Marc Nuri

Marc Nuri

@marcnuri.com

Senior Principal Software Engineer at Red Hat. I build and secure the open source that developers and AI agents depend on. Creator of the Kubernetes MCP Server, lead maintainer of Fabric8 Kubernetes Client and Eclipse JKube. Valencia, Spain.

201 followers154 following80 posts

Every framework ranks AI coding tools by how autonomous they are. The axis nobody names is what each is built around: project, agent, or you. At professional grade, the person-centric corner is still empty. blog.marcnuri.com/project-agen...

blog.marcnuri.comProject, Agent, Person: The Missing Axis of AI Coding Tools - Marc NuriEveryone ranks AI coding tools by how autonomous they are. The axis that predicts the gaps is what they're built around: the project, the agent, or you.

This Thursday in Valencia 🦇 AI coding agents don't make every codebase faster. They amplify the one you already have. I'll show what makes a Java project actually AI-ready, live at Valencia JUG. 🗓️ Thu 18 Jun, 18:30 📍 Xplore Group Fresh off DevTalks Romania. RSVP link below 👇

What AI coding agents are missing isn't capability. It's discipline. Superpowers ships TDD, debugging, planning, and verification as a folder of markdown files. Same folder runs on six AI agent runtimes. blog.marcnuri.com/superpowers-...

blog.marcnuri.comSuperpowers: The Claude Code Skills Framework Shipped as Markdown - Marc NuriSuperpowers is the Claude Code skills framework that ships TDD, planning, debugging, and verification as markdown across six AI coding agents.

Steve Yegge's 8 levels of AI-assisted development have a very large missing rung between "juggling 10 agents by hand" and "build your own orchestrator". I wrote up the five rungs I think are there, from my own climb so far. blog.marcnuri.com/missing-leve...

blog.marcnuri.comThe Missing Levels of AI-Assisted Development: From Agent Chaos to Orchestration - Marc NuriSteve Yegge's 8 levels of AI-assisted development skip the hardest climb: the five missing rungs between juggling coding agents and your own orchestrator.

I've been running multiple AI coding agents in parallel across different machines. The problem: it's chaos. I'd forget about sessions and recover them days later. So I built a dashboard to fix it. Quick demo 🧵👇

View full thread

I added more features! bsky.app/profile/marc...

Marc Nuri@marcnuri.com

Last week I shared a dashboard for managing multiple AI coding agents. This week it got the upgrade I've wanted the most: remote session spawning. I can now launch new Claude Code sessions from any browser, even my phone. 🧵👇

Last week I shared a dashboard for managing multiple AI coding agents. This week it got the upgrade I've wanted the most: remote session spawning. I can now launch new Claude Code sessions from any browser, even my phone. 🧵👇

View full thread

Last week I shared a dashboard for managing multiple AI coding agents. This week it got the upgrade I've wanted the most: remote session spawning. I can now launch new Claude Code sessions from any browser, even my phone. 🧵👇

Here's the workflow: • Tap "spawn" on any device group • Pick a project • A terminal opens on that machine with a fresh Claude Code session, ready to work I can trigger this while commuting and review the results when I sit down.

Two more upgrades this week: • PR-aware sessions — each session card now shows the linked pull request • Smoother terminal scrolling — rebuilt the scroll pipeline for trackpads and touch Small details, but they add up fast when you're orchestrating multiple agents.

I've been running multiple AI coding agents in parallel across different machines. The problem: it's chaos. I'd forget about sessions and recover them days later. So I built a dashboard to fix it. Quick demo 🧵👇

View full thread

The killer feature: remote terminal access. I can connect to any session from my browser and interact with it—even if I'm away from the machine running the agent. WebSocket-based, works anywhere.

I've been running multiple AI coding agents in parallel across different machines. The problem: it's chaos. I'd forget about sessions and recover them days later. So I built a dashboard to fix it. Quick demo 🧵👇

The dashboard shows all my Claude Code sessions in real-time: • Which device and project • Current status (working, idle, needs my input) • Context usage and tokens • Git branch and model info No more guessing what's happening where.

I went all-in on AI-assisted development this year. Not autocomplete. Not faster typing. Asynchronous AI agents working in parallel. The result: 2× daily GitHub contributions and a fundamental shift from coding to orchestration. blog.marcnuri.com/boosting-dev...

blog.marcnuri.comBoosting My Developer Productivity with AI in 2025 - Marc NuriHow I restructured my software development workflow with asynchronous AI coding agents to dramatically boost developer productivity in 2025.

