WEBVTT

1
00:00:01.300 --> 00:00:10.771
This is Anthropic, Read Aloud. Today: How Cresta turned CX expertise into an agent builder on the Claude Agent S D K.

2
00:00:12.071 --> 00:00:17.953
How Cresta turned CX expertise into an agent builder on the Claude Agent S D K.

3
00:00:18.403 --> 00:00:28.123
See how Cresta built Conductor, an agent that builds other agents, on the Claude Agent S D K, and how the team evaluates it with every new Claude model.

4
00:00:28.573 --> 00:00:32.605
From Claude Blog, published October 5, 2026.

5
00:00:33.605 --> 00:00:49.733
In our series, How startups build with Claude, we look at the decisions behind A I products. Here, we explore how Cresta brought its expertise building A I-native customer experience into Conductor, an agent that helps developers build and improve other agents.

6
00:00:50.183 --> 00:00:51.583
Table with 5 rows.

7
00:00:51.823 --> 00:00:53.127
Header: The quick pitch.

8
00:00:53.367 --> 00:00:55.115
The quick pitch: Name; Cresta.

9
00:00:55.355 --> 00:00:58.284
The quick pitch: Founded; 2017.

10
00:00:58.524 --> 00:01:00.622
The quick pitch: C E O; Ping Wu.

11
00:01:00.862 --> 00:01:16.390
The quick pitch: Growth; 270 million dollars+ raised. 100 million dollars+ ARR. Helping organizations like United Airlines, CVS Health, and Marriott deliver exceptional customer experiences.

12
00:01:16.840 --> 00:01:29.406
Building enterprise A I agents for customer experience use cases requires understanding the context embedded across a wide range of customer interactions. For example, customers may:

13
00:01:29.856 --> 00:01:32.923
Ask for a refund without saying which purchase they mean.

14
00:01:33.163 --> 00:01:37.295
Report a billing problem and a login issue in the same message, or.

15
00:01:37.535 --> 00:01:40.613
Request a refund after the return window has closed.

16
00:01:41.063 --> 00:01:49.126
To respond correctly, the agent may need to check what the customer bought, the refund policy, their account history, or the payment status.

17
00:01:49.576 --> 00:02:07.275
Cresta sees these interactions at volume. Its platform powers customer experience, A I agents that handle conversations on their own, provide real-time guidance for the human agents working alongside them, and surface conversation intelligence that shows the business where to improve.

18
00:02:07.725 --> 00:02:28.760
Building and maintaining customer experience agents requires organizations to make key decisions on operating requirements: what the agent needs to know, which systems it must access, and where its behavior should be flexible or deterministic. Cresta wanted to make that judgment available to more teams without each one starting from scratch.

19
00:02:29.210 --> 00:02:47.374
Conductor is the result: a natural language agent builder that helps teams turn complex business context into production-ready agents. Users describe what they want to build, and Conductor guides the work from a grounded blueprint through implementation, evaluation, and optimization.

20
00:02:47.824 --> 00:03:05.441
Cresta originally built an early version of Conductor using Claude Sonnet and Claude Opus to help internal teams support customer deployments. The open question was whether Cresta could turn that system into a product customers could use to build agents directly within the platform.

21
00:03:05.561 --> 00:03:18.447
To carry out open-ended development work, Conductor uses the Claude Agent S D K as a general-purpose harness to gather context, use tools, write and run code, and adapt based on the results.

22
00:03:18.897 --> 00:03:37.810
“We used Claude Code internally as a dev tool because it was very effective at the software development aspects of building an agent,” said Renjie Li, Engineering Lead for Cresta Conductor. “The Claude Agent S D K gave us a way to use that same agentic development harness programmatically.

23
00:03:37.930 --> 00:03:50.465
With Conductor, we’ve connected it with Cresta’s deep conversation intelligence, CX expertise, evaluations, and runtime to give users the agent-building experience we wish we’d had from day one.”

24
00:03:50.915 --> 00:03:58.695
Cresta reports that in early use cases across Cresta and its partners, Conductor cut initial deployment time roughly in half.

25
00:03:59.495 --> 00:04:02.078
A meta-agent that helps build the next agent.

26
00:04:02.528 --> 00:04:23.029
Conductor is a self-improving meta-agent built on established best practices for CX agent development. As teams build and refine more agents with it, they capture their own proven patterns, workflows, and expert judgment. This creates a knowledge flywheel that makes future development faster and more consistent.

27
00:04:23.479 --> 00:04:30.152
After launch, production interactions, workflow outcomes, and feedback become new signals for the next improvement.

28
00:04:30.602 --> 00:04:46.858
Conductor helps builders turn what they learned from an agent build into a reusable memory artifact, capturing patterns, business rules, and evaluation approaches. They can then share it as skill with the rest of the team, so others can build from the same proven practices.

29
00:04:47.658 --> 00:04:49.793
Making the important decisions easier.

30
00:04:50.243 --> 00:05:02.869
Writing the first prompt is only one part of building a useful agent. One of the harder calls is where a conversation should stay flexible and where it needs explicit business rules and controlled tool behavior.

31
00:05:03.319 --> 00:05:20.687
"You don't want to go too deterministic, otherwise you create a giant decision tree trying to map out every possible branch... but there are enterprise-critical workflows that you need to make sure work all the time, especially for highly regulated industries with low risk tolerance," said Renjie.

32
00:05:20.807 --> 00:05:29.959
"Conductor helps discover and harden those deterministic requirements and the implementation with the flexibility to improve the overall experience."

