Decagon builds conversational AI agents for enterprise customer service, with autonomous agents that can answer product questions, process refunds, and cancel subscriptions for brands like Avis Budget Group, Chime and 1-800-FLOWERS.COM . This launch expands Duet Autopilot, the company's self-improving CX agent, which now handles what Decagon describes as 70% of the ongoing agent building work that support and operations teams used to do by hand. The update is aimed at enterprise CX, ops, and product leaders who need to keep an autonomous agent moving without giving up control of tone, policy, or outcomes.
The centerpiece is a governed vault where teams store the context files that steer Duet before it acts, covering guidelines, tone, and naming conventions with version history and an audit trail. Connectors pull guidance directly from where SOPs already live, including Notion, so Duet works from the current version rather than a stale copy. On top of that, teams can now decide which outcome an improvement should optimize for, whether that is deflection, CSAT, handle time, or a custom metric, and then measure the actual impact of each change on agent performance.
The launch matters now because it lands alongside Decagon's broader Duet Autopilot rollout, positioned as the first agent to deliver automatic and verifiable self-improvement for CX agents , with every change surfaced as a versioned update requiring human approval before going live . Decagon was founded in 2023 by Jesse Zhang (CEO) and Ashwin Sreenivas (CTO) , and this release pushes the product further toward a model where operators guide and review rather than manually rewrite prompts and flows.
Comments (11)
Sign in to join the discussion.
N
Nikhil RaverkarSep 10, 2026
70% of agent building work is doing what exactly, writing the intents nobody wanted to write? I'll believe it when my ops lead stops Slacking me at 11pm about a broken flow.
P
Priya VenkateshSep 10, 2026
ok wait, the part where teams feed in their own goals and then measure impact is the actual unlock here. Most 'autopilot' pitches skip the measurement half and hope you don't notice.
J
June A.Sep 10, 2026
Retention on autopilot features is the whole ballgame. Are teams still using this in month 6 or does it quietly become another dashboard nobody opens?
T
Tomasz KwiatkowskiSep 10, 2026
The launch video does that thing where the UI zooms are half a beat too slow. Trim 200ms off each cut and the whole thing suddenly feels like a product, not a webinar.
R
Rasmus LindqvistSep 10, 2026
Tweet copy buries the actual news under a percentage. Lead with 'you can now steer it,' not with a stat that sounds like a McKinsey slide.
M
Marcus O.Sep 10, 2026
'AI concierge for every customer' is one letter away from being a Marriott slogan. I want the version that judges my refund request and offers me a mint.
C
Chidi OkaforSep 10, 2026
Curious how deep the docs go on the guidance layer. If I can't see example configs in the first five minutes I'm bouncing to whatever ships a public sandbox.
H
Hana BeristainSep 10, 2026
hot take: 70% is the number founders pick when they mean 'a lot but please don't audit us.' still, the framing beats yet another 'copilot for support' deck.
D
Devon L.Sep 10, 2026
We shipped something adjacent inside a big co in 2019 and the hard part was never the autopilot, it was convincing ops leads to trust the diff view. Curious how they handle that trust gap.
Y
Yuki TanabeSep 10, 2026
Every support automation company says they take on more work each quarter, yet the total headcount in CX keeps growing. Either the pie is bigger or someone's math is generous.
A
Anaya SridharSep 10, 2026
naive question: if autopilot builds 70% and I guide the other 30%, who owns it when the bot confidently refunds a whale? asking for a friend who is me on Monday.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Decagon
The AI concierge for every customer
The AI concierge for every customer
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Decagon builds conversational AI agents for enterprise customer service, with autonomous agents that can answer product questions, process refunds, and cancel subscriptions for brands like Avis Budget Group, Chime and 1-800-FLOWERS.COM . This launch expands Duet Autopilot, the company's self-improving CX agent, which now handles what Decagon describes as 70% of the ongoing agent building work that support and operations teams used to do by hand. The update is aimed at enterprise CX, ops, and product leaders who need to keep an autonomous agent moving without giving up control of tone, policy, or outcomes. The centerpiece is a governed vault where teams store the context files that steer Duet before it acts, covering guidelines, tone, and naming conventions with version history and an audit trail. Connectors pull guidance directly from where SOPs already live, including Notion, so Duet works from the current version rather than a stale copy. On top of that, teams can now decide which outcome an improvement should optimize for, whether that is deflection, CSAT, handle time, or a custom metric, and then measure the actual impact of each change on agent performance. The launch matters now because it lands alongside Decagon's broader Duet Autopilot rollout, positioned as the first agent to deliver automatic and verifiable self-improvement for CX agents , with every change surfaced as a versioned update requiring human approval before going live . Decagon was founded in 2023 by Jesse Zhang (CEO) and Ashwin Sreenivas (CTO) , and this release pushes the product further toward a model where operators guide and review rather than manually rewrite prompts and flows.
Comments (11)
70% of agent building work is doing what exactly, writing the intents nobody wanted to write? I'll believe it when my ops lead stops Slacking me at 11pm about a broken flow.
ok wait, the part where teams feed in their own goals and then measure impact is the actual unlock here. Most 'autopilot' pitches skip the measurement half and hope you don't notice.
Retention on autopilot features is the whole ballgame. Are teams still using this in month 6 or does it quietly become another dashboard nobody opens?
The launch video does that thing where the UI zooms are half a beat too slow. Trim 200ms off each cut and the whole thing suddenly feels like a product, not a webinar.
Tweet copy buries the actual news under a percentage. Lead with 'you can now steer it,' not with a stat that sounds like a McKinsey slide.
'AI concierge for every customer' is one letter away from being a Marriott slogan. I want the version that judges my refund request and offers me a mint.
Curious how deep the docs go on the guidance layer. If I can't see example configs in the first five minutes I'm bouncing to whatever ships a public sandbox.
hot take: 70% is the number founders pick when they mean 'a lot but please don't audit us.' still, the framing beats yet another 'copilot for support' deck.
We shipped something adjacent inside a big co in 2019 and the hard part was never the autopilot, it was convincing ops leads to trust the diff view. Curious how they handle that trust gap.
Every support automation company says they take on more work each quarter, yet the total headcount in CX keeps growing. Either the pie is bigger or someone's math is generous.
naive question: if autopilot builds 70% and I guide the other 30%, who owns it when the bot confidently refunds a whale? asking for a friend who is me on Monday.