Decagon review: AI agents for customer support across chat, voice and email, built for enterprises

What Decagon does

Decagon is a platform for building, optimizing and scaling AI agents that handle customer conversations. Its homepage describes it as an AI concierge, aimed at enterprise customer experience teams across industries such as retail, travel, financial services and healthcare.

Agent logic is written as Agent Operating Procedures (AOPs), which describe workflows in natural language. A companion called Duet reviews customer conversations, spots gaps in workflows, and generates, tests and refines AOPs. The vendor says this lets non-technical teams change agent behavior, while technical teams keep visibility over guardrails, integrations and versioning.

The same intelligence layer covers chat, voice and email. The voice agent is built on the vendor's own speech model, supports 70+ languages, handles interruptions, can place outbound calls, and can pass a call to a human agent with a summary. Decagon Assist is a related tool that autogenerates summaries, replies and in-tool actions for human care teams.

For quality control, the platform offers simulated-conversation testing, versioning, A/B experiments, decision tracing, and Watchtower, which monitors conversations around the clock against custom criteria. It connects to ticketing platforms, CRMs, knowledge bases and CCaaS providers.

Pricing is not published on the official site. The pages direct visitors to get a demo or contact sales, and no free plan or trial is mentioned.

• Natural-language AOPs
• Duet AOP generation and refinement
• Chat, voice and email agents
• 70+ language voice support
• Outbound calling
• Simulated-conversation testing
• A/B experiments and versioning
• Watchtower monitoring and QA

Our take

Decagon looks aimed at larger support organizations that want AI agents across chat, voice and email, with workflows edited in plain language rather than code. The testing, versioning and monitoring features suggest a team that expects to keep tuning its agents. The trade-off is transparency on cost: there are no published prices, no free plan and no self-serve trial on the pages we read, so budgeting means a sales conversation. Small teams wanting to try a tool quickly, or buyers who need to compare costs up front, may find this a poor fit. Results figures on the site are customer claims, not guarantees.