Introducing Duet Autopilot, the first verified self-improving AI agent for CX. It automates agent improvement by turning conversation signals into validated improvements ready for human review, helping agents get better with every conversation. 🧵

This week, @firstharmonic hosted its first sales-focused Builders Table (Sellers Table?) with top AEs from @DecagonAI, @Rippling, @fal, @getserval, @_hex_tech, @stripe, @coderabbitai, @figma, and other great companies. Some interesting takeaways from the room: 1/ Most sellers are using AI for meeting prep and follow-up, not as much for building decks or writing emails (!). 2/ The thing sellers most WISH AI could do for them is automate CRM input (hard relate). 3/ Most sellers think that there will be fewer SDRs in five years, and that they'll have different skills than they do today. 4/ Most sellers think Salesforce will eventually get displaced, but not for 10 years. 5/ The "AI tool" that sellers value most isn't an AI-native tool at all; it's @Gong_io. Personally, I learned that the most cracked sellers are the ones who show up at a dinner on the last day of the quarter, the week before 4th of July! Thanks to all who joined us. We want to do more of these dinners: for engineers, salespeople, marketers, you name it. DM or email me if you want to get on the list.

Voice agents are one of the hottest use cases in enterprise today, but also one of the hardest to go live. → How do you reduce latency to something that feels natural while also prioritizing intelligent responses? → How do you reliably make tool calls to your CRM without losing the customer on the line? → How do you build a system of fallback models in case Anthropic and OpenAI are running hot? Last week, I hosted a fireside chat with Basia Sudol (Head of Enterprise Solutions, @DecagonAI), @vr000m (CPTO, @trydaily), Steven Diaz (FDE Manager, Vapi), Tyler D'Silva (Founding FDE, @retellai), and @kamath_sutra (Founder, @smallest_AI) to deep dive into the world of voice agents and answer some of those questions. Some of my favorite parts and link to full episode in comments ↓

Building production-grade agents shouldn't be difficult or tedious. In fact, over the last few months, several customers have begun sharing major agent improvements, built E2E without deployed support. The big unlock has been Duet, Decagon's AI agent for building AI agents!

our engineers >>

World cup szn at Decagon ⚽️

Forward Deployed Engineering is the most critical ingredient for Enterprise AI adoption. It’s why the most important companies on the planet are investing so heavily into it. That’s why we put together the Forward Deployed track at this year’s AI Engineer World’s Fair on Tuesday, June 30. Learn about the current state of the industry and where its headed from the people and companies leading the charge. Everything it REALLY takes to deploy agents in production and transition our economy to the new world order Don’t miss it :) (more details and link in comments) ↓

out of all the config week side quests, the @harvey × @DecagonAI event last night probably gave me the most to think about. they had an incredible panel that dropped some sharp pro-tips. @harvey also has a ridiculously gorgeous office. a few things that stuck:

Still don't think people fully appreciate how big dflash can be for inference latency/throughput. Genuine game changer for latency-sensitive workloads.

design night with @harvey x @DecagonAI

We have some insane people at Decagon... Sebastian Chiu just ran 220 miles continuously from NYC to Boston. Just had to share. He recorded a video of it 👇

Modal Auto Endpoints provide state-of-the-art open source inference perf with a click. Learn how we developed our low latency inference playbook with @DecagonAI, delivering responses 60ms faster than the best proprietary provider. modal.com/blog/achieve-sota-…

SaaS products accumulate a ton of features over time (for good reason, B2B is complex) and for decades we just threw them all into menus because the software didn't know what you actually wanted, so it made you do the work. Companies have whole orgs dedicated to training customers on their product. Now that agent interfaces are here they tend to get written off as a marketing gimmick, but it seems to me it’s actually the first real fix to that problem. We just built one @DecagonAI with Duet Autopilot, and it has already made more changes to our customers' agents than humans have - no one seems to miss the menus. More in the comments!

Decagon vs Sierra is the AI CX debate to watch. Sierra is a comfortable choice. But more CIOs are choosing @DecagonAI because of its product and approach. Decagon puts the people closest to the customer in control. CX and product teams can define, test, and improve how their agents behave without overreliance on a vendor. The next generation of AI CX will be defined by the operators who learn from their customers and iterate the fastest.