October 28, 2026 • 10 min read
7 Best Practices for Dealing with Personal Agents In the Contact Center
Director of Content & Market Research
October 28, 2026
In September 2026, Meta launched Muse, a personal AI agent that can carry out tasks on a user’s behalf.
Within six weeks, it had reached ten million users.
Such personal agents aren’t new. Since Google Duplex first teased their potential in 2018, a number of them have emerged, with some, like Instinct, demonstrating just how far the technology can go.
But Meta’s reach and distribution could push them from the periphery to the mainstream.
Consider how Muse is available on WhatsApp, which has 2.3 billion daily users. That gives it enormous potential to scale.
However, Meta isn’t the only consumer tech giant innovating here.
Less than a month after Muse’s launch, OpenAI debuted Dots, its personal agent.
Meanwhile, smartphone providers like Apple and Samsung could be next, making personal agents available via on-device voice assistants such as Siri and Gemini.
As these consumer tech giants invest, personal agents will become increasingly common, raising important questions for organizations and their contact centers.
What Are Personal Agents?
A personal agent is an AI assistant that acts on a user’s behalf to complete tasks.
Take Muse as an example. It can connect with the user’s accounts to plan trips, shop, send emails, manage subscriptions, and more.
It also has a beta feature that lets it make phone calls.
Over time, personal agents will increasingly act as machine customers, buying products and contacting customer support on the users’ behalf. They’ll do so in two ways:
- They’ll follow a customer’s instructions to contact the business directly and pursue a specific outcome.
- They’ll try to complete a task on the company’s website and escalate to customer support when they fail.
As personal agents take on more tasks, contact centers will likely face increasing pressure.
How Will Personal Agents Impact Contact Centers?
Personal agents will fundamentally challenge how organizations engage with customers and build relationships.
However, contact center leaders should brace for the following three immediate impacts.
1. Hiked Contact Volumes
Last year, Forrester predicted that, in 2026, at least three major brands would each experience six instances of single-day call volumes spiking to 100 times their normal levels.
Why? Because AI agents will “flood” their call queues.
With the release of Muse and Dots, that prediction is looking increasingly on the money.
2. Increased Customer Persistence
Personal agents will make customers more persistent by pursuing issues, challenging decisions, and escalating complaints.
As a result, contact centers will face more sustained, informed, and harder-to-deflect demands.
3. Contact Center Analytics Woes
Personal agents make it easier for customers to send companies complaints, which they will largely generate themselves.
“These complaints are likely to be longer, more detailed, and potentially contain misleading information,” warns Maria McCann, Co-Founder of Neos Wave.
With contact centers using their own AI to process these interactions, analytics become increasingly synthetic, potentially distorting sentiment analysis, NPS, and CSAT scores. That reduces the reliability of customer insights fed back into the wider business.
Can Organizations Simply Turn Personal Agents Off?
Yes is the answer. In fact, Amazon already has. But a better question is: should they?
Imagine a business telling a human customer to leave because they don’t want to deal with them. Blocking personal agents is essentially the same practice.
“Those agents represent human customers. They might not be buying something for themselves, but they’re acting on behalf of someone who is.”
Moreover, Pihlaja notes how, when rejected, personal agents may turn back to the customer and recommend a competitor they can communicate with instead.
Given this, contact centers should focus on building out personal agent strategies, and many vendors are offering a helping hand…
The New Protocols for Personal Agents
To help contact centers manage interactions with personal AI agents, CX technology and personal agent providers are introducing new services and protocols.
Zendesk was among the first to act, introducing its Open Service Directory (OSD) to help personal AI agents find and verify a brand’s official support channels.
“The Open Service Directory gives assistants a trusted way to find that official route, while leaving brands in control of what an assistant can access and do.”
Zendesk also joined Sierra, Meta and others to launch the Personal Agent Protocol.
The protocol establishes rules and an identity system for personal agents, helping answer key questions, such as:
- Who does the agent represent?
- Has the customer authorised it?
- What is it allowed to do?
- Can it access personal information, make purchases, or cancel services on the customer’s behalf?
This information helps the agent and contact center navigate the interaction with greater clarity and trust.
Other providers are developing similar gateways to connect personal agents with businesses. Decagon has introduced its Personal Agent Gateway, while Cresta offers a Sentry Gateway.
However, while these solutions are still in early development, they currently only address part of the challenge.
Organizations must also define what agents can do, how consent is managed, when additional verification is required, and, most importantly, when a human needs to step in.
That takes us to the best practices…
7 Best Practices for Dealing with Personal Agents in the Contact Center
From conversations with prominent analysts, consultants, and thought leaders on all things personal agents, here are seven best practices for contact center leaders to keep in mind.
1. Send In Your Own Agents
One starting point is to send a friendly AI agent through existing customer service channels and see how well it fares.
In doing so, ask:
- Does anyone recognize that it’s an AI agent?
- Can it get through the front door?
- Do different channels produce consistent outcomes?
- Where does the journey stall?
- Are the customer’s goals and constraints preserved throughout the process?
- What happens when the company can’t provide the requested outcome?
