After two years of deploying AI on the front lines of customer service, a problem emerged: the AI was working faster than humans could monitor and tweak it.
Last Thursday, the company formerly known as Intercom—now rebranded as Fin—held a launch event in San Francisco, unveiling a new product called Fin Operator. Its job isn't to answer customer calls or write tickets. Its sole purpose is to oversee another AI, named Fin, which handles customer service.
This marks the first time in the company's decade-long history of customer service software that it has turned 'AI managing AI' into a sellable product.
Frontline and Second-Line Duties Assigned to Different AIs
Fin's sales model over the past two years has been straightforward: let AI handle the front line, with tens of thousands of customer conversations per month processed directly by Fin, and companies pay based on 'resolution rate.' But every Fin customer had to maintain a team of 'AI ops' in the background—checking prompts when customers complained, updating knowledge bases when products changed, and pulling reports to find bugs when weekly numbers didn't add up.
The newly launched Operator packages these tasks into a single product.
It has two external roles:
- Data Analyst: Ask directly, 'How did the customer service team perform last week?' and it pulls up charts, trend reports, and resolution rates broken down by channel—things that previously required clicking through three layers of dashboards, now accessible with one sentence.
- Knowledge Base Manager: Feed it a three-page product update PDF, and it scans the entire document library to identify which articles are outdated, which need revision, and which are missing—then lists suggested changes in diff format for you to confirm.
Fin's customer service AI has been running for over two years across thousands of companies, each with a backend team doing this 'second-line AI ops.' Operator aims to eliminate that human overhead.
A Less Glamorous but Lucrative Track
In the software world, the idea of 'AI supervising AI' has been floated many times over the past two years, but few products have reached the market. Microsoft's AutoGen, Anthropic's multi-agent systems, and various agent-swarm papers have all discussed it—mostly remaining at the demo stage.
Fin's advantage lies in its existing customer base. Before rebranding to Fin, Intercom was already a leader in the space, with data from running customer service AI, configuration pipelines, and knowledge base structures already in place. Having one AI manage another is much easier when the materials are ready.
Fin's financial logic is also straightforward: currently, companies pay per resolved customer query. With Operator, they add a subscription layer, effectively selling back the 'configuration and maintenance manpower' for AI.
Who Will Be Replaced First
Operator is now available in early access for Pro-tier users, with a general release expected this summer.
For enterprises, the first roles to be compressed aren't frontline customer service (which already saw cuts last year) but the backend AI ops teams. In short, the group responsible for 'writing rules, debugging, and reviewing reports for the AI customer service' will see its workload further reduced.
As for Fin's competitors—Salesforce's Agentforce has reached $800 million in ARR, and Sierra is valued at $15 billion—they are likely working on similar solutions, though none have officially launched yet.
The next question for this track isn't 'Can AI do customer service?'—that was settled last year. It's 'When AI makes a mistake, who fixes it?' Operator offers one answer: send in another AI.
Sources: Intercom, now called Fin, CocoLoop, launches an AI agent whose only job is managing another AI agent (VentureBeat)