AI Agents Weekly Digest - May 18, 2025

Claude 3.7's full 24,000 token context window, the current state of Agents, and guides from all the big companies

Author

Yujian Tang
May 18, 2025

Hello everyone, many of you are new and haven’t received one of these AI Agents Weekly recap newsletters, so welcome to the crew. If you’re receiving this, and you didn’t sign up directly, it’s because I’ve moved over the people who have signed up for events from the OSS4AI calendar. This is a newsletter that covers weekly news from AI Agents as we see it on our 130k+ person subreddit + my personal meta commentary.

For those of you that are used to receiving this newsletter, you may notice that we’ve rebranded from “AI Agents Weekly” to “The AI Hatchery”. This reflects our focus on providing all developer the resources they need to understand, use, and contribute to the development and direction of AI. You will still continue to receive these weekly updates, and on top of that, we’re in the process of adding a "Community Expert Contributions” which will consist of educational pieces from community experts who have made significant contributions to software engineering and AI as a whole.

Special Announcements

We opened up our Community Experts Program this last Thursday and received over 100 applications already! If you’re interested in being a community expert, please apply here.

We’re partnering with Arize AI for Observe on June 24th in SF and I will be giving a talk on Agent Architectures! Use code OSS4AI or buy tickets through this link for 75% off.

  1. Claude 3.7’s prompt leak

    1. The link to the leaked gist

    2. The top comment is about how this post is AI generated and how AI generated posts all sound the same now

    3. I agree, there’s a ton of AI generated posts at the moment, and they have this “influencer” or “hype man” tone to them. I’m sure this will get changed as LLMs get better

  2. What people are getting wrong about AI Agents in 2025

    1. This is a great perspective + value post

    2. Here’s some tips from this person’s POV summarized:

      1. Multiple specialty agents are outperforming one generalized agent

      2. Backend optimization agents are more valuable than current customer facing ones

      3. Human in the loop > autonomous agents

      4. Multi-agent systems > single agents

      5. Proactive agents > reactive agents

  3. Free Guides from Anthropic, Google, and OpenAI

    1. Prompting Guide – Google

      Building Effective Agents – Anthropic

      Prompt Engineering Guide – Anthropic

      A Practical Guide to Building Agents – OpenAI

      Identifying and Scaling AI Use Cases – OpenAI

      AI in the Enterprise – OpenAI

    2. An addition from another user of HuggingFace guides:

      1. Vision Language Models

        How to Build an MCP Server

        Agents Course (can get certified)

        Using AutoRound for Quantization

Reply

or to participate.