AI Agents Weekly Digest - April 12, 2025

MCP, A2A, and Prompting - a Big Week for Agents

April 14, 2025

(I know this got sent late, I was drafting it on Saturday but didn’t get around to finishing it until just now)

Google’s Agent-to-agent protocol and Agent Development Kit releases at Google Next were huge this last week. Our Sub also featured some discussion around MCP best practices and discussions around prompting.

  1. 3 Rules from Anthropic to Build Agents

    1. The three ideas provided are:

      1. Don’t build agents for everything

      2. Keep it simple

      3. Think like your agent

    2. There’s a lot more, but that’s the basics

    3. I disagree with one of the statements about agents running in loops - I don’t think agents should be in loops, but rather event based

  2. 4 “levels” of Prompt Engineering

    1. Very main character energy

    2. Reported for spam, but also very popular so I kept it

    3. The levels proposed are:

      1. Tourist (basic)

      2. Template User (basic)

      3. Engineer (what you do if you can play around with things)

      4. Architect (probably what you are if you’re reading this)

  3. Google’s A2A Announcement

    1. This was a surprisingly horny comment section, I don’t know why

    2. Really exciting though, combined with the “multi-xyz” theme in the projects section, I think it’s clear that the agent to agent protocol is gonna be big

  1. Multi-Agent Framework “Prompits”

    1. Open source framework for peer to peer agent communication

    2. Basically the agent to agent protocol but it’s not by Google

    3. Great vision

  2. Multi-Agent Cross Verification

    1. I don’t really consider this multi-agent but the theory makes sense

    2. It’s basically LLM as a judge

  3. Complex, Multi-database Queries

    1. Multi-xyz things are getting popular

    2. This is actually a hard problem because you need a lot of understanding of the underlying database structures

Community Message Board

Special Announcements

Remember our hackathon is coming up in a month!

There are currently:

  • Four Judges

    • A Director from Databricks

    • A GP from Banyan VC

    • A Senior Solutions Architect from AWS

    • A Senior SDE from AWS

  • Participation from

    • LanceDB

    • GMI Cloud

    • CopilotKit

    • AgentR