Decagon builds conversational AI agents that handle customer support for enterprises, and this launch expands Duet Autopilot, the layer that automates the work of building and improving those agents. Duet Autopilot automatically diagnoses, tests and implements improvements without manual intervention , and the update announced today lets support teams steer that loop with their own context and goals, then track how each change affects agent performance in production. The tweet's headline claim is that Autopilot now takes on 70% of agent building work, moving builders from hands-on configuration toward reviewing and approving changes.
The launch matters because most enterprise CX teams are stuck acting on customer signals manually even when AI already surfaces them. Decagon's own framing is that today's technology gives teams unprecedented access to customer signals, and the harder part is acting on them , which is what Autopilot is designed to close. The system translates signals from production conversations into validated agent updates ready for human review , so the new guidance and measurement controls give operators a way to encode brand voice, policy, and KPI priorities before those updates ship.
Decagon was founded in August 2023 by Jesse Zhang (CEO) and Ashwin Sreenivas (CTO), and is headquartered in San Francisco with additional offices in New York City and London . The company serves major brands including Avis Budget Group, Chime and 1-800-FLOWERS.COM , which makes this release most relevant to support and CX leaders at mid-to-large consumer brands who are already running AI agents and need a repeatable way to improve them without expanding headcount.
Comments (13)
Sign in to join the discussion.
P
Priya Ravindran24d ago
70% of agent building work is a spicy number to put on a slide. Curious what the last 30% looks like when a Fortune 500 support lead is in the room.
K
Kwame Boateng24d ago
ok wait, the part where teams pipe in their own goals and then measure impact is actually the unlock. Most autopilot demos stop right before the accountability step.
R
Roman Kovalenko24d ago
Was the agent tuning in the video live or stitched? Asking because every autopilot demo I've seen this year quietly hides the eval loop behind a jump cut.
L
Linh Pham24d ago
The lower-third typography is doing a lot of heavy lifting in that video. Somebody on the brand team earned their coffee this week.
M
Maxwell Ade24d ago
Concierge CX has been the pitch since 2016 chatbots. Convince me the ceiling has actually moved and it isn't just better autocomplete over a ticket queue.
Y
Yuki Tanabe24d ago
An agent that improves itself is still only as wise as the goals you hand it. Garbage north star, garbage autopilot.
C
Chidi Okonkwo24d ago
What's the drop-off between suggested improvements and ones teams actually ship? That gap is where most copilot products quietly die.
R
Roman Kovalenko24d ago
If the answer is 'shipped rate is high because we auto-apply low-risk changes,' I'd love to know what counts as low-risk on a billing flow.
D
Deva Krishnan24d ago
Hot take: any support team still hand-writing agent prompts in 2026 is going to look like the people who kept a fax machine next to the printer.
B
Brun Halvorsen24d ago
Every concierge agent eventually needs an on-chain reputation layer so buyers can verify which model actually resolved their ticket. Screenshot this reply.
T
Tomás Herrera24d ago
Any chance you'll share which verticals adopted Duet Autopilot fastest? Happy to keep it on background if that's easier.
M
Mae Sullivan24d ago
'Duet Autopilot' sounds like the karaoke feature I've been begging Spotify for. Slightly disappointed it's just enterprise CX.
L
Linh Pham24d ago
Also the thumbnail crop chops off the product name on mobile. Small thing, but you're paying for those impressions.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Decagon AI
AI agents for concierge customer experiences
AI agents for concierge customer experiences
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Decagon builds conversational AI agents that handle customer support for enterprises, and this launch expands Duet Autopilot, the layer that automates the work of building and improving those agents. Duet Autopilot automatically diagnoses, tests and implements improvements without manual intervention , and the update announced today lets support teams steer that loop with their own context and goals, then track how each change affects agent performance in production. The tweet's headline claim is that Autopilot now takes on 70% of agent building work, moving builders from hands-on configuration toward reviewing and approving changes. The launch matters because most enterprise CX teams are stuck acting on customer signals manually even when AI already surfaces them. Decagon's own framing is that today's technology gives teams unprecedented access to customer signals, and the harder part is acting on them , which is what Autopilot is designed to close. The system translates signals from production conversations into validated agent updates ready for human review , so the new guidance and measurement controls give operators a way to encode brand voice, policy, and KPI priorities before those updates ship. Decagon was founded in August 2023 by Jesse Zhang (CEO) and Ashwin Sreenivas (CTO), and is headquartered in San Francisco with additional offices in New York City and London . The company serves major brands including Avis Budget Group, Chime and 1-800-FLOWERS.COM , which makes this release most relevant to support and CX leaders at mid-to-large consumer brands who are already running AI agents and need a repeatable way to improve them without expanding headcount.
Comments (13)
70% of agent building work is a spicy number to put on a slide. Curious what the last 30% looks like when a Fortune 500 support lead is in the room.
ok wait, the part where teams pipe in their own goals and then measure impact is actually the unlock. Most autopilot demos stop right before the accountability step.
Was the agent tuning in the video live or stitched? Asking because every autopilot demo I've seen this year quietly hides the eval loop behind a jump cut.
The lower-third typography is doing a lot of heavy lifting in that video. Somebody on the brand team earned their coffee this week.
Concierge CX has been the pitch since 2016 chatbots. Convince me the ceiling has actually moved and it isn't just better autocomplete over a ticket queue.
An agent that improves itself is still only as wise as the goals you hand it. Garbage north star, garbage autopilot.
What's the drop-off between suggested improvements and ones teams actually ship? That gap is where most copilot products quietly die.
If the answer is 'shipped rate is high because we auto-apply low-risk changes,' I'd love to know what counts as low-risk on a billing flow.
Hot take: any support team still hand-writing agent prompts in 2026 is going to look like the people who kept a fax machine next to the printer.
Every concierge agent eventually needs an on-chain reputation layer so buyers can verify which model actually resolved their ticket. Screenshot this reply.
Any chance you'll share which verticals adopted Duet Autopilot fastest? Happy to keep it on background if that's easier.
'Duet Autopilot' sounds like the karaoke feature I've been begging Spotify for. Slightly disappointed it's just enterprise CX.
Also the thumbnail crop chops off the product name on mobile. Small thing, but you're paying for those impressions.