How Do I Get Started Building Agents?
Read the Portuguese translation →
Author: LangChain · Watch on YouTube
Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.
Original language: English
Original video: How Do I Get Started Building Agents? · Channel: LangChain · 01:01 · Watch on YouTube
Topics: AI agents, AI evaluation and benchmarks
Summary
Section titled “Summary”The speakers recommend different ways to start building agents depending on a developer’s experience and comfort with code. Developers comfortable building agents can use Deep Agents, which includes task planning, file systems for context management, sub-agents, and long-term memory. Beginners or those who prefer not to code can use Agent Builder to create and share agents using natural language. LangSmith can be used to test, deploy, and monitor agents, evaluate performance, and validate improvements, and it works with any framework or model.
Key points
Section titled “Key points”- Deep Agents is presented as an easy starting point for developers already comfortable building agents.
- Deep Agents includes built-in task planning and file systems for context management.
- Deep Agents supports creating sub-agents and long-term memory.
- Agent Builder is a no-code option for people new to agents or less comfortable working with code.
- Users describe what they want in natural language, and Agent Builder builds the agent.
- Agents built with Agent Builder can be shared across a team.
- LangSmith provides tools for testing, deploying, and monitoring agents.
- LangSmith evals can measure agent performance and help validate improvements before shipping.
- LangSmith works with any framework or model.
Chapters
Section titled “Chapters”| Start | Chapter | Summary |
|---|---|---|
| 00:00 | Starting with Deep Agents | The speakers recommend Deep Agents for developers comfortable building agents and describe its built-in capabilities. |
| 00:31 | Building agents without code | Agent Builder lets users create agents with natural language and share them with their team, while LangSmith supports testing, deployment, and monitoring. |
| 01:01 | Evaluating agents with LangSmith | The speakers describe using evals to measure performance and validate improvements, noting that LangSmith works with any framework or model. |
Quotes
Section titled “Quotes”Deep Agents is the easiest way to start building agents. — 00:00
It has built-in task planning, file systems for context management. — 00:00
You describe what you want, Agent Builder builds it for you — 00:31
LangSmith, which is our platform for testing, deploying and monitoring agents. — 00:31
LangSmith actually works with any framework or model you choose. — 01:01
Concepts and entities
Section titled “Concepts and entities”Deep Agents, Agent Builder, LangSmith, LangChain, Sean
Full transcript (original language)
Section titled “Full transcript (original language)”00:00 How do I get– Oh, wow. The suspense is killing me. Started building agents. Oh, this is a great– this is a great final question, actually. It’s an all-encompassing question. Yeah, yeah. The way I think about it is, if you’re a developer who is, I’d say, pretty comfortable building agents, Deep Agents is the easiest way to start building agents. It has built-in task planning, file systems for context management. You can create sub-agents and long-term memory. Or if you’re brand new to building agents and you’re not really comfortable or you don’t want to get hands on.
00:31 With code, you could start building agents with Agent Builder. That’s our no code agent builder where you build agents using natural language. You describe what you want, Agent Builder builds it for you, you can then share it across your entire team. So it’s kind of cool then. It’s like kind of flexing a little bit like, “Look, there’s cool agent I just built.” - Either way, you can integrate your agent into LangSmith, which is our platform for testing, deploying and monitoring agents. - Yeah, great. I mean, as we cover that’s way more important than actually building the agent. And I think with LangSmith, you can actually identify where decisions went wrong,
01:01 what your agent was reasoning. You can use these evals that Sean’s super excited about to measure agent performance and also validate your improvements before shipping. But I think the best part is that LangSmith actually works with any framework or model you choose. Oh, so it’s not just LangChain, it’s not just Deep Agents, it’s not just Agent Builder, like– Anything. Yeah.