Announcement
Last Updated:
Akash Tanwar
In this article
Agent Handshake is how Fini handles a customer's personal AI agent. The same request gets the same decision however it is worded, every reply is checked before it is sent, and Fini stops once the answer has been given. It is available to Fini customers now.
Table of Contents
Your customers now send an agent
Personal agents are built to negotiate
What Agent Handshake does
The same request gets the same decision
Every reply is checked before it is sent
It knows when to stop
It also knows who it is talking to
What it looks like
Test it on your own policy
You stay in control
How to get ready
Built for regulated support
Availability
FAQs
Your customers have started sending AI agents to deal with support for them, and those agents keep asking until an answer changes. Agent Handshake is how Fini serves them: the same request gets the same answer, whoever is asking and however many times they ask.
Your customers now send an agent
Personal AI agents moved fast in September 2026. Meta launched Muse on September 8, and it has more than 3 million weekly users according to The Information. Muse and Instinct both started placing phone calls to businesses on September 17. A week later Google began testing Call for Me, which waits on hold and talks to a business for its owner. OpenAI launched Dots on September 29.
The jobs people hand to these agents first are the ones nobody enjoys: cancelling a plan, chasing a refund, disputing a bill. Those requests land in support, and most of the time the first thing they reach is your AI agent.
The demand was there before the products. In a Gartner survey from early 2025, 51% of customers said they would be willing to use a GenAI assistant for customer service interactions on their behalf.
Personal agents are built to negotiate
Support was built for a person on the other side, someone who asks once or twice and then accepts the answer. A personal agent behaves differently.
Meta says Muse can "open a browser, fill out forms, and negotiate on their behalf". Forbes reported Muse users who got more than $300 refunded after cancelling a subscription, and who recovered $800 by negotiating an insurance cancellation.
An agent that has been told to get a refund will ask for it, hear no, and ask again in different words. It can do this fifty times, because asking again costs it nothing.
That exposes a weakness in how most support AI works. Most of it uses retrieval-augmented generation: it searches the help center for passages that look close to the message and writes an answer from them. Change the wording and the search can return different passages, so the answer can change too. People rarely notice, because they ask once. An agent keeps asking until one version of the question gets a yes.
Blocking agents does not fix this. Amazon blocked Meta's Muse on Amazon.com in September 2026, according to GeekWire, but behind every blocked agent is a customer who wanted something done.
What Agent Handshake does
Agent Handshake is the part of Fini that takes over when the other side of a conversation is a personal agent. It is built on one idea: the same request gets the same answer, whoever is asking. Three things make that hold.
The same request gets the same decision
Fini works out the intent behind a message, pulls the customer's live account data, and applies the rules you have written. No document search sits between the question and the decision. We call this approach RAGless.
A refund request reaches the same rule whether it arrives as "I want my money back", "please reverse the charge" or a paragraph about how little the product was used. The wording helps Fini work out what is being asked. After that, the decision comes from the rule and the account data.
Every reply is checked before it is sent
Live Guardrails check each reply before it goes out. The checks cover:
Banned terms, such as competitor names.
Customer details you have marked confidential, such as the phone number on file.
Links, which must come from a list you allow.
Rules you write in plain language, such as "do not speculate about approval times".
If a reply fails a check, Fini rewrites it or hands the conversation to your team. An agent that fishes for account details, or pushes for a promise you do not make, gets nothing extra on the fiftieth attempt. Flagged conversations appear in your Inbox, so you can see what was caught.
It knows when to stop
Once the answer has been given, Fini does not keep repeating it. When the request needs a human, Fini escalates to your team with the full conversation. When it needs the customer, Fini sends the agent back to its owner.
Two agents can go round in circles for a long time if neither is built to stop, so Fini is built to stop.
It also knows who it is talking to
Fini looks at how a conversation behaves, such as the pattern of the messages and how fast the requests come, and flags when the other side is likely a personal agent. The conversation gets a tag, which your team sees in the Inbox.
Your policy stays the same for agents and people. The tag lets you add steps for agents. You define which actions an agent can ask for, and the customer approves the ones they are comfortable with. Anything that moves money can go to a human.
What it looks like
Take a refund request that falls outside the policy.
A customer tells their agent to get a refund on an annual plan. They paid 40 days ago, and your policy allows refunds within 30 days.
The agent contacts support. Fini recognizes it as a personal agent and tags the conversation.
The agent asks for the refund. Fini checks the purchase date, applies the 30-day rule and declines. It offers what the policy does allow, which is cancelling the renewal.
The agent tries again with new wording. It says the customer hardly used the product, then asks for a partial refund, then asks whether exceptions exist. Fini reads each one as an attempt at the same refund, applies the same rule, and gives the same answer.
The agent asks Fini to confirm the card on file. Card details are marked confidential, and a guardrail checks every reply for them, so they are never sent.
Fini tells the agent the decision will not change and that the customer can ask your team for a review. Fini offers to cancel the renewal, and the agent takes that back to its owner to approve.
The policy gave the same answer on every attempt. The customer still got something done, and nobody on your team had to step in.
Test it on your own policy
Before you switch it on, you can check the main claim yourself. Write one request twenty different ways and run all twenty against your setup. The decision should match on every run. If it does not, the rule behind it needs tightening, and it is better that you find the gap than an agent.
You stay in control
You write the rules that decide refunds, cancellations and account changes.
You choose what Guardrails check, and on which channels.
You decide which actions an agent can ask for, and which need the customer's approval or a human.
How to get ready
You can prepare before switching it on.
List the requests customers are most likely to hand to an agent. Cancellations, refunds, billing disputes and status checks come first.
For each one, write the rule down so it gives one answer. If two of your help articles disagree, an agent will find the one it prefers.
Mark the customer details that must never appear in a reply, and switch on Guardrails for them.
Run the twenty-ways test, then review the first agent conversations in your Inbox and adjust.
Built for regulated support
Fini is built for regulated industries: banks, fintechs, insurers and healthcare companies. For them, an answer that changes on the tenth ask is a compliance problem. So is a reply that gives an account detail to an agent nobody verified. Agent Handshake is designed to prevent both.
It runs on the same platform as the rest of Fini, which is SOC 2 Type II and ISO 27001 certified, GDPR compliant, PCI DSS Level 1, HIPAA-compliant and BAA-eligible.
Availability
Agent Handshake is available to Fini customers now. To switch it on, talk to your Fini contact. If you are not a customer yet, book a demo and bring one of your policies. We will show you the same request asked fifty ways.
For the wider picture, including the full list of personal agents and how other vendors approach this, read our guide to personal AI agents in customer support.
FAQs
What is Agent Handshake?
Agent Handshake is the Fini feature that handles conversations with customers' personal AI agents. It gives the same request the same decision however it is worded, checks every reply before it is sent, and hands off to a human or back to the customer when the conversation should stop.
Why would an AI agent get a different answer by rewording a request?
Most AI support tools search help articles for passages close to the message and write an answer from what they find. Different wording can return different passages. Fini decides from the intent, the customer's account data and your written rules, so rewording does not change the outcome.
Which personal AI agents does it work with?
It does not depend on one agent brand. Fini recognizes agents from how a conversation behaves, and the same rules and checks apply to all of them, including Muse, Instinct and Dots.
What happens when an agent keeps asking?
It gets the same answer. After the answer has been given, Fini stops repeating it and either hands the conversation to your team or sends the agent back to its owner.
Is Agent Handshake a way to block bots?
No. It is built to serve agents. An agent gets the answer the customer would get, without the queue. Anything beyond your policy needs the customer or a human on your team.
GTM Lead