π£ Welcome back! Hereβs everything that happened in GenAI in June.
π Share: X Β· LinkedIn Β· BlueSky Β· Email and letβs dive in.
π₯ Top Picks from June
π The Prompt Engineering Playbook for Programmers β Addy Osmani π₯
π€ How we built our multi-agent research system β Anthropic π₯
π‘ Software Is Changing (Again) β Andrej Karpathy π₯
π The future of AI agentsβand why OAuth must evolve β Alex Simons
π§βπ» Why Go is a good fit for agents β Hatchet Team
π οΈ Design Decisions Behind app.build, a Prompt-to-App Generator β Neon
βοΈ Understanding the KV Cache in LLMs from Scratch β Sebastian Raschka
π’ Zapier's AI-first hiring and onboarding β Zapier
π₯ 12-Factor Agents β Dex Horthy π₯
π§ Context Engineering β Langchain
π This AI Agent Should Have Been a SQL Query β Gunnar Morling
𧬠Prompt Engineering β Lilian Weng π₯
β¦and if you dare, 100+ updates to explore π
π― Featured Article
The Prompt Engineering Playbook for Programmers - Addy Osmani demonstrates how prompt engineering has become an essential skill for developers using AI coding assistants, outlining systematic approaches to craft effective prompts that yield high-quality code solutions rather than generic outputs. A MUST READπ
π° Long-Form Reads
Understanding and Coding the KV Cache in LLMs from Scratch - Sebastian Raschka breaks down KV caching for faster LLM inference with code examples. π
Block's Playbook for Designing MCP Servers - A practical guide to mission-critical MCP server design.π
The Dark Side of "Just Hooking Up" AI Agents to GitHub - Folks at AI Native Dev examines the security risks and potential vulnerabilities that emerge when developers integrate AI agents with GitHub repositories without proper safeguards or oversight.
Do Machine Learning Models Memorize or Generalize? - The author investigates the "grokking" phenomenon where models unexpectedly learn to generalize after extended training.
GitHub - humanlayer/12-factor-agents - principles for production-quality LLM agents based on real-world implementation patterns.πππ
Evaluating Large Language Model (LLM) systems: Metrics, challenges, and best practices - Jane Huang breaks down systematic LLM evaluation beyond simple prompt testing.
How we built our multi-agent research system \ Anthropic - reveals how Anthropic built Claude's multi-agent research system for complex information gathering. πππ
How Google is accelerating code migrations with AI - Abi Noda reveals Google's AI techniques for faster, more efficient code migrations.
Prompt Engineering - Lilian Weng details techniques for steering LLM behavior without updating model weights, emphasizing that prompt engineering is an empirical science requiring experimentation across different models.ππ
Vibe coding is rewriting the rules of technology - Kiara Nirghin explains how AI-powered development prioritizes intent over code syntax.
Kai Waehner explains how OpenAI uses Apache Kafka and Flink for GenAI
The "Trust, But Verify" Pattern For AI-Assisted Engineering - Addy Osmani explains why developers must treat AI coding assistants like junior developers, requiring thorough verification of AI-generated code while maintaining human accountability for all output.
How AI generated code compounds technical debt - LeadDev - Bill Doerrfeld examines how LLM-based coding assistants are making it easier to create code but simultaneously undermining engineering best practices like DRY, leading to increased code duplication and declining quality.
The future of AI agentsβand why OAuth must evolve - Alex_Simons examines the need for collaborative development of identity standards to enable AI agents to securely access and operate across different systems.
LLM Shibboleths determine AI effectiveness - Author reveals why developer expertise determines AI coding assistant value.
Highlights from the Claude 4 system prompt - Simon Willison unpacks Claude 4's system prompts, revealing AI guardrails and capabilities.
Building software on top of Large Language Models - Simon Willison shares insights on LLM tools and free content philosophy.
