The support queue changed before the software did
Something changed in the support queue while nobody was redesigning it. Customers stopped calling themselves. Their personal AI agents — tools like Muse and Dots — started calling on their behalf, asking for refunds, changing bookings, resolving disputes. According to data published by Fini, queries originating from personal AI agents have grown 20x since those consumer tools emerged, and the infrastructure on the other end was never built for them.
Support software was designed for a human conversation: a person who can rephrase, clarify, and tolerate a little friction. An agent rewords the same request and can receive a different answer, or gets blocked by a bot check and simply gives up. The customer behind the agent never sees the failure — they just hear that it could not be done. Deepak Singla, Fini's CEO and co-founder, put it bluntly in the company's announcement: most businesses are still thinking about customer support in terms of human interactions, and that framing is already behind. The agents are here, the volume is real, and the systems built for humans are not ready for them.
Four million interactions a month at Uber
Singla learned what breaks at scale before he tried to fix it. He graduated from IIT Delhi in mechanical engineering with a minor in business management, started at PwC, and then spent several years at Uber working on customer experience and growth, where he was responsible for resolving over four million customer interactions a month. That vantage point was, in his telling, transformative: he saw both the power of AI to transform support and the size of the gap in what other enterprises could actually buy.
At Uber, sophisticated machine learning models were already doing real work. Outside it, the options were basic chatbots that frustrated customers or expensive, low-quality solutions that did not actually solve problems. The gap was not a lack of models. It was a lack of systems trustworthy enough to let an AI answer for a company in the first place — and that observation became the company. Singla met his co-founder, Hakim Khalafi, at Uber; Khalafi brought a software engineering and analytics background of more than ten years and a master's in engineering mathematics from École Polytechnique. Together they set out to bring enterprise-grade AI support to companies that could not build it themselves.
Accuracy first, because a wrong answer is a liability
Fini's answer was to be accuracy-first from day one, aimed squarely at regulated industries — fintech, banking, healthcare — where a hallucination is not an embarrassment but a liability. When an AI gives a wrong answer about a refund policy, a benefits claim or a medication question, the company argues, it creates legal exposure, regulatory action and press damage from a single incorrect response. That is why, in Fini's telling, many financial firms have reversed AI support initiatives entirely rather than risk it.
The pitch that convinced investors was built on that framing. In January 2026 Fini announced a $3.6 million seed round led by Matrix Partners, with Y Combinator and angels from Uber, Intercom, SoftBank, McKinsey and Twitter participating. The company says its platform resolves 90% of tickets at 99% accuracy across voice, chat and email, goes fully live in 14 days and fully autonomous in 30 days, and ships with SOC 2 Type II, ISO 27001 and HIPAA-aligned compliance built in. Whether those numbers hold for every customer is a claim only deployments can settle — but the positioning is deliberate: resolution accuracy and compliance first, deflection volume second.
The handshake: same request, same answer
Agent Handshake, introduced in October 2026, extends that same bet to the agent-to-agent world. It is Fini's way for a personal AI agent to get help from support: the same request gets the same answer, however it is worded. The customer decides which actions the agent can ask for and approves the ones they are comfortable with, and anything that moves money can go to a human on the team.
The design is deliberately boring in the best sense. There is no new app for the customer to learn and no separate queue for agents to join. The personal agent talks to the support agent Fini already runs, and the conversation stays consistent because the underlying knowledge and permissions are the same. That continuity is the whole point: a refund or a booking change raised by an agent should complete instead of bouncing between systems that each assume a human is on the other end.
Why this gap, and why now
Agent Handshake matters beyond Fini because it names a shift every support team is about to feel. For years, support was measured on how well it served the person who showed up. Now a growing share of visitors are representatives — software acting with a customer's authority, at a customer's request, and with a customer's patience running out. A business that blocks those representatives is not blocking bots; it is turning away the customer behind them.
Singla's arc explains why Fini got there first among support vendors rather than last. He did not start from the technology and look for a market; he started from four million monthly interactions at Uber, saw that existing tools could not be trusted with the answers, and built a company around trustworthiness. Agent Handshake applies the same logic one step further: if the future customer arrives as an agent, then trust has to extend to the agent too — the same answer, the same permissions, and a human where the stakes demand one.
The playbook
- Treat a wrong answer as a liability, not a bug — Fini's founding bet is that in regulated support, a wrong answer is not a bad experience but a legal liability. Accuracy is the product, not a feature to add later.
- Build the system that finds its own gaps — Singla's Uber years taught him that support at scale breaks on knowledge gaps, not on volume. Fini detects its own gaps and learns from real resolutions instead of waiting for retuning.
- Serve the new visitor through the old queue — Agent Handshake does not ask businesses to serve a new channel. It asks their existing support agent to answer a personal agent the same way, every time — continuity over novelty.
- Put approval where the money moves — The customer approves which actions an agent can ask for, and anything that moves money can go to a human. Permission is designed in, not bolted on after the first incident.
- Ship live in two weeks, autonomous in thirty days — Fini went live in 14 days and fully autonomous in 30, per the company. Speed to value is part of the pitch to enterprises that have already watched AI support projects stall.