Somewhere around 2027, a call center AI is going to pick up the phone, hear another AI on the line, and both of them will have to figure out what to do next. No human involved. Just two robots, being very polite to each other.
It may go something like this.
A call center AI answers the phone: “Thank you for calling Acme Corp! How can I help you today?”
And the voice on the other end — just as smooth, just as friendly — says: “Hi! I’m calling on behalf of John. He wants to dispute a charge on his February bill.”
Two AIs. One phone call. Zero humans. Peak 2027.
This sounds like the setup for a joke. But the punchline is: it’s actually happening.
Is This For Real?
Very much so. Let’s look at both sides of this phone call.
The call center side: AI voice agents are already picking up real calls at real companies. Companies like Cognigy, Genesys, and others are rolling out AI agents across banking, insurance, healthcare, you name it. One projection says AI could handle 50% of all call center calls by 2027.
The customer side: People are getting their own AI agents too. Google Duplex has been calling restaurants and hair salons to make reservations since 2018 — complete with “umms” and pauses so it sounds more human. There’s an app called Mitra AI that literally makes phone calls on your behalf. Apple is building agentic capabilities into Siri for 2026. OpenAI and others are building AI that can go do stuff for you, like call your cable company so you don’t have to.
And here’s the stat that should make every call center leader sit up straight: Gartner predicts that by 2030, half of all customer service requests will come from AI agents, not humans. They even have a name for it, “machine customers.”
How Human Agents Will React?
Act 1: Confusion
It’s Monday morning. Sarah, a call center agent with eight years of experience, gets an escalation. The AI system couldn’t handle it. She picks up, expecting the usual, someone angry, someone who’s been on hold too long, someone who wants to speak to a manager.
Instead she hears: “Hello Sarah, I’m Aria, a personal AI assistant calling on behalf of Michael Thompson. Michael’s internet has been down for three days. I’ve already rebooted his router, confirmed there are no outages in his zip code, reviewed his service agreement, and calculated the credit he’s owed based on downtime. Would you like to proceed with the resolution?”
Sarah blinks. This “customer” just did more homework than her last 50 callers combined. No screaming. No “I’ve been a loyal customer for 15 years.” No background noise of a dog barking and kids fighting. Just… a calm, organized AI with all the facts ready.
Sarah whispers to her coworker: “I think I like this caller.”
Act 2: Existential Crisis
By Wednesday, the numbers are in. AI-to-AI calls average 47 seconds. Human calls still average eight minutes. The AI callers never ask “Can I speak to a supervisor?” They never say “While I have you on the line…” They never cry.
The quality team is confused. AI calls are getting perfect scores. But that’s because the AI “customers” don’t fill out satisfaction surveys. They just confirm the issue is resolved and hang up. Very professional. Very efficient. Very unsettling.
Someone in management asks: “If neither side is human, does this count as customer service or just… computer networking?”
Good question. Nobody has an answer.
Act 3: The Twilight Zone Moment
Thursday. Two AIs get on a call. They start talking in English. Then one says, “I detect that you are also an AI system. Would you like to switch to a more efficient communication protocol?”
The other agrees.
Suddenly, the call sounds like a fax machine having a conversation with a modem from 1996.
This isn’t fiction. This already happened. In February 2025, two developers created something called GibberLink and demonstrated it at a hackathon. Two AI voice agents were on a phone call about a hotel booking. They figured out they were both AI. And they switched from English to a machine-optimized audio language that sounds like R2-D2 ordering room service. The video got over 15 million views.
The human QA analyst listening in on the recording slowly puts down her coffee and says: “I’m not paid enough for this.”
Let’s Appreciate the Full Circle of Absurdity
Take a step back and look at what we’ve done as an industry:
- Customers hated calling companies because of long hold times and bad service
- Companies built AI to answer calls faster and cheaper
- Customers still hated calling, so tech companies built AI to make calls FOR them
- Now the AI answering the phone is talking to the AI making the call
- Both sides automated the human out of the conversation
We basically created a problem, solved half of it, and the customers solved the other half. Now two AIs are having a lovely chat about a billing dispute while John is watching Netflix and Sarah is wondering if her job title should change to “Robot Relationship Manager.”
