# Tips for building AI agents

> For agents: start with the [agent guide](https://ratstack.sh/llms.txt). Every page is Markdown by default; add `Accept: text/html` for HTML.

[Anthropic](https://www.youtube.com/@anthropic-ai) · Published 2025-02-13 · 18:19 · B tier · Agents and MCP

[Watch on YouTube](https://www.youtube.com/watch?v=LP5OCa20Zpg) · [All RatTube videos](/tube)

## Summary

The speakers distinguish predetermined workflows from agents that choose what to do next. Fixed sequences can transform inputs through known steps, while agentic work needs room for exploration.

They emphasize tool descriptions the model can understand, valuable and complex tasks where errors are cheap to catch, and a way to measure results. The discussion also explores interactions between multiple agents and the uncertainty around where those approaches help.

## Section notes

- [0:29](https://www.youtube.com/watch?v=LP5OCa20Zpg\&t=29s): The conversation begins by separating definitions of agents from broader LLM applications.
- [3:18](https://www.youtube.com/watch?v=LP5OCa20Zpg\&t=198s): A fixed workflow has known steps and intermediate checks.
- [6:43](https://www.youtube.com/watch?v=LP5OCa20Zpg\&t=403s): Tool use still requires instructions that make sense from the model's perspective.
- [9:56](https://www.youtube.com/watch?v=LP5OCa20Zpg\&t=596s): Task value, complexity, and the cost of errors affect agent suitability.
- [13:18](https://www.youtube.com/watch?v=LP5OCa20Zpg\&t=798s): A game illustrates multiple agents communicating in different roles.
- [16:30](https://www.youtube.com/watch?v=LP5OCa20Zpg\&t=990s): Measuring results provides feedback on whether the system works.

## Key concepts

No matching wiki page yet.

## Our take

Anthropic's applied AI and research people define agents against workflows, look at the anatomy of an agent prompt, and give advice on where agents help and where they don't.

## Related videos

- [Code Mode: Let the Code do the Talking - Sunil Pai, Cloudflare](/tube/code-mode-let-code-do-talking)
- [All Code Is Slop (Except Mine) — Lauren Tan, SpaceXSI](/tube/all-code-is-slop-except-mine-lauren-tan)
- [How We Build Effective Agents: Barry Zhang, Anthropic](/tube/building-effective-agents-barry-zhang)
- [The Future of MCP — David Soria Parra, Anthropic](/tube/future-of-mcp-david-soria-parra)

Summary based on the transcript. Written 2026-11-15.