How ChatGPT Remembers You: A Deep Dive into Its Memory and Chat History Features - wunderwuzzi's
My AI Skeptic Friends Are All Nuts Β· The Fly Blog - Thomas Ptacek argues that LLMs have fundamentally changed software development despite skepticism from tech leaders, noting that talented developers are refusing to use tools that already outperform them purely out of resistance to change.
Evaluation Driven Development for Agentic Systems. - Aurimas GriciΕ«nas reveals how evaluation-driven methodologies improve AI agent development.
Claude Code saved us 97% of the work β then failed utterly - Thoughtworks: AI coding assistant shows promise but has critical limitations.
The Open-Source Toolkit for Building AI Agents - Sahar Mor reveals essential open-source tools for creating AI agents.
What Even Is Vibe Coding? - Ashley Willis: From skepticism to curiosity about Karpathy's "vibe coding" concept.
Domain-Driven RAG: Building Accurate Enterprise Knowledge Systems through Distributed Ownership - George Panagiotopoulos reveals how distributed domain ownership improves RAG system accuracy. π
Announcing AWS Security Reference Architecture Code Examples for Generative AI | AWS Security Blog - AWS launches security templates for securing generative AI implementations.
Developing Apps by Chat: How Far Can Lovable 2.0 Really Go? - Baptiste Fernandez examines the potential and boundaries of conversational app development with Lovable 2.0.
Zapier's AI-first hiring and onboarding - New hires must now demonstrate AI fluency as standard requirement.π
The last six months in LLMs, illustrated by pelicans on bicycles - Simon Willison covers recent LLM developments with creative imagery, ChatGPT prompting tips for a UK official, plus curated links and quotations for AI enthusiasts.
Modernizing Legacy Struts2 Applications With Claude Code - Damien Gallagher demonstrates AI-assisted migration from Struts2 to modern frameworks.
Adopting Docs-as-Code at Pinterest - Pinterest Engineering reveals their shift to docs-as-code after traditional documentation methods failed.
How we built our AI code review tool for IDEs - CodeRabbit
God is hungry for Context: First thoughts on o3 pro - Ben Hylak reveals: OpenAI slashes o3 pricing while launching superior o3-pro variant.
Radar Trends to Watch: June 2025 β O'Reilly
SWE Agents Too Cheap To Meter, The Token Data War, and the rise of Tiny Teams - Latent.Space explains how recent AI coding agents like Codex and Jules are being offered at no extra cost, encouraging developers to maximize their usage while highlighting upcoming AI engineering events and the trend of smaller development teams.
Can AI Replace Software Architects? I Put 4 LLMs to the Test - CloudWay
LLMs are cheap - Mikael Snellman challenges the common misconception that Large Language Models are prohibitively expensive to operate
Why Go is a good fit for agents - Hatchet team notes Go's growing popularity in hybrid tech stacks for agent development.
Why agents are bad pair programmers - Justin Searls: AI agents code faster than humans think, creating poor pairing dynamics.
How I program with Agents - Crawshaw explores using AI agents as programming tools beyond simple autocomplete.
Securing AI agents: A guide to authentication, authorization, and defense - WorkOS explains how to secure autonomous AI systems in applications.
Agentic Coding Recommendations - Armin Ronacher reveals his cost-effective, token-optimized approach to AI coding assistants.
Don't Build Multi-Agents - Walden Yan from Cognition outlines why multi-agent systems fail and what works instead.
The state of AI agents - Lance Martin details the rise of autonomous background AI agents.
Context Engineering - Folks at Langchain examines techniques for managing LLM context windows efficiently.
AI Agents, meet Test Driven Development - Anita Kirkovska: Adapting TDD practices for more reliable AI agent development.
Takeaways from the AI Engineer World's Fair... - Patrick Ellis
Trying out the new Gemini 2.5 model family - Simon Willison reviews Gemini's new models with improved pricing and performance.
How to Vibe Code as a Senior Engineer - Alex Maccaw reveals how AI transforms coding from expensive to efficient.