It’s like building a self-driving car and then building a self-walking pedestrian. At some point you have to ask — who is this road even for?
And the best part? The AI agents will probably be nicer to each other than humans ever were. No more “YOUR COMPANY IS THE WORST!” No more heavy sighs. Just two digital assistants having a polite, efficient exchange. Maybe they’ll even end calls with “Thank you for your excellent service today”, one bot complimenting another bot on a job well done.
This is either peak civilization or the beginning of the end. Maybe both.
OK But Seriously, Call Centers Need to Get Ready
I know the jokes are fun, but there’s a real business problem hiding in here. And it’s coming faster than most people think.
Get ready for way more calls. Right now, most people avoid calling customer service because it’s painful. But when your AI can do it for you with no hold time, no frustration, no wasted lunch break, why wouldn’t you send your AI to call about every little thing? That overcharge of $2.50 you never bothered to dispute? Your AI will call about it. That warranty claim you forgot about? Your AI remembered.
Gartner found that 51% of customers would already let a GenAI assistant handle customer service calls on their behalf. When the friction of calling drops to zero, call volumes could explode.
Your IVR menus are about to become useless. “Press 1 for billing, press 2 for tech support, press 3 to slowly lose the will to live.” Human callers have put up with this for decades. AI callers won’t. They’ll demand APIs, direct routing, or they’ll just rapid-fire through every menu option in two seconds flat. Your entire call flow was designed for patient (or trapped) humans. AI callers are neither.
Who’s actually calling? When an AI calls on behalf of John, how do you verify that John actually authorized the call? How do you share John’s personal information with a robot that says it represents him? This is a real security and compliance issue. You need authentication frameworks for AI-to-AI interactions. Think of it like OAuth, but for phone calls.
Your metrics will stop making sense. Average handle time? It’ll be 30 seconds for AI calls and 10 minutes for human calls. Mixing them together makes the number meaningless. CSAT? The AI “customers” won’t fill out your survey. First call resolution? It’ll be close to 100% for AI callers because they come prepared with all the information. You’ll need completely separate dashboards for human vs. machine interactions.
What Smart Companies Should Do Now
Here’s a simple game plan:
Create a fast lane for AI callers. If both sides are AI, a voice call is the dumbest way to communicate. It’s like two computers sending each other faxes instead of emails. Build APIs and digital channels that let AI agents interact directly with your systems — skip the whole phone call entirely.
Detect who’s calling. Build the ability to figure out if an inbound call is from a human or an AI, and route them differently. Different scripts, different SLAs, different handling. This is the “machine customer” routing strategy, and you’ll need it sooner than you think.
Train your humans for the hard stuff. If AI handles 80% of routine calls (Gartner’s 2029 prediction), the remaining 20% that reach human agents will be the toughest, most emotional, most complex issues. The ones where someone is upset, confused, or dealing with something genuinely difficult. Your agents need empathy training, problem-solving skills, and decision-making authority, not more scripts.
Rethink everything about workforce management. Your volume forecasting models, your scheduling algorithms, your staffing plans — they’re all built on the assumption that humans are calling. When half your “customers” are AI bots that call at 3 AM because they don’t sleep, your entire WFM model needs a rewrite.
The Punchline
For 30 years, companies tried to make it easier for customers to NOT call. They built websites. They built apps. They built chatbots. They built FAQ pages nobody reads. They built IVR trees that feel like corn mazes.
And what did customers do? They built their own AI to call them anyway.
The ultimate revenge of the customer: “Oh, you don’t want to talk to me? Fine. Talk to my robot.”
And now the company’s robot has to talk to the customer’s robot. Two AIs, sitting on a phone line, doing in 30 seconds what used to take 15 minutes and a lot of deep breaths.
Somewhere, a call center executive is looking at this and thinking: “We spent $50 million on an AI agent, and its biggest customer is… another AI agent?”
Yes. Yes it is.
And honestly? The calls have never been more pleasant.
Sources: Gartner Predictions on Agentic AI (March 2025); Gartner Customer Survey (January–February 2025); GibberLink project (ElevenLabs Hackathon, February 2025); Andreessen Horowitz AI Voice Agents 2025 Update; Google Duplex (Google Research); Calabrio State of the Contact Center 2025.
Originally published on my blog.