As AI coding tools become more common, I've been thinking about testing strategies. Blackbox (focusing on behavior) vs whitebox testing (focusing on implementation) hits differently when working with AI assistants that suggest optimizations regularly. blog.marcnuri.com/blackbox-whi...

blog.marcnuri.comBlack Box vs White Box Testing: When to Use Each Approach - Marc NuriDiscover key differences between blackbox and whitebox testing, and why blackbox should be your default for robust, maintainable test suites.

🚀 Dynamic Tool Selection is coming to ai-cli! Too many tools overwhelm LLMs, especially smaller ones with tiny context windows. In ai-cli the LLM can enable MCP servers automatically based on the user's prompt 🔥 🎥 www.youtube.com/watch?v=892d... #AI #MCP #LLM

youtube.comSolving the too many tools problem in ai-cli with Dynamic Tool SelectionIn this demo, I’ll show you how we're solving the too many tools problem in ai-cli: https://github.com/manusa/ai-cli When you provide too many tools to an LLM, it becomes confused and is not able to…

Next Wednesday at #DevBcn I’m distilling months of work into 40 minutes: Model Context Protocol Servers 101: Unlocking the Power of AI 🧠 What is MCP 🔧 How to build a server 💡 Real-world lessons from Kubernetes + Podman MCP servers 🕥 10:25am — Not recorded!

Just updated my MCP intro post to reflect the latest version of the protocol. MCP is moving fast, and I’m keeping the article fresh to help folks stay current. Great starting point if you’re curious about how AI agents gain context. 🔗 blog.marcnuri.com/model-contex... #MCP #AI #LLM #OpenSource

blog.marcnuri.comIntroduction to the Model Context Protocol (MCP): The Future of AI Integration - Marc NuriDiscover the Model Context Protocol (MCP), an open standard revolutionizing how AI systems interact with external data and tools.

🧪 We ran an experiment using SonarQube’s AI CodeFix on Eclipse JKube. ✅ Some issues were fixed impressively well ⚠️ Others showed that dev oversight is still essential 🤖 Overall? It’s a promising AI assistant for Java devs! Check out article from @vilojona.bsky.social : 🔗 dev.to/jonathanvila...

dev.toDiving Deep into SonarQube's AI CodeFix: A Hands-On Experiment with Eclipse JKubeSonarQube has long been a staple in a Java developer's toolkit for maintaining code quality. With the...

🚀 Updated blog: Connect your Java apps to an MCP server using LangChain4j, now compatible with the latest LangChain4j 1.0.0-beta5 & 1.0.0! Covers STDIO, SSE, testing, & full assistant setup 🧠 👉 buff.ly/98F5MG3 #Java #LangChain4j #AI #MCP #LLM #AIagents

buff.lyConnecting to a Model Context Protocol (MCP) Server from Java using LangChain4j - Marc NuriLearn how to integrate Java applications with an MCP server using LangChain4j. The article covers STDIO & SSE transports along with tips for testing servers.

🚀 Just published a new demo showing how to get function calling working in smaller language models like IBM Granite 3.2 by combining it with a solid MCP (Model Context Protocol) server and prompt design. buff.ly/FvahjtA #AI #LLM #MCP #Kubernetes #Granite #OpenSource

youtu.beSmaller LLMs, Bigger Power: IBM Granite 3.2 Meets Kubernetes MCP ServerIn this video, I demonstrate how to enable function calling in Small Language Models (SLMs) by combining a well-designed Model Context Protocol (MCP) server ...

🚀 Want to integrate JavaScript apps with MCP servers? Now you can with LangChain.js! 🔗 Learn how to connect MCP tools via STDIO & SSE 🤖 Build AI-powered agents with LangGraph.js Check out my latest post 👇 buff.ly/DUPBzSe #AI #JavaScript #LangChain #MCP #LLM

blog.marcnuri.comConnecting to an MCP Server from JavaScript using LangChain.js - Marc NuriLearn how to integrate JavaScript applications with a Model Context Protocol (MCP) server using LangChain.js and LangGraph.js

Vibe coding a game & deploying on Kubernetes with my MCP servers 🚀 Using VSCode + GitHub Copilot Chat, Claude Sonnet, Podman Desktop & RedHat Developer Sandbox to streamline the process. Watch the video and level up your vibes!

youtube.comVibe Coding: Build & Deploy a Game on Kubernetes with MCP Servers 🚀Join me as I vibe code a simple game and deploy it to OpenShift (Kubernetes), making it accessible for everyone to play!In this demo, I showcase the power of...