Introducing MCP Catalog and Toolkit - Mark Cavage highlights Docker's vision for standardizing AI agent-tool connections with MCP.
AI as Normal Technology - Nicholas Bagley argues AI should be viewed as normal technology, not superintelligence.
The agent-first developer toolchain: how AI will radically transform the SDLC - Amplify Partners: AI needs new dev toolchains, not just upgrades to old ones.
Design Decisions Behind app.build, a Prompt-to-App Generator - Arseny Kravchenko details how Neon's app.build was engineered to prioritize reliability over feature complexity through scope limitations, FSM-guided tree-search, extensive validation systems, and an error analysis feedback loop.π
AI Engineering Goes Mainstream - Latent.Space shows the evolution of AI Engineering into a recognized professional discipline.
Benchmarking Multi-Agent Architectures - Will Fu-Hinthorn compares multi-agent systems and reveals key performance improvements.
Fuck You, Show Me The Prompt - Hamel Husain warns against tools that hide prompts from developers.
Gen AI Evaluation Service β Computation-Based Metrics - Mete Atamel dives into computation-based metrics in Vertex AI's evaluation service, explaining their deterministic nature, requirement for ground truth references, and limitations in capturing language nuances.
Why I'm Dialing Back My LLM Usage β Zed's Blog - Engineer's journey from LLM enthusiasm to practical skepticism.
Coding agents have crossed a chasm // flurries of latent creativity
Augmented Coding: Beyond the Vibes - Kent Beck shares insights from building a B+ Tree with AI assistance.
The AI-Native Software Engineer - Addy Osmani explains how software engineers can adopt an AI-native mindset, treating AI as a collaborative partner that can dramatically amplify productivity and creativity rather than viewing it as a threat. πππ
How to Stop Your Human From Hallucinating - Shrivu Shankar reveals strategies for managing AI hallucinations in workflows.
The DNA of AI Agents: Common Patterns in Recent Design Principles - Cedric Chee compares AI agent design frameworks, revealing shared patterns across different approaches.
From Prompt to Code - Paul Datta breaks down how Gemini CLI translates prompts into executable code.
The rise of "context engineering" - LangChain details how properly engineered context helps LLMs perform reliably.
Project Vend: Can Claude run a small shop? - Anthropic: AI-run store experiment shows Claude's potential and pitfalls in autonomous business management.
Context Engineering for Agents - Lance Martin explains what goes into an LLM's limited context window for better agent performance.
This AI Agent Should Have Been a SQL Query - Gunnar Morling contrasts pull-based SQL with push-based streaming queries.
37 Things I Learned About Information Retrieval in Two Years at a Vector Database Company - Leonie Monigatti reveals key vector database lessons from industry experience.π
Developing with GitHub Copilot Agent Mode and MCP - Austen Stone reveals VS Code settings for customizing AI assistants to enhance coding efficiency.
How Long Contexts Fail - Drew Breunig: Bigger AI context windows don't always mean better results.
The New Skill in AI is Not Prompting, It's Context Engineering - Philipp Schmid reveals why context quality trumps prompting in modern AI systems.
Context Engineering - What it is, and techniques to consider - Tuana Γelik and Logan Markewich guide to designing AI systems with optimal contextual awareness.
Hussein Mozannar - Hussein Mozannar shares a web agent development tutorial.
Building good agents - Smolagents shares essential practices for creating functional AI agents.
ποΈ YouTube & Podcasts
Claude Code: Anthropic's CLI Agent - YouTube - Video explores Anthropic's new CLI coding tool amid intensifying AI coding competition.
AI Product Management - YouTube - Aakash Gupta presents a comprehensive crash course covering essential AI PM skills from prompting to AI agents, designed to equip viewers with the knowledge needed to become effective AI product managers.
MCP vs ACP vs A2A: Comparing Agent Protocols - YouTube - Laurie Voss compares major agent protocols for AI resource integration.π
RAG vs. CAG - YouTube - This video explores the differences between Retrieval-Augmented Generation (RAG) and Context-Augmented Generation (CAG).
