Developers spend more and more time building an AI harness instead of writing code themselves. But what is a good harness, according to the community? I benchmarked 246 open-source reporisories and read 57 publications to find out. marmelab.com/blog/2026/09...

The State Of AI Harness Engineering 2026

We audited 246 open-source repositories and read 57 publications to better understand the trends of tooling for agentic coding.

marmelab.com

Profitez de cet aperçu de la conférence de @francoisz.bsky.social au Forum PHP 2026 dans l'interview du jour ! Il explique notamment que selon lui, l'IA ne va pas remplacer les devs, mais qu'elle va changer le type de projets sur lesquels on travaille. Venez l'écouter ! buff.ly/fBmeX24

StyleX makes CSS-in-JS interesting again. Typed styles in, static atomic CSS out. No runtime injection. Tokens, linting & predictable composition make inconsistency harder to ship. Sure, it's more verbose than Tailwind, but with coding agents, does it still matter? flaviocopes.com/stylex/

A deep dive into StyleX

Learn how StyleX turns JavaScript style objects into regular CSS classes at build time, with React and Astro setup, variants, themes, and responsive styles.

flaviocopes.com

AI produces too much code for engineers to review. Perhaps it's a good time to ask why we need code reviews in the first place. Code review isn’t just about finding bugs. It’s how teams share knowledge, teach junior engineers, build collective ownership and spread architectural understanding.

Do you really need a RAG? If you just want a natural language search over a large corpus of mostly cold documents, many alternatives will provide great results without the hassle associated with embeddings. www.lighthousenewsletter.com/p/rag-is-sim...

RAG Is Simpler Than You Think

Six approaches to retrieval-based AI, from minimal to elaborate

lighthousenewsletter.com

Quel est le meilleur agent de codage ? D'après leurs éditeurs, ils sont tous meilleurs que les autres... Alors j'ai fait mon propre benchmark : redévelopper la même application from scratch avec 3 agents différents. marmelab.com/blog/2026/07...

OpenAI Codex vs Claude Code vs Kimi Code : la battle

Quel assistant de codage choisir ? Le meilleur benchmark, c'est un même projet recodé depuis la spécification par 3 agents différents.

marmelab.com

TIL the /insights command on #ClaudeCode lets the agent inspect your session logs, detect when Claude went wrong, suggest fixes in your harness (CLAUDE.md, skills, hooks) and inform you of features you could have used to improve your workflow.

"But the bottleneck was never the code. It was always the human attention, the engineering judgment, the ability to hold a coherent vision for a system. We just didn't notice because writing code felt like the hard part."

Add a realtime vizualizer of the carbon impact of your Claude Code usage with ecologits-statusline, based on the EcoLogits LLM consumption model and tweaked with the data from the ML.ENERGY Leaderboard. #OSS #Sustainability github.com/DuarteVi/eco...

Handoff: verify you actually understood the code Claude wrote by answering a quiz. Code review was always as much about comprehension as correctness. github.com/arimet/handoff

GitHub - arimet/handoff: Verify you actually understood the code Claude wrote — before you ship it

Verify you actually understood the code Claude wrote — before you ship it - arimet/handoff

github.com

Software that builds software is here. I call them Agentic Software Factories. Unlike Retool or Excel, they understand natural language. As a result, any company can now get tailor-made tools for ~$10k. Custom software stops being a Fortune 500 privilege. marmelab.com/blog/2026/05...

Agentic Software Factories: The Future Of Programming?

What if anyone in a company could build the exact software they need, without writing a single line of code? Agentic software factories, powered by GenAI, are making this possible - and they're redefi

marmelab.com

Wait for it... github.com/marmelab/ato...

getagentseal/codeburn: See where your AI coding tokens go. Interactive TUI dashboard for Claude Code, Codex, and Cursor cost observability. github.com/getagentseal...

I keep seeing MCP servers get overlooked or misunderstood. I think that's a mistake: they're shaping up to be a must-have for any web service. Here's the pitch I've been making to non-tech clients. #AI #MCP marmelab.com/blog/2026/04...

MCP Servers: Your Next User Is An AI Agent

MCP servers are the best thing since sliced bread. They extend the capabilities of AI assistants and let them connect with third-party tools. If your service doesn't have one yet, it's time to priorit

marmelab.com

Steve Blank's latest piece is a wake-up call: if your company is more than two years old, the world has shifted under your feet, and most of your founding assumptions probably aren't true anymore. steveblank.com/2026/03/17/y...

Steve Blank Your Startup Is Probably Dead On Arrival

Your Startup Is Probably Dead On Arrival If you started a company more than two years ago, it’s likely that many of your assumptions are no longer true. You need to stop coding, building, recruitin…

steveblank.com

I finally managed to implement an MCP App, i.e. an interactive widget that renders inside a conversation in ChatGPT, Claude, or any other MCP-compliant agent. marmelab.com/atomic-crm/d...

How much of your project code was written by #AI? Whether to track coding agent adoption in engineering teams or security risks introduced by vibe coding, it's time to track what Claude/Cursor/Copilot/Codex do. usegitai.com

Git AI - Track AI Code all the way to production

Cross-agent observability from prompt to production. Track AI-generated code from Cursor, Claude Code, GitHub Copilot, Gemini, and more through the entire SDLC.

usegitai.com

The axios npm package, widely used for data fetching, has been compromised. Pin your version now. #NodeJS #Security socket.dev/blog/axios-n...

