Introducing Duet Apprentice: AI that learns like humans do

Today, we're introducing Duet Apprentice, a new way for Duet to learn your business like your best people do. Apprentice arrives alongside Voice 3, Personal Agent Gateway, and Agent Modules at Decagon Dialogues.

Every company has systems to turn a new hire into an expert: onboarding docs, team channels, and undocumented expertise held by the tenured teammate. A new hire reads it, sits in on internal discussions, and picks up what nobody wrote down. None of it was designed for an agent, though. Teaching an agent the same context has meant manually piecing together what the team knows and perpetually keeping it current.

Duet arrived earlier this year as an agent-building partner carrying what Decagon has learned across hundreds of enterprise deployments, and it now writes more than 70% of the Agent Operating Procedures (AOPs) running on the platform. But most of what it knows comes from what happens within the Decagon platform. 

With Apprentice, Duet widens that aperture in two directions:

• Onboards the way a human does, reading your onboarding material and learning from how your best reps work
• Works alongside your team in the messaging platforms they already use, so it keeps proactively learning after launch

Apprentice is in beta today and will be generally available soon.

Onboard your agent the way you onboard your team

Apprentice enables Duet to connect to the systems where your team already keeps its knowledge and read what's there. Onboarding docs and process wikis in Notion, Confluence, SharePoint, or Google Drive stay where they are, with nothing migrated or reformatted.

Documentation only covers part of what a new hire learns. The rest comes from watching someone who has done the job for years. Duet analyzes the escalated conversations where a tenured rep handles a situation no AOP anticipated, such as how to navigate an unusual edge case. That judgment is usually a team's most valuable asset but is rarely written down.

Everything Duet learns through Apprentice becomes context it can draw on. Ask Duet to draft a workflow, run Simulations, or analyze conversations, and it grounds its approach in your business context.

It reaches Duet Autopilot too, steering how the agent self-improves. Every run draws on what your team knows, not only what the agent has seen in production, so a proposed fix can account for a decision that no transcript would ever reveal.

That context belongs to you. It stays in your workspace and isn’t use to train any models.

Part of the team, wherever work happens

A teammate that only learns during onboarding starts falling behind in its first month. Policies get revised, a promotion launches, and an edge case becomes routine.

Apprentice's plugins for Slack and Microsoft Teams let Duet follow along where those decisions actually get made in real time, picking up what's changing before anyone writes it down. Ahead of a product launch, it can track which customers get access first and what the team settled on for the migration path, then be tagged in to draft the AOP updates before the launch date. 

Duet also proactively unblocks itself. When a thread is ambiguous, it asks for clarity instead of guessing, in the channel where the conversation is already happening for greater visibility. A teammate that asks when it isn't sure is easier to trust than one that assumes an answer.

"Duet lets our team build and iterate on agent logic without pulling resources away from the roadmap. That matters at Perplexity, where we ship product changes every day and our agent has to keep pace." – Jenn Palk-Cogley, Head of CX and User Operations at Perplexity

‍

"Before Decagon, launching a new feature meant 2 to 4 weeks of manual, spreadsheet-driven A/B testing, and model updates could cost us 6 weeks of tweaking individual prompts. With Duet, we confidently launch new features on a weekly basis. It has drastically tightened our cycle time, allowing us to focus on the client experience instead of getting bogged down in under-the-hood versioning." - Anne Young, Senior Manager, AI & Process Automation at Questrade

‍

You shouldn’t need to build the map

Nobody would accept a self-driving car that asked you to survey the roads first; the car is supposed to know the route. AI agents have been sold on a similar promise of autonomy, but have required enterprises to absorb the cost of building the map.

Your agent's ceiling is set by what it's allowed to learn from. Frontier models keep getting better at reasoning, and none of it teaches an agent how your team actually operates. That knowledge already exists inside your company, and Apprentice is how the agent reaches it.

Reach out today if you'd like to be one of the first customers to try Apprentice.