The newly rebranded Fin (formerly Intercom) has launched Operator, an AI agent designed to manage, optimise, and continuously improve customer support operations. Now available in early access, Operator works across both Fin's AI agent and the Intercom helpdesk, taking on the operational work that most support teams either handle “reactively, or not at all”.
The problem it's solving
As Fin explains, running a modern customer support function means managing AI performance and human teams simultaneously. On the AI side, that means keeping help content current as products evolve, diagnosing why an AI agent mishandled a conversation, and identifying where automation can expand. On the human side, it means tracking rep performance, managing incident response, and surfacing what to prioritise next.
The company’s argument is that this work consistently outpaces what teams can realistically keep on top of, and that Operator can relieve them of this burden. The results, it says, can provide a new way to “understand, manage, and improve customer experience”.
What Operator does
Operator's capabilities span four broad areas. On the data side, teams can query their support operation in plain language, asking why a metric changed or what drove escalations last week. Operator analyses real conversations to surface patterns, produce charts, and identify root causes, with the option to schedule recurring reports that deliver automatically to a Fin workspace.
Knowledge management is a significant focus. Teams can brief Operator on a product update or policy change, and it will identify every affected help article, draft revisions in the team's tone of voice, and flag content gaps. Beth-Ann Sher, Fin's Senior AI Knowledge Manager, described working with Operator as having "five additional knowledge managers" on the team.
For AI configuration, Operator can debug conversations where Fin underperformed, propose fixes, and run simulation tests before anything goes live. It can also build ‘Procedures’ from a plain-language description, configure rules and data connectors, and flag where automation opportunities exist based on what human agents are still handling manually.
On the human operations side, Operator can identify affected customers during an incident and draft targeted responses, help team leads prepare for one-to-ones by pulling rep metrics and surfacing outliers, and give individual agents a prioritised queue when they return to their desk. It is, in effect, a working example of how human and AI collaboration can be designed into a single workflow rather than bolted on as an afterthought.
What sets it apart
The most original aspect of Operator is its positioning as an AI agent that manages AI support operations, rather than one that handles customer conversations directly. Most CX AI investment to date has concentrated on customer-facing agents, human assist tools, and analytics dashboards. Operator, on the other hand, is essentially an agent to oversee agents, on its path to becoming “increasingly agentic”.
It also appears to be meeting a genuine pain point. As many have observed, deploying a support AI is the easy part, keeping it accurate and effective at month six is where most operations begin to come apart at the seams. Instead of having to rely on fully autonomous edits or manual admin work, Operator enables operational improvements for human review before anything deploys. This is also key in addressing trust and governance concerns, which are very much front of mind right now with the EU AI Act soon to come into effect.
Built for purpose
In a related blog post, Fin positions Operator explicitly against the "well-prompted LLM”, pointing instead to fundamental differences in design. Rather than giving a foundation model access to APIs and relying on it to reason its way through support data, Operator has more than 50 purpose-built tools and 10 skills, each encoding specific domain expertise. That kind of precision takes months of engineering to develop, which represents tangible value for organisations weighing up building versus buying.
While none have packaged their technologies in quite the same way, competitors such as Zendesk, Salesforce, and Freshworks are moving in a similar direction. As with any agent operating at this level of autonomy another critical question is whether it performs reliably at scale beyond controlled demos. With over 200 teams already deploying Operator, the results will begin showing soon.