Supply Chain Attack on Axios Pulls Malicious Dependency from...

A supply chain attack on Axios introduced a malicious dependency, plain-crypto-js@4.2.1, published minutes earlier and absent from the project’s GitHu...

socket.dev

AI is getting better at coding every day. Time to diversify your skills. UX sense is hard to automate, easy to neglect. We built a free browser game to train it: spot the usability flaw in mini-apps, describe it, get scored by ML. #UX #OpenSource #AI marmelab.com/ux-quest/

UX Quest - Train Your Usability Skills

Spot the UX flaw! Examine mini-apps with deliberate usability problems, describe what's wrong, and get scored by an AI model.

marmelab.com

"Who needs devs now that agents can do the job" reminds me of "who needs banks now that we have the blockchain". Same confusion. It was never about the technology. It's about trust. marmelab.com/blog/2026/03...

Software Will Never Die

AI agents can now rewrite an app in hours. But software value lies in judgment, trust, and specification — not code. Here's why the craft of software engineering is evolving, not dying.

marmelab.com

We've just releaased Shadcn Admin Kit version 1.5 with tons of new features, improvements and bug fixes. #OpenSource #React github.com/marmelab/sha...

Release v1.5.0 · marmelab/shadcn-admin-kit

v1.5.0 (Feb. 2026) 🚀 Features Add <TextArrayInput> component for handling arrays of text values. Add <DateTimeInput> component for date and time input support. Add <ImageField> component for displ...

github.com

Generative UIs are becoming a reality: both ChatGPT and Claude now support MCP Apps, a new standard that allows tools to render tables, charts, and interactive UI widgets. #GenerativeUI #AI #MCPApps blog.modelcontextprotocol.io/posts/2026-0...

MCP Apps - Bringing UI Capabilities To MCP Clients

Today, we’re announcing that MCP Apps are now live as an official MCP extension. Tools can now return interactive UI components that render directly in the conversation: dashboards, forms, visualizati...

blog.modelcontextprotocol.io

Laravel admins often mean Nova. It works well. But it’s paid. There’s a free alternative. #Laravel + #APIPlatform + #ReactAdmin = a powerful admin UI on top of your Eloquent models. JSX instead of PHP, but more flexibility and UI power. marmelab.com/blog/2026/01...

A Free Laravel Admin With API Platform and React-Admin

Replace Laravel Nova with a flexible admin built on open standards and open-source tools.

marmelab.com

Why do coding agents work great in some projects but fail badly in others? In my experience, the difference is rarely the agent. It’s the codebase. Here are 40+ best practices to make coding agents faster, safer, and more productive on real-world codebases. marmelab.com/blog/2026/01...

Agent Experience Best Practices for Coding Agent Productivity

Why do coding agents shine in some projects and fail in others? This article explains how to design your codebase so agents work autonomously and effectively.

marmelab.com

🛠️ Vous développez un agent IA ? Le prototype n’est que le début. Optimiser le coût, la rapidité et la pertinence représente 90 % du temps. Découvrez comment les outils d’évaluation peuvent simplifier ce processus dans le dernier talk de @francoisz.bsky.social!👇 @afup.org youtu.be/NjrbQ8aWDv8?...

L'évaluation des IAs : la recette secrète des agents pas trop bêtes - François Zaninotto - Forum PHP

YouTube video by AFUP PHP

youtu.be

When using a backend-as-a-service like Supabase, you can only access the database through an HTTP API. Yet it is still possible to group queries into a transaction, without compromising on security. Read on to learn more. #database #security #NodeJS #supabase marmelab.com/blog/2025/12...

Transactions and RLS in Supabase Edge Functions

Edge functions are a powerful way to run server-side code close to your users. But how to handle transactions and Row Level Security (RLS) in these functions?

marmelab.com

Generative UI is about letting agentic apps create user interfaces on the fly. A new specification called MCP Apps emerges, backed by Anthropic & OpenAI, to standardize it.It's very early stage ; to get an idea of the potential, try the popular MCP UI. blog.modelcontextprotocol.io/posts/2025-1...

MCP Apps: Extending servers with interactive user interfaces

Today we’re introducing the proposal for the MCP Apps Extension (SEP-1865) to standardize support for interactive user interfaces in the Model Context Protocol. This extension addresses one of the…

blog.modelcontextprotocol.io

Have you come across Spec-Driven Development yet? One AI writes the specs, another writes the code. Sounds smart, but it's basically waterfall in disguise. Here’s why we should stay agile instead. marmelab.com/blog/2025/11...

Spec-Driven Development: The Waterfall Strikes Back

Kiro, Spec-kit, Bmad, Tessl, and other SDD frameworks turn business analysts into Markdown reviewers. Isn't there a more agile way to use Coding Agents?

marmelab.com

Etsy's Web Performance team implemented the Speculation Rules API (SRA) to prefetch product listing pages from search results, resulting in significant performance gains (20-24% improvement in metrics like TTFB, FCP, and LCP). #WPO www.etsy.com/codeascraft/...

etsy.com

"The ability to phrase our intent in natural language and receive working code does not replace the deeper understanding that comes from learning each language's design, constraints, and trade-offs." This analysis of #CodingAgents on #SoftwareDevelopment Is spot on. martinfowler.com/articles/llm...

The Learning Loop and LLMs

How continuous learning shapes effective software development with LLMs

martinfowler.com