Decagon launched Duet Autopilot, an AI agent that continuously diagnoses, tests and ships improvements to customer experience agents with minimal manual intervention. The system translates production signals into proposed updates, validates each change against regression tests and real customer personas, and surfaces versioned diffs for human approval before deployment.
To benchmark capabilities, Decagon also released DuetBench, an evaluation framework designed to measure AI agent self-improvement end-to-end. Autopilot passed 93% of diagnostic tasks, exceeding the average human score. The system feeds its own correction history back into its improvement loop, enabling performance gains to compound over time.
Decagon shows how self-improving AI agents can help enterprise customer experience teams move from periodic optimization cycles toward more continuous and automated improvement.
Image Credit: Decagon
What's Driving This Trend
- Self-improving AI Agents
- Autonomous systems that diagnose failures, test fixes, and refine their own workflows represent a shift toward software that compounds performance gains without traditional release cycles.
- Continuous CX Optimization
- Customer experience operations are moving from periodic tuning to always-on improvement loops where production signals become validated updates for service agents.
- Agent Evaluation Frameworks
- Benchmarking tools that measure end-to-end agent improvement create new value around trust, regression prevention, and measurable AI performance governance.
Who This Affects Most
- Customer Service
- AI-driven support environments can evolve into adaptive service layers that reduce manual quality assurance while improving consistency across customer interactions.
- Enterprise Software
- Platforms with built-in self-testing and versioned approval workflows point to enterprise applications that update themselves while preserving managerial oversight.
- Contact Centers
- Automated agent optimization introduces a new operating model for contact centers where staffing, training, and performance management are increasingly augmented by continuous AI refinement.
