Connect, a single integration enabling merchants to accept payments from AI agents across all major agentic protocols. American Express announced Agent Purchase Protection, extending its consumer guarantees to purchases made by registered AI agents on behalf of cardholders.
On the browser front, while Google’s new Skills in Chrome let users save reusable AI workflows for browsing tasks, more advanced AI browsers and extensions, including Perplexity’s Comet and Anthropic’s Claude in a Chrome extension, are already going significantly further, autonomously navigating websites, booking services, buying on your behalf, and handling multi-step tasks in the background while you do something else entirely.
None of this is voice-based service delegation yet, but it signals something important: the consumer-side agent is being normalized fast across every channel. Voice for service requests is the natural next step.
The financial infrastructure giants are not waiting. The browser is becoming an agent. A startup can ship a consumer-side tool in a day. The consumer-side gap that looked stable six months ago is closing in real time.
The frontier: AI calling AI
Here is where it gets truly interesting and genuinely complex.
What happens when both sides have deployed agents? A consumer-side AI calls a company-side AI. No human is in the loop on either end. Calls can loop. IVRs designed for humans fail. Authentication breaks down. And the user gets a vague summary of an inconclusive exchange.
And there is the deeper question underneath all of this: when an AI agent calls on your behalf, how does the organization on the other end know it is genuinely you behind the agent? I have explored this identity gap in depth in two separate pieces because it may be the most consequential unanswered question in the entire machine customer conversation.
But done well, it can indeed be frictionless and instant. The consumer agent formulates the request, the business agent routes it to the right system, the loop is logged and traceable, and the outcome lands in the user’s inbox before they’ve finished their coffee.
This is the 2026–2028 horizon. The organizations that figure out how to handle machine customers gracefully, designing M2M-ready APIs, agent-friendly entry points, and transparent outcome logs, will define the next era of CX.
Those that don’t will face rising call volumes, authentication failures, and a trust gap with a new class of non-human callers they were never designed to serve.
What CX leaders should do now
Three things, starting today—not when machine customers become mainstream, because by then it will already be too late.
- Assume the consumer-side gap closes faster than you expect. Fifty-one percent of consumers are already willing. The tools are being built. Design your service infrastructure for machine customers now, not after they arrive.
- Distinguish between agentic commerce and agentic service delegation. They are different problems, different trust relationships, and different compliance requirements. Your AI strategy may cover the former. It almost certainly underweights the latter.
- Build for the AI-to-AI handshake. Machine-to-machine protocols, voice authentication, and transparent outcome layers are not future concerns, but 2026 infrastructure requirements.
The next era of CX will not be defined by how well your AI handles human callers. It will be defined by how well your organization handles AI callers.
AI calling AI is the horizon I find most fascinating and most urgent. The organizations that design for machine customers gracefully and the services that act as genuine delegates for consumers will define what CX means next. That’s exactly what we’re building at Partu AI.
The question is whether you are building for that conversation or still optimizing for the one that’s already being automated away.
The article originally appeared on LinkedIn.