It may be uncomfortable, but it’s a fast way to expose the gaps in current processes.
“It will at least give you an alarm bell that you can send up to somebody else and say, we can’t do it.”
2. Structure Online Information for Personal Agents
Many personal AI agents will only contact customer service when they can’t complete a process on the website first.
Typically, that online content is designed for human users and is not sufficiently accessible or structured for AI.
Consider a personal agent trying to resolve a billing dispute. If policies, FAQs, transaction data, and resolution options are scattered across systems or available only through difficult-to-navigate interfaces, the agent may fail and escalate unnecessarily.
To avoid unnecessary escalations, organizations should ensure that knowledge, policies, and transactional capabilities are accessible through machine-readable formats, structured data, and APIs.
In doing so, they’ll enable personal AI agents to find accurate answers and complete tasks autonomously.
“We measure how much of the website’s content is server-side rendered versus browser-side rendered. If content is rendered only in the browser, some agents may not be able to see it because they don’t execute JavaScript in the same way as a human browser.”
3. Establish Boundaries for the Interaction
When agents reach the contact center, protocols should be there to help establish who a personal agent represents, what it’s authorized to do, and whether it should be granted access.
However, those protocols don’t necessarily define the boundaries of the interaction itself.
Consider a customer who sends their personal AI agent to resolve an issue with a damaged delivery. They authorize it to seek a replacement or a refund, but explicitly rule out store credit.
There isn’t necessarily a mechanism to communicate and enforce these constraints throughout the interaction.
Contact center leaders should therefore consider how their AI agents can scope out these constraints, align on a shared goal with the customer’s personal agent, and agree on fallback options if the two agents can’t reach a resolution.
“Without a clear fallback mechanism, the agent may keep trying different approaches, repeatedly contacting the business, and consuming more tokens.”
4. Design for Persistent Personal Agents
AI agents will make it easier for customers to compare offers, challenge decisions, and pursue resolutions, removing the friction that once discouraged them from taking action.
“There was a cost for me as a customer to sit in a queue or send an email. I couldn’t just spam it. Now I can. There is no cost for an agent to continue, to continue, to continue.”
Consider an insurance customer at renewal time. Previously, they might have compared two or three quotes before accepting. Now, their personal agent can compare 100 renewal quotes and challenge their insurer if the premium is no longer competitive. There is no extra effort involved for the human behind the agent.
To address this challenge, McCann recommends that organizations assess their ten most common customer interactions from the perspective of an extremely persistent AI agent acting on the customer’s behalf.
Such an exercise will identify unnecessary friction, ensure decisions are justifiable, and help redesign processes to deliver fair outcomes that don’t depend on customers’ limited time, knowledge, or persistence.
5. Maintain a Selective Approach to Supporting These Agents
Contact centers that support personal AI agents reduce the risk of those agents turning around and recommending a competitor instead.
However, supporting these agents should not mean opening up to every customer interaction. Contact centers should take a selective approach, prioritizing common intents where the risks are understood and manageable.
Why? Because mistakes could have significant consequences.
For instance, what if the company’s AI agent provides bad information that leads a personal agent to make the wrong decision? There is a significant liability risk here.
Given such risks, brands should only incrementally expand support for personal AI agents, starting with common, low-risk intents.
6. Reconsider the Contact Center Metric Mix
Traditional contact center metrics, such as First Contact Resolution (FCR) and Average Handling Time (AHT), become less relevant as customers increasingly delegate tasks to AI agents.
After all, when an AI agent acts on the customer’s behalf, the number and duration of interactions matter less than whether the desired outcome is achieved.
Given this, when monitoring personal agent interactions, the focus should shift from measuring interactions to measuring task completion.
That could involve measuring the rate at which both the business and the customer’s agent agree that a task has been successfully resolved.
7. Think About How Vulnerable Customers May Use Personal Agents
Personal AI agents offer significant opportunities to help vulnerable customers navigate organizations and access services that might otherwise be difficult to use.
However, customers may become overly reliant on their agents without fully understanding the decisions made on their behalf.
To address this risk, contact centers may consider dedicated routes for agents acting on behalf of known vulnerable customers.
They could also examine the types of queries personal agents ask to identify previously unknown vulnerable customers and adapt their future experience accordingly.
What Does the Future Hold for Personal Agents?
Personal agents are likely to be much more widely adopted if and when they become integrated into phone assistants.
Imagine consumers simply asking Siri or Gemini by voice to cancel a contract, query a bill, or chase an update on a case.
Already there are signs that these personal voice agents are coming.
In 2025, Google patented a smartphone capability that automatically answers phone calls on the user’s behalf.
If this materializes, contact centers will also need to reconsider their outbound strategies.
Yet, first things first, customer support leaders may wish to consider these three questions:
- How do you verify that an agent is genuinely authorized to do what it is trying to do?
- What are the rules of engagement?
- Who is ultimately responsible if an agent makes a mistake? (There are three parties involved: the LLM provider, the personal AI agent provider, and the company interacting with the agent.)
The answers may help service leaders put their best foot forward.