AI's Dropout Problem: Why the Best Agents Never Stop Learning | Decagon | RAISE Summit 2026

Decagon's Core Value Proposition

• Builds AI agents designed to interact with end customers via voice, chat, email, and text to solve problems.
• Differentiates itself by focusing on the "deployability layer," enabling hundreds of enterprise employees to collaboratively build, test, and manage agents at scale.

Enterprise Motivations and ROI

• Enterprises primarily seek to increase revenue and reduce churn by handling higher volumes of customer interactions rather than simply reducing headcount.
• Business model is outcome-based (proof of work), pricing against the cost of human agent intervention rather than a per-seat SaaS fee.
• Customers express significant concern regarding deployment confidence and potential brand damage from AI errors.

Operational Mechanisms and Safety

• Testing Suite: Simulates agent behavior against historical human conversation data (potentially millions of interactions) prior to launch to identify and fix regressions.
• Production Monitoring: Implements real-time detection of deviations from expected agent behavior to catch issues before they reach customers.
• "Duet Autopilot" Learning Loop: Automatically analyzes escalated human-AI interactions to identify patterns; pre-drafts fixes based on human agent solutions for admin review and deployment.

AI Strategy and Model Architecture

• Model Selection Strategy: Employs a sequential approach using frontier models for new, complex tasks requiring high intelligence, then transitions to smaller, tuned open-source models for specific, optimized tasks once the landscape is understood.
• Internal Development: Decagon Labs creates custom models for specific use cases while maintaining partnerships with major frontier labs.
• Cost Management: Treats AI spend as an ROI problem, evaluating cost per use case (e.g., PR reviews, sales briefs) rather than adopting AI universally.

Market Position and Competitive Landscape

• Target Market: Focuses exclusively on large, regulated enterprises (financial services, airlines, telcos) rather than SMBs.
• Moat Strategy: Builds comprehensive infrastructure surrounding the core agent, including permission structures, QA workflows, compliance monitoring, and insight extraction, which generalist labs do not provide.
• Incumbent Competition: Views the acquisition of Finn by Salesforce as a validation of the market but maintains that the complexity of regulated, large-scale enterprise needs leaves the market "thinner" than perceived, favoring specialized providers.

Forward-Looking Statements

• Adoption Trend: Predicts a shift in focus from model capability to the infrastructure required to make agents deployable and safe within enterprise environments.
• Immediate Focus: The primary development goal for the next 12 months is the automation of agent improvement, allowing systems to learn and evolve from millions of daily customer interactions without significant manual effort.