Welcome back to the latest weekly issue! This week, we explore agent design patterns, examine stats about prompt formatting impact on LLM performance, and look at a new protocol for LLM integration. If you like the content please support the newsletter by sharing with your network π¦ Twitter, πΌ LinkedIn, π¦BlueSky, βοΈ Email
This Week in Prompt Patterns π€
4 Agentic Design Patterns π§©
JSON Prompts are 40% Better π
New Protocol for LLM Integration π
2024 State of ML in Production π
Lessons Learned from Notion using LLMs π§
And more π
π― In the Spotlight
Andrew Ng introduces agent design patternsπ providing a framework to conceptualize and implement AI agents, categorized into four fundamental types.
These patterns combined with the diagrams from Pankaj (link further down) are a must check!
π° Interesting Articles
Does Prompt Format Matter? - New research examines impact of different prompt formats (plain text, Markdown, JSON, YAML) on LLM performance. JSON templates outperform others by 40% in code translation tasks, while GPT-4 shows more format resilience compared to GPT-3.5.
What is an AI Agent? - LangChain team breaks down the spectrum of agent capabilities, moving beyond binary definition.
State of Production ML in 2024 - Industry survey revealing current practices and challenges in production ML deployments.
Agentic AI Design Patterns - Practical guide to implementing the top 4 design patterns in AI systems (visual above).
Good Enough Prompting - Pragmatic approach to effective prompt engineering without overcomplication.
State of AI Agents - Comprehensive overview of current agent architectures, implementations, and emerging patterns.
ποΈ YouTube & Podcasts
Building Notion AI: Lessons Learned - Simon Last, Notion Co-Founder and CTO, shares practical insights from implementing AI in a production environment.
Compound AI + Open Source - Lin Qiao, Fireworks AI CEO, discusses why the combination of open source and AI creates stronger solutions.
A Practical and Tactical Approach to Temporal and AI - Deep dive into common challenges of orchestrating agentic system design.
Systems Observability with AI and LLMs - Jason Hand examines how LLMs are transforming observability through automated analysis, anomaly detection, and incident response.
π Social of the Week
Agentic AI Architectures - Visual breakdown of key agent design patterns with implementation examples.
π AI Laughs
π New Tools & Releases
Model Context Protocol (MCP) - Anthropic's open protocol for standardizing LLM integration with external tools and data sources.
Amazon Bedrock Agents Implementation - SA-ready patterns for building AI assistants using Amazon's managed service.
Data Engineering Handbook - Comprehensive guide covering modern data engineering practices and tools.
π Learning Hub
Prompt Engineering for Generative AI - The most comprehensive guide to prompt writing for text and image generation, by @hammer_mt & @jamesaphoenix12. Check out my short review here.
π Until Next Week
Thanks for reading Prompt Patterns π If you like the content please support the newsletter by sharing with your friends. If you have any topics for this newsletter, just drop me an email!
Have a great rest of the week π