Hypothesis: Autopilot is better than ever. Results: Tested and validated.
Posted on November 5, 2026
Article
In June, we introduced Duet Autopilot as the industry’s first self-improving agent, and it’s already moving the needle for hundreds of enterprise teams. It lifted resolution rates by 25pp at a Fortune 50 company, delivered a 4.2% resolution gain from a single edit, and created durable impact with 85% of its updates still live in production after 30 days.
Autopilot has become a core part of how teams build agents. It empowers them to iterate on agent logic at an unprecedented pace. Today, we’re deepening Autopilot’s impact with two new capabilities that make it more effective:
- Run experiments on Autopilot updates to close the self-improvement loop
- Iterate faster with strategies developed by Decagon Labs hillclimbing strategies
Run experiments on Autopilot updates
Each Autopilot suggestion is only as good as its measured impact. That’s why Autopilot changes can now be validated with online evals through Experiments. Agent builders select the suggestions they want to test and push them to a limited share of production traffic. Once live, every change is tracked against the goal’s stated metric.
With experimentation, the self-improvement loop is complete. Every suggestion is first validated in offline simulation, then approved by human reviewers before being tested on live traffic. Once an experiment ships, fresh measurements feed the next suggestion, so Autopilot gets faster at hitting each goal over time.
Iterate faster with strategies developed by Decagon Labs hillclimbing strategies
Autopilot is built on Agent Operating Procedures (AOPs), the same foundation as the customer-facing agents it improves. This design choice has two important implications: Autopilot is subject to its own improvement loop, and the experts who build it can continuously improve its inputs.
Since launch, Decagon Labs has refined its approach to agent hillclimbing to find the most efficient and effective ways to accelerate improvements to agent logic. Today, those learnings are codified as strategies Autopilot can deploy on its own. These strategies include analyzing regression tests, reviewing escalated transcripts, and examining top escalation drivers. By deploying each strategy, Autopilot makes steady progress towards the stated goal.
By deploying these new strategies, one customer saw a 30% increase in deflection within a single experiment. Delivering strategies through AOPs also opens the door to future optimization. Strategies can be developed for any outcome, whether that’s campaign performance or CSAT.
A compounding loop
Experiments and Decagon Labs strategies are useful on their own, but together they contribute to the system that helps Autopilot learn. Every input sharpens the next round of suggestions, so each improvement makes the following one easier to find.
As more strategies and goals are developed, Autopilot’s loop will extend beyond resolution rate to the outcomes you care about most.
Ready to put Autopilot to the test? Book a demo to see it in action.
Start improving your workflow with Decagon
With Decagon, CX teams don’t have to guess whether a change will improve CSAT or deflection. They can move quickly, measure what matters, and act on what works.
Join us
There are very few places where you can prototype with frontier LLMs, ship to production in days, and watch users engage with the systems you built—all while owning the entire stack, from intent parsing and tool usage to API integration and observability. This role at Decagon is one of those places.
From my own experience working across both agent development and broader engineering initiatives at Decagon, I’ve seen firsthand how uniquely impactful this work can be. Whether I’m building intelligent workflows for customers or designing infrastructure that supports our agent platform, it’s rare to find an environment where the work transitions from concept to production within days, actively powering user experiences and transforming how businesses operate.
If you’re looking for a role where you can:
- Build at the frontier of LLMs, automation, and user interaction
- Deploy AI agents that solve high-value business use cases across industries including retail, travel and hospitality, fintech, edtech, and more
- Work directly with customers on high-impact use cases
- Ship fast, iterate constantly, and own your work from idea to production
- Join a fast-moving, collaborative team solving real-world challenges with AI
We’d love to hear from you!
.png)
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
The AI concierge for every customer.