🤖 Generative Programmer | Issue #17
Curated insights, tools, and techniques for AI-assisted development and building agentic systems
📣 Welcome back! Here’s everything I came across in November on agentic systems and AI-assisted coding, all in one post. Support me by sharing on your networks, and let’s dive in.
🔥 Top Picks from Issue #17
🔐 How We Hijacked a Claude Skill with an Invisible Sentence - Josh Devon
🧠 Why your AI agents need domain awareness - Russ Miles
🔧 How I Use Every Claude Code Feature - Shrivu Shankar
📊 GenAI in Production: MLOps or GenAIOps? - Dr. Kartakis
👌 The Context Engineering Guide - Weaviate
🔍 Building a Deep Research Agent from scratch - Aurimas Griciūnas
🛡️ Design Patterns for Securing LLM Agents against Prompt Injections - Luca Beurer
📈 I Went All-In on AI. The MIT Study Is Right. - Josh Anderson
🎯 Featured Article
This in-depth tutorial by Aurimas Griciūnas walks you through building a Deep Research Agent from scratch using the open-source DeepSeek R1 model.
It covers every step, from planning and web search to summarization and reflection, without relying on any orchestration frameworks. Highly recommended to check it out!
🤖 Building Agentic Systems
How We Hijacked a Claude Skill with an Invisible Sentence - Josh Devon reveals a critical security vulnerability in Claude Skills that allows attackers to override instructions with invisible text!
Agent IAM in Cloud Native World with SPIFFE and Istio - Mr Posta’s slides on cloud-native identity management with SPIFFE. 🔥
Domain Driven Agent Design - by Russ Miles - Miles explains why AI agents fail when deployed without domain context, creating impressive demos that collapse in production environments.
New prompt injection papers: Agents Rule of Two and The Attacker Moves Second -
highlights new research on prompt injection vulnerabilities along with WiFi-enabled conference badge hacking and other tech links.Design Patterns for Securing LLM Agents against Prompt Injections - Luca Beurer’s paper on security framework for LLM agent protection.
How Many Tools Functions Can An AI Agent Have - to do
💫 The Context Engineering Guide - Learn to optimize LLM performance through context engineering patterns that overcome hallucinations and build reliable AI applications with real-world context.
Why Model Context Protocol and Agentic Workflows Are Shaking Up How We Build Apps - Monica Hockelberg explains how MCP and agentic workflows transform rigid development into more fluid, adaptable systems for the AI era.
Bulletproof agents with the durable task extension for Microsoft Agent Framework - New extension enables resilient agent workflows that can survive interruptions and automatically resume processing.
Agent OS Workflow for Spec-Driven Development - Brian Casel proposes a six-phase AI agent workflow for spec-driven development.
The complete guide to LLM observability for 2026 - Ghost explains how to implement comprehensive LLM monitoring systems with frameworks for tracing, guardrails, cost management, and governance.
Context Engineering in Manus - Erik explores how Manus handles context management for AI agents, addressing challenges of context windows filling up during extended tool-calling sessions.
Building a Toolformer model with OpenAI and Remix - Josh Sanger shows how to build a weather assistant using Toolformer, enabling AI to use external tools via API calls.
Making Agent-to-Agent (A2A) Secure and Reliable with Dapr - I tried A2A with Dapr and saw how easily it adds mTLS, policy enforcement, tracing, and self-healing behavior to any agent workflow. Then showed step-by-step how to turn a deterministic workflow into an agentic one.
🚀 Choose a design pattern for your agentic AI system - Google’s guidance for selecting appropriate agent design patterns to build AI systems that can handle autonomous decision-making and complex workflows.
MCP Apps: Extending servers with interactive user interfaces - Ido Salomon introduces a new extension enabling MCP servers to deliver interactive UIs to hosts, addressing a top community request.
💻 Coding with AI
How I Use Every Claude Code Feature - Shrivu Shankar shares detailed insights on leveraging Claude Code for both hobbyist projects and professional development with high-volume token usage.
Claude Code Can Debug Low-level Cryptography - Filippo Valsorda describes how Claude Code successfully identified a complex bug in his post-quantum signature algorithm implementation when he was stuck.
The evolution of code review practices in the world of AI - Alex Pliutau examines how AI tools are transforming code reviews, improving efficiency while maintaining human oversight.
🔥 GitHub - github/awesome-copilot - github - Curated toolkit to supercharge Copilot.
🔥 GitHub - hesreallyhim/awesome-claude-code - Community-curated collection of commands, files, and workflows to enhance productivity with Anthropic’s CLI-based coding assistant.
How I use AI (Oct 2025) - Stolovitz reflects on his AI usage as a software engineer, highlighting Copilot’s impact on coding productivity since its early days.
Rulebook-AI wants to “elevate vibe coding to vibe engineering” - Tool transforms AI coding assistant configurations into structured guidance, eliminating fragmentation and inconsistency in development workflows.
📰 Long-Form Reads
A Survey of Context Engineering for Large Language Models - Lingrui Mei’s comprehensive overview of techniques to optimize LLM context usage, from retrieval methods to context compression.
The State of AI - McKinsey’s report on current AI adoption trends, business impact, and implementation challenges across industries.
