In a recent LinkedIn post, Teresa Torres highlights the limitations of current AI support tools in handling complex customer service scenarios, particularly within the fintech sector. Torres uses the analogy of a stolen credit card to illustrate the multifaceted nature of customer issues that go far beyond simple query responses.
“What happens when a customer reports a stolen credit card? The frontline answer is simple—freeze it. But underneath lies a cascade of follow-ups: dispute filings, fraud investigations, merchant communications, and proactive outreach to gather more details. Most AI support tools handle only the tip of the iceberg.”
Torres’s post introduces a discussion with Jack Taylor, Product Engineer, and Ibrahim Faruqi, AI Engineer from Gradient Labs. They are developing AI agents designed to automate the entire customer support process, moving beyond basic chatbots to handle intricate workflows.
The Iceberg Metaphor in Customer Support Automation
As Teresa Torres points out, the common perception of customer support automation often focuses on the easily visible, front-line interactions. However, the true complexity lies beneath the surface, involving a series of interconnected tasks and follow-ups. Gradient Labs, according to Torres’s reporting, is addressing this by building AI agents capable of managing these deeper, more involved processes.
The core of Gradient Labs’ approach, as detailed in the post, involves architecting a platform with three distinct but coordinating agents: inbound, back office, and outbound. This structure allows for a more comprehensive handling of customer issues.
“They share how they’ve architected a platform with three coordinating agents—inbound, back office, and outbound—all built on a shared foundation of natural language procedures, modular skills, and configurable guardrails.”
Enabling Non-Technical Experts and Orchestrating Complex Conversations
A significant aspect highlighted by Torres is how Gradient Labs empowers subject matter experts. Instead of relying solely on engineers, their system allows non-technical individuals to define agent behavior using natural language.
Teresa Torres emphasizes that this is achieved through “natural language procedures,” which means “no coding required” for defining how agents should behave. This democratizes the AI training process, making it more efficient and accurate by leveraging the deep knowledge of those closest to the customer issues.
Managing Long-Running and Asynchronous Interactions
Furthermore, Torres reports on the technical architecture that enables these agents to handle complex, multi-step customer journeys. Gradient Labs has developed a sophisticated state machine orchestrator.
“Architected a state machine orchestrator that manages turns, triggers, and skill selection across long-running conversations.”
This orchestration is crucial for fintech support, where conversations can span days and involve various internal and external communications. The system’s ability to manage these “turns” and select the appropriate “skills” ensures continuity and efficiency.
Ensuring Safety and Continuous Improvement
Torres also draws attention to the critical role of safety and compliance in AI-driven support. The Gradient Labs team has implemented “guardrails” designed to prevent errors and ensure adherence to regulations.
According to the post, these guardrails are built as binary classifiers, with a focus on “high recall on critical regulatory checks.” This meticulous approach to safety is complemented by an auto-evaluation system. As Torres explains, this system “samples conversations for human review to catch edge cases and build labeled datasets,” fostering continuous learning and improvement for the AI agents.
In essence, Teresa Torres’s coverage of the Gradient Labs discussion provides a compelling look at how advanced AI agents are moving beyond simple task automation to manage the full, often messy, reality of customer support in sensitive industries like fintech.
📝 About This Content
This article is based on insights shared by Teresa Torres on LinkedIn.
📅 Originally posted on December 18, 2025 | View original post on LinkedIn →