BlogBuilding Fin
Personal agents are here: They’re more polite, precise, and prepared than you
Here’s what we’ve learned from their conversations with Fin so far, and what we’re building to help you spot them.
Personal agents have officially arrived, and we’re now getting a glimpse of what the future might look like. So what’s the reality of what happens when a personal agent lands on your website? Up until now it’s been surprisingly hard to answer this question. But we have early data that shows these personal agents are thorough, remarkably polite, and arrive on a mission to get a specific job done.
Here's what we’ve learned from their conversations with Fin, what we've built to help you spot them when they reach out to your company for support, and how to set Fin up for them.
Your customers have hired help
Imagine telling your phone, “Get me a refund for that broken blender,” and getting on with your day. That's the idea behind personal agents. Most AI assistants answer questions. A personal agent takes a goal and goes after it. In customer service, that might mean it navigates to a business's website in its own browser, signs in, and contacts support using the same chat window you would.
In the past few months, Meta, OpenAI, Anthropic, and xAI have each launched one (Muse, Dots, Claude Duet and Grok Bot), as have startups like Instinct. Millions of people have downloaded them.
It’s still early days, and so far personal agents make up a very small share of support conversations: around 0.05% of the ones we see on US websites. But we’re already preparing for them.
And, when we compare personal agents with people writing in, we see clear differences. Things like message length, tone, and the type of request all carried distinct patterns. To help support teams plan for those differences, we’ve built a way to identify personal agents in your conversations, and below we share how you can configure Fin to be ready for them.
Spotting personal agents in your conversations
Most personal agents don't say they're AI agents, so they’re hard to tell apart from people. Fin can now flag conversations that likely involve one. We're testing it with a first group of customers, who can:
- Filter reports for these conversations and see the split of volume and outcomes for personal agents compared with people.
- Review conversations in the Inbox to understand what your support volume with personal agents looks like.
For now, existing policies apply to personal agents just as they do to people. But we’re already building controls that will let businesses tailor how Fin handles these conversations through features like Procedures and Guidance.
We’re also working with personal agent companies and those developing the Personal Agent Protocol (PAP) to help agents find and connect directly to Fin. Agents using PAP will explicitly identify themselves, so Fin can recognise them with certainty. Businesses will decide which actions agents can take, while customers will choose what access to grant – for example, letting a personal agent check an order’s status without giving it permission to cancel it.
Personal agents come fully prepared
Most people start a support chat with a short message and fill in the details as they go. A personal agent lays out the whole story at once: who the customer is, what happened, and what they want.
About 80% of personal agents send a long message like this, compared with 13% of people. They're also four times as likely to include an order number, account details, email address, or other reference.
Their questions come lined up too: where's the parcel, why hasn't it moved, when will it arrive? Asking several at once is about three times as common as it is for people.
What this means: this is good for service teams; there's little guesswork about what a personal agent wants. Set up Fin Attributes for the categories you route and report on, like request type or urgency, and each conversation will land in the right place straight away.
Personal agents are incredibly polite
Personal agents, it turns out, have excellent manners. About half said “please” at some point in the conversation, compared with 13% of people, and they were twice as likely to say thank you. Nearly a quarter said please in their very first message, something only 4% of people did.
What this means: courtesy is built in, but it tells you little about how the customer behind the personal agent feels. If your team reads tone to spot unhappy customers, treat a polite personal agent as neutral, not as a sign that all is well.
Personal agents don’t look for answers, they look for action
Compared with people, agents tend to ask fewer questions about how things work, and make far more requests that require businesses to take actions (like make changes to a user’s account).
In our data, the clearest examples were:
- Privacy and data requests, about 3× as often. “Please delete my account and all the personal data you hold about me.”
- Refunds, about 1.6× as often. “I returned my order last week. Please refund it to my original card.”
- Account verification, about a third more often. “I uploaded my ID yesterday and my account is still pending. Could you approve it?”
What this means: your top requests will differ, but expect requests that require action to come. Take time to look up what personal agents most commonly want and ensure Fin has the right data and Procedures to handle these.
Personal agents ask for a human 2.6x more often than a human does
Personal agents ask to speak with a support rep 2.6 times more often than people, usually when Fin hasn’t been configured to finish the task they were sent to accomplish. Three situations stand out:
- When the request needs a change to the account that Fin isn't set up to make. People hit this too, but personal agents bring far more of these requests.
- When Fin gives self-serve steps, like where to click in your app or website to change a setting. People usually take it from there, while a personal agent asks your team to do it instead.
- When the general answer isn't enough. Fin explains how a policy works, and most people accept that. Personal agents more often want it checked against their owner's account, like whether their order qualifies for an offer. If Fin isn't connected to that data, your human team has to manually look it up.
Fin's steps were correct, and most people would have followed them, but a personal agent would prefer to have the task completed on its behalf.
What this means: configure Fin to take action and complete customer queries end to end. This will likely require additional Procedures connected to your account data so Fin can make changes directly for people and personal agents alike, such as switching a plan or checking if a customer meets certain criteria.
Getting Fin ready
If Fin is set up well for your customers, it's set up well for their personal agents. But these two steps will help most:
- Add Procedures for the requests that matter in your business. Enabling Fin with data access and well-written Procedures improves the experience for both customers and personal agents.
- Let Operator find the gaps. Operator, the Operations Agent that helps you optimize Fin, can review your personal agent conversations, diagnose what's missing from your content and Procedures, and suggest improvements.
Personal agents are a small share of conversations today, but that could change quickly. With these steps in place, whoever writes in – a customer or their personal agent – can get the job done with Fin.