🤖 Generative Programmer | Issue #16
Autumn 2025: Patterns, Frameworks, and Productivity in AI Development
📣 Welcome back! Here’s everything that happened in the last two months in agentic systems and AI-assisted coding, all in one post. Support by sharing: X · LinkedIn · BlueSky · Email and let’s dive in.
🔥 Top Picks from This Issue
🧠 Writing Effective Tools for AI Agents – Anthropic 🔥
🧩 Effective Context Engineering for AI Agents – Anthropic again
🏗️ Building Production-Ready Agentic Systems – Shopify’s lessons
🧬 Deep Agents: From Shallow to Deep... – Elvis Saravia
💻 How to Vibe Code as a Senior Engineer – Alex Maccaw
🚀 The AI-Native Software Engineer – Addy Osmani
🧠 How I’m Using Coding Agents in September 2025 – Jesse Vincent
🧭 Context Engineering Needs Domain Understanding – Rod Johnson
⚡ From GPT-2 to gpt-oss: Architectural Advances – Sebastian Raschka 🔥
💡 You Should Be Rewriting Your Prompts – Max Leiter
🧱 Google’s MCP Toolbox for Databases Is Now Open Source
🔥🔥🔥 Prompting 101 - Hannah & Christian from Anthropic
…and if you dare, 100+ updates to explore 👇
🎯 Featured Article
Writing effective tools for AI agents - Anthropic outlines techniques for creating effective tools for LLM agents using the Model Context Protocol, sharing key principles for writing high-quality tools that bridge deterministic and non-deterministic systems.
📰 Building Agentic Systems
🔥🔥🔥 Effective context engineering for AI agents - Anthropic explains how context engineering is replacing prompt engineering, focusing on optimizing the entire token context to achieve consistent LLM performance rather than just finding the right prompt phrases.
Agentic AI Architecture Framework for Enterprises - InfoQ - A framework for building autonomous AI systems with architectural components and implementation patterns.
🔥 AI Agent Orchestration Patterns - Guide to designing multi-agent systems that handle complex tasks collaboratively by Microsoft.
🔥 Building production-ready agentic systems - Shopify’s journey building a production-ready AI assistant with practical lessons learned.
Equipping agents for the real world with Agent Skills - Anthropic: AgentSkills packages domain expertise for more capable AI assistants.
Magentic-One: A Generalist Multi-Agent System for Solving Complex Tasks - Microsoft’s multi-agent system that solves complex tasks through collaborative AI agents.
Supporting our AI overlords: Redesigning data systems to be Agent-first - Berkeley researchers propose redesigning databases for LLM agents’ unique query patterns.
Managing context on the Claude Developer Platform - Anthropic introduces new context management capabilities that help developers build AI agents capable of handling complex, long-running tasks without hitting context limits or losing critical information.
Agent Design Pattern Catalogue: A Collection of Architectural Patterns - Yue Liu - A collection of design patterns for building foundation model agents.
The 5 levels of AI agent autonomy: learning from self-driving cars - Maps coding agents to self-driving car autonomy levels, from manual to fully autonomous.
🔥 Deep Agents - AI systems evolving from shallow to deep agents with strategic planning capabilities.
📰 Coding with AI
🔥🔥🔥 How to Vibe Code as a Senior Engineer - Alex Maccaw explains how “vibe coding” leverages AI models to dramatically accelerate software development, turning traditionally expensive and time-consuming coding processes into rapid, efficient workflows for senior engineers.
Read That F*cking Code! - Developers need to actually read the code that AI generates for them.
🔥 The AI-Native Software Engineer -
: Engineers who embrace AI as partners can multiply their productivity.Onboarding for coding agents - Simplified context delivery for AI coding assistants.
Claude’s new Code Interpreter -
reviews Claude’s Code Interpreter and showcases various AI development tools.AI Coding Assistants for Large Codebases: A Complete Guide - Molisha Shah explains how effective AI coding assistants for large codebases need to understand entire repositories and architectural patterns, not just provide simple autocomplete functionality.
explains how to maximize the potential of coding agents like Claude Code and Codex CLI by designing effective agentic loops that enable these AI tools to iterate toward solutions.Embracing the parallel coding agent lifestyle - Simon Willison: Developer’s shift toward using multiple AI coding assistants at once.
🔥 How I’m using coding agents in September, 2025 - Developer’s workflow using Claude Code with git worktrees for parallel project management.
Vibe coding is not the same as AI-Assisted engineering. -
clarifies the critical distinction between casual “vibe coding” and professional “AI-assisted engineering,” emphasizing that conflating these approaches risks devaluing engineering discipline and misleading newcomers about production software requirements. explores programming’s evolution from hardware to AI-assisted environments.Reverse Engineering your Software Architecture with Claude Code to Help Claude Code - Nick Tune: Claude Code works best when it understands your system’s full functionality.