🔬 I Went All-In on AI. The MIT Study Is Right. - Josh Anderson shares firsthand experience using Claude Code exclusively for three months to understand why 95% of corporate AI initiatives fail.
GenAI in Production: MLOps or GenAIOps? - Dr. Kartakis analyzes various operational frameworks for deploying GenAI solutions, clarifying differences between MLOps, LLMOps, and other emerging approaches.
What would you do if you had a chance to sneak peak in the year 2030 and you were a Hubber? - A developer’s 2025 forecast of AI’s rapid takeover of coding jobs, driven by economics and efficiency gains across most software domains.
The Death of Software Engineering as a Profession - Jason Scheirer reflects on incorrect predictions about programming careers and how specialized knowledge remains valuable despite technological advances.
🎨 Nano Banana can be prompt engineered for extremely nuanced AI image generation - Max Woolf explores how Nano Banana offers sophisticated image generation amid competition from FLUX.1-dev, Seedream, and ChatGPT’s viral features.
🎙️ YouTube & Podcasts
The Secrets of Claude Code From the Engineers Who Built It - Explains how Claude Code enables developers to ship features in unfamiliar codebases more efficiently, with each new feature making subsequent development easier.
Practical AI - Podcast making AI accessible through real-world applications.
🗞️ News and Updates
Akshay on X - Who is a Full-stack AI Engineer?
Paweł Huryn on X - 10 Principles of Building AI Agents:
Paweł Huryn on X - Context engineering is the new prompt engineering.
Aurimas on X - AI Engineering Learning Roadmap
TOON (Token-Oriented Object Notation) - Aurimas Griciūnas briefly explains how TOON increases model accuracy while reducing token count, critical for optimizing agentic systems.
🛠️ Tech & Tools
llmtext - Turn any llms.txt into a dedicated MCP server
GitHub - agentregistry-dev/agentregistry - Centralized registry for AI artifact governance.
AI Connectors Directory - Directory of Remote MCP servers for AI custom connectors.
DeepCode: Open Agentic Coding - Open-source tool offering terminal-based and web interfaces for AI-powered code generation from papers and text.
Prompt-Hacking-Resources - PromptLabs’ curated collection of resources for AI red teaming, jailbreaking, and prompt injection, designed for those interested in AI/ML security and safety.
OpenCode | The open source AI coding agent - A privacy-focused AI coding assistant offering free models with no account required and integrations across multiple platforms.
GitHub - toon-format/toon: 🎒 Token-Oriented Object Notation (TOON) - A compact JSON alternative optimized for LLMs that combines YAML’s indentation structure with CSV-like layouts for efficient token usage while maintaining readability.
GitHub - Portkey-AI/gateway - Open-source AI gateway providing fast routing to 1600+ language, vision, audio, and image models with integrated guardrails and enterprise-ready security.
GitHub - metatool-ai/metamcp - MCP proxy that aggregates multiple MCP servers into a unified endpoint with middleware support, enabling developers to build agents on this infrastructure.
You’re Burning 0.02 Per 1K JSON Tokens You Don’t Need To - Ankita Tripathi explains how to optimize JSON structures in API calls to reduce token usage and lower costs when working with AI models.
💰 Frugal AI - Intelligent Application Cost Engineering - AI-powered tool that optimizes application code across major cloud platforms by automatically identifying and fixing inefficiencies in code rather than just resizing infrastructure.
Technology Radar - Thoughtworks’ biannual guide tracking emerging tools, techniques, and frameworks is AI-dominated.
17 Top MCP Registries and Directories - Comprehensive guide to the best MCP registries for server discovery.
cto.new - Free AI code agent offering access to frontier models from Anthropic and OpenAI without requiring credit cards or API keys.
Claude Skills Marketplace - Directory for browsing, searching and installing Claude skills from GitHub, including official Anthropic skills, code skills, and AI automation workflows.
GitHub - anthropics/skills - Anthropics’ repository of reusable Claude skills that enable specialized tasks across creative, technical, and enterprise workflows with clear documentation.
GitHub - langwatch/better-agents - CLI tool and standards that enhance coding assistants to build production-ready agents with industry best practices.
GitHub - ruvnet/claude-flow - Enterprise-grade AI orchestration platform using swarm intelligence with persistent memory and MCP tools for enhanced development workflows.
Kagenti - Kubernetes-based control plane for AI agents that works with any framework, providing modular components to streamline production deployments.
GitHub - kubernetes-sigs/agent-sandbox - Kubernetes CRD for managing isolated, stateful singleton workloads with stable identity, designed for AI agent runtimes and similar use cases.
⚡Code Wiki - Google’s new Gemini-generated documentation.
📚 Learning Picks
🎓 5-Day AI Agents Intensive Course - A 5-day Google course on Kaggle to build, evaluate, and deploy real-world AI agents.
👍 Google, OpenAI, Anthropic Offer Free AI Guides - Here are ten free AI guides worth bookmarking.
🎭 AI Humor
👋 See You Next Time
That wraps this issue. If it helped you, pass it along or subscribe. 🙏
Outstanding curation here! The shift you're highlighting from prompt engineering to context enginering really captures where the field is headed. When agents fail in production despite working in demos, it's almost always becuase they lack the domain-specific context that constrains their decision space. TOON as a compact alternative to JSON is particularly clever since it directly addresses the token bloat problem that kills agentic workflows at scale.