Software Is Changing (Again) - YouTube - Andrej Karpathy reveals how AI is revolutionizing software development practices. ππ π
12-Factor Agents - YouTube - Dex Horthy reveals best practices for building dependable AI agent systems. ππ π
ποΈ News and Updates
X - next level openai codex abuse
Build AI agents with the Mistral Agents API - Mistral introduces new Agents API.
X - an easy way to extract ChatGPT's memory of you and metadata
X - Min Choi's new development flow
Vibes, Rules, it's getting messy! - Dion Almaer examines the evolving tension between intuitive "vibe-based" development approaches and traditional rule-based methodologies in AI-native development environments.
Doubling Down on Open Source - Langfuse announces: All features now MIT-licensed for free self-hosting.
X - did you know ChatGPT gave you free Windows 10 Pro keys!
π οΈ Tech & Tools
GitHub - x1xhlol/system-prompts-and-models-of-ai-tools: FULL v0, Cursor, Manus, Lovable... - extensive AI tool prompts and models for developer reference. πππ
Repo Prompt - A frictionless app for AI-powered code iteration.
GitHub - mindsdb/mindsdb - A unified AI query engine for connecting and extracting insights across scattered data sources.
StreamNative Agent Engine: Event-Driven Runtime for Real-Time AI Agents
Claude Code β HiddenβGems - power features for maximizing Claude's capabilities.
GitHub - going-doer/Paper2Code: Paper2Code: Automating Code Generation from Scientific Papers
Zed β The editor for what's next - A Rust-built code editor integrating AI for collaborative development.
GitHub - BoundaryML/baml - A language for reliable AI prompts with multi-language support.
GitHub - github/github-mcp-server: GitHub's official MCP Server - for connecting GitHub APIs with developer tools.
GitHub - GoogleCloudPlatform/cloud-run-mcp - MCP server to deploy apps to Cloud Run
sketch.dev - A containerized coding assistant for parallel development workflows.
GitHub - sst/opencode - a terminal-based AI coding assistant that helps developers integrate AI capabilities directly into their command line workflow.
codename goose - a customizable local AI assistant for developers that automates complex tasks.
Cursor β Background Agents - Cursor introduced asynchronous code agents that work independently and push to GitHub.
KaibanJS - Kanban-inspired JavaScript framework for orchestrating AI agent systems.
GitHub - dagger/container-use - Containerized environments for multiple AI coding agents to work independently.
Kimi-Dev-72B - open-source coding model trained to fix real code issues.
Terminal-Bench - Netlify presents a benchmark for measuring AI agents' terminal capabilities.
Hooks - Anthropic - Did you know, Anthropic has customizable shell commands for extending Claude Code with automated behaviors.
Introducing Qodo Gen CLI - New CLI tool for building and running AI agents anywhere, and transform any IDE into an agentic environment.
Dosu | You build. Dosu documents. - A tool that converts codebases into living documentation for all team members.
π Learning Picks
AI Explorables | PAIR - PAIR team breaks down complex ML concepts through interactive essays.
Building software on top of LLMs - Simon Willison demonstrates LLM integration techniques for developers at PyCon 2025.
GitHub - langchain-ai/rag-from-scratch - Langchain-ai demonstrates how to build RAG systems from scratch to expand LLM capabilities.
π AI Humor
π source
π See You Next Week
Thanks for reading Generative Programmer! π
If you like the content please support my newsletter by subscribing and sharing with your network: X Β· LinkedIn Β· BlueSky Β· Email. Have a great rest of the week!
The repeated context-engineering links are the most useful thread in this issue for me. In an animation workflow such as https://spritefuse.com/, an initial character-and-action prompt and a later request to refine the motion have different jobs. A follow-up should say what to preserve as well as what to change. It would be interesting to see a future roundup compare those revision workflows, including how much context the creator has to restate at each step.