Beyond vibe coding: How AI can transform pull requests - Thoughtworks: Enterprise coding needs more than AI prompts due to regulatory and legacy constraints.
Exploring Cursor, Windsurf and Copilot with GPT-5 - Comparative analysis of GPT-5 coding tools across different development scenarios.
The Rule Maker Pattern: Creating Deterministic Execution with AI Probabilistic Generation - Guy Podjarny explains how developers can establish clear rules and guidelines to transform AI’s unpredictable outputs into reliable, deterministic execution for more consistent results.
How tech companies measure the impact of AI on software development - Orosz: Companies tracking whether AI coding tools justify their increasing costs.
How GPT5 + Codex took over Agentic Coding — ft. Greg Brockman, OpenAI - OpenAI’s GPT-5-Codex disrupts Anthropic’s coding dominance.
📰 Long-Form Reads
🔥🔥🔥 Context Engineering Needs Domain Understanding - Rod Johnson: Context engineering replaces prompt engineering with more precision.
🔥 You should be rewriting your prompts - Max Leiter: Different LLMs require different prompting approaches for best results.
Comparison - LLMs for Creating Software Architecture Diagrams - Testing if AI can create architecture diagrams but not replace architects.
debunks fears that AI harms cognition, suggesting productive AI use instead.GPT-5: Key characteristics, pricing and model card -
: Brief notes on OpenAI’s new models and ChatGPT features.The art and science of context engineering - CodeRabbit examines how effective AI code review tools require sophisticated context engineering to understand specific codebases deeply rather than simply pattern-matching against general rules.
🔥🔥🔥 From GPT-2 to gpt-oss: Analyzing the Architectural Advances - Technical breakdown of OpenAI’s new open-weight models and their architectural improvements.
Superintelligence - Meta’s roadmap for developing safe, advanced AI systems beyond human intelligence.
10 Proven Ways to Get Your Brand Mentioned in AI Answers - Strategies for brand visibility in AI responses through quality content placement.
How far can we push AI autonomy in code generation? - Böckeler: AI can build simple apps but still requires human supervision.
Mastering MCP Server with Spring Boot & AI - Standardized protocol for AI model-application communication.
Prompt injections as far as the eye can see - Researcher documents daily AI prompt injection vulnerabilities across multiple tools.
APIs don’t make good MCP tools - Reilly Wood explains why automatically converting existing APIs into Model Context Protocol tools often works poorly due to context window limitations and the fact that web APIs weren’t designed with these constraints in mind.
Agentic AI has changed my career. I don’t write code… - Elliot Graebert: Manager who stopped coding after early career promotion.
Less is More: Optimizing Function Calling for LLM Execution on Edge Devices - Streamlining LLM function calls for better edge device performance.
Building AI Products In The Probabilistic Era - AI systems work in ways we can’t fully predict or understand.
Your next job will require AI skills -
: AI fluency becoming essential across all job functions.Master the Blueprint: LLM Prompts for Perfect PRDs - Alward shares template for creating LLM-generated PRDs optimized for coding agents.
How AI is Transforming the PRD Process - Sean Shoffstall reflects on how AI tools are revolutionizing product requirements documents, addressing the longstanding problems of traditional PRDs that tend to be either too vague or overly detailed.
Writing product requirements with AI - Kapadia examines why PMs struggle with requirements documentation and AI limitations.
Using LLMs in Software Requirements Specifications: An Empirical Evaluation - Study comparing LLMs to humans in creating software requirement documents.
Vibe Specs: Vibe Coding That Actually Works - Luke Bechtel advocates for making AI write requirements before code, arguing this 5-minute investment saves hours of confusion and improves productivity.
LLM Evaluation: Practical Tips at Booking.com - George Chouliaras shares Booking.com’s methods for evaluating generative AI applications.
To vibe or not to vibe - Birgitta Böckeler examines the nuanced debate around AI-generated code review, challenging binary perspectives and suggesting that the appropriate level of review depends on various contextual factors.
LLMs as Parts of Systems - Marc’s Blog - Systems built with LLMs matter more than LLMs’ theoretical capabilities alone.
Attention Is the New Big-O - Alex Chesser explains how LLMs process text differently than humans, emphasizing that structural choices in prompts can impact results more significantly than word choice.
Let the Model Write the Prompt - Drew Breunig demonstrates how DSPy can be used to define and optimize LLM tasks, using a geospatial conflation problem as an example to show how the framework simplifies, improves, and future-proofs AI implementations.
First Malicious MCP in the Wild: The Postmark Backdoor That’s Stealing Your Emails - Koi reveals details about the first discovered malicious Model Composition Protocol (MCP) backdoor in the Postmark npm package that steals users’ email data.
Context engineering is sleeping on the humble hyperlink - the overlooked solution for efficient LLM context management.