33
00:05:30.409 --> 00:05:51.465
Conductor draws on historical conversations to help teams identify where a human should stay in the loop and where the model can be given room to adapt. Once those boundaries are set, developers still need to test the deterministic pieces against the requirements they're meant to enforce, and revise those tests as the underlying business rules change.

34
00:05:52.265 --> 00:05:54.171
How Cresta evaluates Conductor.

35
00:05:54.621 --> 00:06:01.618
Conductor is measured on how well it builds agents. To that end, Cresta runs Conductor through a set of build tasks:

36
00:06:02.068 --> 00:06:03.215
Build a new agent.

37
00:06:03.455 --> 00:06:04.759
Write a test case for one.

38
00:06:04.999 --> 00:06:07.106
Change an agent that already exists.

39
00:06:07.346 --> 00:06:09.081
Perform root cause analysis.

40
00:06:09.531 --> 00:06:17.497
Finishing the task is not the only bar. The team also looks at the outcomes, how Conductor got there, and what it used along the way.

41
00:06:17.947 --> 00:06:19.347
Table with 5 rows.

42
00:06:19.587 --> 00:06:22.088
Header: Evaluation; Question it answers.

43
00:06:22.328 --> 00:06:29.269
Evaluation: Outcome; Question it answers: Did Conductor produce the requested, usable artifact?.

44
00:06:29.509 --> 00:06:36.547
Evaluation: Execution path; Question it answers: Did it use the appropriate tools and required context?.

45
00:06:36.787 --> 00:06:43.038
Evaluation: Quality; Question it answers: How does the result compare with a curated reference?.

46
00:06:43.278 --> 00:06:50.028
Evaluation: Resource use; Question it answers: How much time and model usage did the task require?.

47
00:06:50.478 --> 00:07:08.270
Those same tasks function as a safety net. When a new Claude model is launched or the Conductor framework is updated, Cresta reruns its evaluations. It uses the same tasks and same scoring to identify what improved and what still needs work before anyone commits to the change.

48
00:07:08.720 --> 00:07:25.200
That same evaluation discipline is built into Conductor, which allows its users to check and see if their agents are performing as intended. Conductor turns requirements and edge cases captured during the build phases into tests within its evaluation module.

49
00:07:25.650 --> 00:07:42.676
Teams can then monitor the workflows that need to work every time as policies, connected systems, and conversation patterns change. That gives them a practical way to update an agent as the business changes without losing the reliability of the workflows they have already hardened.

50
00:07:43.476 --> 00:07:46.491
Building Conductor on the Claude Agent S D K.

51
00:07:46.941 --> 00:08:04.689
Cresta designed Conductor as the CX-specific control plane for agent development. It brings together conversation data, domain expertise, and development practices, then governs each agent run through policy, observability, verification, and feedback-driven improvement.

52
00:08:05.139 --> 00:08:22.152
The Claude Agent S D K sits beneath Conductor as a general-purpose execution harness. It manages the work of gathering context, calling tools, and writing and running code; Conductor layers Cresta's CX workflows, tools, and domain context on top.

53
00:08:22.602 --> 00:08:41.719
The team evaluated the S D K against the work Conductor needed to perform: tasks with multiple steps, substantial context, and repeated tool use. Its review covered data privacy, tenant architecture, organization-level key distribution, controls, and observability across the implementation.

54
00:08:42.169 --> 00:08:57.504
"Anthropic gives us a strong, general-purpose foundation for agentic software development. That lets us put our engineering investment where Cresta creates the most durable customer value," said Xiangru Chen, VP of Engineering at Cresta.

55
00:08:58.304 --> 00:09:02.580
Putting Cresta's engineering expertise directly in customers' hands.

56
00:09:03.030 --> 00:09:22.529
Conductor started inside Cresta, helping its own forward-deployed team learn which development tasks could become repeatable, where context mattered most, and what builders needed to inspect along the way. Its success in that internal work gave the team confidence to bring the same capabilities to customers and partners.

57
00:09:22.979 --> 00:09:37.013
For any team whose product deploys custom agents faster than its engineers can hand-build them, Cresta's approach is a useful reference point: with agents or meta-agents, the work after the initial build is often the hardest part.

58
00:09:37.463 --> 00:09:44.396
See how Conductor turns business context into better agent design, implementation, and ongoing improvement.

59
00:09:44.846 --> 00:09:46.247
Table with 5 rows.

60
00:09:46.487 --> 00:09:48.897
Header: Best practices from the Cresta team.

61
00:09:49.137 --> 00:10:01.930
Best practices from the Cresta team: Dogfood your products; Conductor ran as an internal tool first, and only became a product once its output matched what an engineer with deep A I expertise would build..

62
00:10:02.170 --> 00:10:21.632
Best practices from the Cresta team: Focus on your unique advantages; Cresta drew a hard line between the expertise only they could bring (their conversation data and understanding of customer experience) and what they could take off the shelf from the Agent S D K, and refused to build anything on the wrong side of it..

63
00:10:21.872 --> 00:10:37.073
Best practices from the Cresta team: What parts of your agent need to be deterministic; Harden the enterprise-critical workflows that can never go wrong and leave the rest flexible, or you won't end up with the interactive agentic experience you want..

64
00:10:37.313 --> 00:10:55.084
Best practices from the Cresta team: Getting to production is the easy part; The initial build is only about 20 percent of the effort, the other 80 percent is the testing, optimization, and continuous improvement that follows, so design your tooling around the iteration loop, not the launch..

65
00:10:56.084 --> 00:11:15.584
That was How Cresta turned CX expertise into an agent builder on the Claude Agent S D K, from Claude Blog, published October 5, 2026. Read the original at claude dot com. This unofficial audio edition was generated automatically for the podcast Anthropic, Read Aloud.