0:00
我很高兴宣布我们最新的Langshane学院课程——构建可靠的
0:03
Langsmith代理。
0:04
我们正在解决代理工程团队如今面临的最大挑战之一，
0:09
无论你是搭建系统的工程师还是负责交付的产品经理，
0:10
将代理从原型推向生产环境都是一项艰巨的任务。
0:13
可靠性并不新鲜，工程师们
0:14
数十年来一直致力于提升系统的正常运行时间、准确性和性能。那么为什么代理这么难处理呢？
0:20
传统软件有明确的规则和确定性的逻辑。
0:21
出现问题时，你查看日志，找到代码行。代理的代码只是骨架，
0:27
它定义了工具和
0:29
提示。真正的决策发生在一个非确定性的模型内。你无法仅从代码库中
0:34
阅读模型的
0:35
推理，但你可以追踪从初始输入到最终输出的决策过程。
0:39
这种可观测性是构建可靠代理的基础。
0:41
本课程的目标是通过持续迭代，将代理从首次运行打磨成生产就绪系统。
0:47
你将学习如何使用
0:48
我们的代理工程平台Langsmith来观察、评估和部署代理。
0:53
你会从第一次分析周期开始，运行代理，找出问题，
0:55
并通过追踪而非日志进行调试。修复失败后，还要确保它们不再出现。
1:00
你会构建评估来捕捉回归并随着时间追踪质量。
1:04
最后进入生产阶段，每天成千上万用户使用你的代理，
1:05
你需要一个既能随业务增长又成本高效的测试系统。
1:09
到课程结束，
1:10
你将拥有完整的工具包来构建、测试和运营任意规模的可靠代理。
1:14
我非常期待你投入学习，掌握用Langsmith构建可靠代理的方法。
1:20
你可以今天就报名参加。
0:00
I'm excited to announce our latest Langshane Academy course, Building Reliable
0:03
Agents with
0:04
Langsmith. We're tackling one of the biggest challenges Agent Engineering teams
0:09
face today,
0:10
whether you're an engineer building the system or a product manager responsible
0:13
for shipping it,
0:14
bringing agents from prototype to production is difficult. Reliability is not
0:20
new. Engineers
0:21
have spent decades refining uptime, accuracy, and performance. So why are
0:27
agents so hard to get right?
0:29
Traditional software has clear rules and deterministic logic. Something breaks,
0:34
you check the logs,
0:35
you find the line of code. With agents, the code is just scaffolding. It
0:39
defines the tools and the
0:41
prompt. The real decisions happen inside a non-deterministic model. You can't
0:47
read a model's
0:48
reasoning in your code base, but you can trace its decisions from first input
0:53
to final output.
0:55
This observability is the foundation of building reliable agents.
1:00
The goal of this course is to take an agent from first run to production-ready
1:04
system
1:05
through iterative cycles of improvement. You'll learn how to do this with Lang
1:09
smith,
1:10
our Agent Engineering platform for observing, evaluating, and deploying agents.
1:14
You'll start with your first analysis cycle, running your agent, finding what
1:20
breaks,
1:20
and debugging with traces, not logs. Once you fix those failures, you'll make
1:25
sure they don't
1:25
come back. You'll build evals that catch regressions and track quality over
1:30
time.
1:30
Finally, production. Thousands of users hitting your agent daily. You'll need a
1:36
testing system
1:37
that scales the grow while staying cost-efficient. By the end, you'll have the
1:42
complete toolkit
1:43
to build, test, and operate reliable agents at any scale.
1:48
I'm really excited for you to dive in and learn how to build reliable agents
1:51
with Langsmith.
1:53
You can enroll today.
回复