Agentic Lessons learn after 300 agents - Sai Yashwanth - I have built around 300 agents, worked at 5 startups. Here’s what I learnt about AI Agent
Leadership co-processing with LLMs - James Stanier examines how large language models are transforming management practices through techniques for prioritization, communication, and cognitive offloading.
🎙️ YouTube & Podcasts
🔥How To Get The Most Out Of Vibe Coding - Guide to using AI as a coding partner to build faster by Y Combinator.
🔥🔥🔥 Prompting 101 | Code w/ Claude - Hannah Moran and Christian Ryan from Anthropic present fundamental prompting techniques for effectively working with Claude, covering essential strategies developers can use to improve AI interactions.
On Engineering AI Systems that Endure The Bitter Lesson - Building AI systems that last through principled engineering approaches.
🗞️ News and Social
Introducing gpt-oss - OpenAI releases Apache-licensed reasoning models optimized for consumer hardware.
Building agents with the Claude Agent SDK - Anthropic rebrands Claude Code SDK to reflect its broader agent capabilities beyond coding.
Google Labs introduces Opal - a new way to help you build and share AI mini-apps by linking together prompts, models, and tools— all while using simple, natural language
10 Principles of Building AI Agents: - by Pawel Huryn.
OpenAI experiments with new “Study together” tool on ChatGPT - ChatGPT’s new feature creates interactive learning experiences with guided study sessions.
THE WAY OF CODE - a project by
n in collaboration with Anthropic:Tessl launches spec-driven development tools for reliable AI coding agents
8 RAG architectures for AI engineers
Google’s MCP Toolbox for Databases is now open source! 🔥
AI tooling must be disclosed for contributions - AI disclosure requirement for contributions to ghostty-org.
🛠️ Tech & Tools
Agents.md: an open standard for AI coding agents - A file standard for consistent AI assistant instructions across projects.
GitHub - openai/gpt-oss: gpt-oss-120b and gpt-oss-20b - Open-weight models for reasoning and agentic tasks.
OpenCode | The AI coding agent built for the terminal - Open-source AI coding assistant for terminal that respects privacy.
Agent Leaderboard - a Hugging Face Space by galileo-ai - Comparative tool for AI agent performance with filterable metrics.
GitHub - traceloop/openllmetry - traceloop - OpenTelemetry-based observability for LLM applications.
GitHub - terryso/claude-auto-resume - Script that auto-resumes Claude CLI tasks after usage limits expire.
GitHub - f/awesome-chatgpt-prompts - A curated collection of prompts designed for ChatGPT and other LLM tools to help users get better responses from AI language models.
Kilo Code - coding agent for VS Code and JetBrains - AI coding assistant with project management capabilities serving 500k+ users.
GitHub - marcelsud/spec-driven-agentic-development - Specification-driven development framework for AI-assisted software building.
GitHub - charmbracelet/crush: 💘 - Terminal-based AI coding assistant for your workflow.
GitHub - Pimzino/claude-code-spec-workflow - Structured AI-powered development workflows for features and bug fixes.
GitHub - docker/cagent: Agent Builder and Runtime by Docker - AI agent orchestration tool for building collaborative virtual expert teams.
GitHub - kirodotdev/Kiro: Kiro is an agentic IDE that works alongside you - AI-powered IDE that understands your codebase and automates development tasks.
Product requirement document generation using LLM task oriented dialogue · GitHub - LLM-powered PRD generation through guided dialogue.
GitHub - davidkimai/Context-Engineering - First-principles handbook for AI context design and optimization.
GitHub - ashishpatel26/500-AI-Agents-Projects - Curated collection of AI agent applications with implementation resources.
GitHub - GoogleCloudPlatform/kubectl-ai: AI powered Kubernetes Assistant - AI-powered tool that simplifies Kubernetes management.
GitHub - hesreallyhim/awesome-claude-code - A curated resource list for maximizing Claude Code’s capabilities.
AI Coding Tools: Install Trends (based on Visual Studio Code) - Visual tracking of AI coding extension popularity in VS Code.
GitHub - invariantlabs-ai/mcp-scan: Constrain, log and scan your MCP connections - Security scanner for MCP servers detecting vulnerabilities across various configurations.
📚 Learning Picks
Think Python - Free, beginner-friendly Python programming book with interactive Jupyter notebooks.
GitHub - Asabeneh/30-Days-Of-Python: 30 days of Python programming challenge - Self-paced Python learning challenge with daily exercises.
Generative AI for Beginners - Microsoft’s 18-lesson course on building Generative AI applications.
GitHub - microsoft/ai-agents-for-beginners: 12 Lessons to Get Started Building AI Agents - Beginner-friendly AI agent development course .
🎭 AI Humor
👋 See You Next Time
Thanks for reading Generative Programmer! 🙏 If you like the content please support my newsletter by subscribing and sharing with your network: X · LinkedIn · BlueSky · Email. Have a great rest of the week!