The Agentic AI Digest (16/09)| Agentic Design Patterns & Prompting Secrets
This week: Agentic Design Patterns, a deep dive into Prompt Engineering, and what the community is building.
Hi everyone,
Welcome to your weekly briefing from the Agentic AI Roundtable. Our goal is to cut through the noise and deliver the most relevant signals, patterns, and community wins to help you build more effectively.
Let's dive in.
📡 On the Radar: What's Moving the Needle
A curated look at the articles, papers, resources and updates that are worth your time this week.
Check out this epic book in a Google Doc, Agentic Design Patterns, and the accompanying NotebookLM by Google Cloud Snr. Director and Distinguished Engineer, Antonio Gulli
There were a ton of A2A and ADK announcements this week including that custom remote A2A Agents are coming to AgentSpace, and that JS and GoLang versions of the ADK are just around the corner
Nano Banana! Try out the new Gemini image generation on AI Studio.
📒From the Workbench: Patterns to Pocket
This week, let’s talk about the most important skill for getting real value from Generative AI: prompt engineering. A well-crafted prompt is the difference between a generic answer and a brilliant one. The core principle is to think of the model as a highly capable person who has zero context about your project. Your job is to provide that context.
Here are a few key techniques to elevate your prompts:
Start Simple and Iterate: Don't copy a complex prompt and tweak it. You might inherit complexity that steers the model in the wrong direction. Start with the core action you want (e.g., "analyze," "generate," "recommend") and build from there.
Prefer Instructions over Constraints: Guide the model with positive instructions ("Use vivid and exciting language") rather than negative constraints ("Don't be boring"). Telling the AI what you want is far more effective than telling it what to avoid.
Provide Examples (Few-Shot Prompting): Show the model what a good output looks like. Providing 3-6 high-quality examples of what you're looking for is one of the most powerful ways to improve the accuracy and format of its responses.
Use Chain-of-Thought Prompting: For tasks requiring logic, force the model to slow down and "show its work." Simply adding a phrase like "Let's think step by step" encourages the model to break down the problem, leading to more accurate reasoning.
The best workflow is iterative: research the topic to ask better questions, craft a simple prompt, test it, and refine it based on the output. A poor answer is a signal that your prompt may not be as clear as you think. For more great insights, check out this discussion on advanced prompting strategies and these resources.
Alis Build AI Skills Boost III: Prompt Strategy
Dive Deeper with These Resources
A well-crafted prompt is the difference between a generic answer and a brilliant one.
Explore the official guide on prompt design. This is your foundation for today: Prompt Design Strategies
Create your own "Prompt Refiner." Write a prompt and test it. Create a Gem with the knowledge from this whitepaper: Google Whitepaper on Prompt Engineering
🛠️ Community Commits: Building in the Open
This is where we track our collective progress and celebrate what you're building.
Watch Me Build: Intro to Building Production Ready A2A Agents in Google Cloud
Each week, Nicholas will incrementally take us along the journey of building production grade agents and multi-agent systems from scratch on Google Cloud. This week’s session introduces the core concepts underpinning agentic and A2A systems.
Show & Tell: The End of Copy-Paste: AI Answers Directly in Your Documents
This week, we're highlighting Madelein and her project, Due Diligence Exchange.
What it is: A generative AI agent that automates the tedious process of completing due diligence questionnaires by generating grounded answers from business documents and previously answered questions, inserting them directly into MS Word.
The Insight: "The real breakthrough wasn't just generating grounded answers, but eliminating the copy-paste nightmare. By allowing the AI to write directly into MS Word, we transformed a multi-day research task into a streamlined, in-document workflow."
Incredible work! Feel free to ask questions in the community forum thread.
Want to get involved?
The best way to contribute is to share your work, any links or useful resources, ask a question or leave a review on the weekly Open Thread in the Community Chat Space [link] or send a note on the Community Google Group.
Until next week, keep building.
The Agentic AI Roundtable Core Team