Joaquin T.’s Post

Next.js 16.2 shipped, and the focus is clear: make AI agents first-class citizens in your dev workflow. Here are the 4 features worth knowing: 1. AGENTS.md in every new project create-next-app now scaffolds an AGENTS.md file by default. It tells AI coding agents to read version-matched Next.js docs from node_modules/next/dist/docs/ before touching any code. Why this matters: Vercel's own research showed agents given bundled docs hit a 100% pass rate on Next.js evals. Agents relying on retrieval-based approaches? 79% max. Always-available context wins over on-demand lookup because agents often don't know when they don't know. 2. Browser errors forwarded to your terminal Client-side errors now pipe directly to your terminal during development. No browser tab needed. For AI agents that live in the terminal, this is a significant unblock. Configurable in next.config.ts from errors-only to full console output. 3. Dev server lock file How many times has an agent spun up a second next dev while one was already running? Now Next.js writes a lock file with the PID, port, and URL. If a second process tries to start, you get an actionable error with the exact kill command. Simple, but genuinely solves a real agent loop problem. 4. Experimental Agent DevTools (@vercel/next-browser) This one is the most forward-looking. It's a CLI that exposes your running app's data as shell commands: React component trees, PPR shell analysis, network activity, screenshots, console logs. The use case: an agent runs next-browser ppr lock, sees the entire page is a loading skeleton, then runs next-browser ppr unlock to find that a single getVisitorCount() call at the top of a component is blocking the entire page from prerendering. It gets told exactly which file, which line, and what to do next. Zero browser required. This is what AI-native tooling actually looks like. Not just autocomplete. Agents that can observe, diagnose, and fix running applications. #NextJS #WebDevelopment #AIAgents #JavaScript #SoftwareEngineering

VibeKit: Stop Burning Tokens, Start Shipping Real Apps If you've ever watched Claude Code, Lovable, or V0 confidently generate a beautiful app that breaks the moment you touch authentication — VibeKit is built for you. Created by JB (Muke Johnbaptist) of Desishub Technologies, VibeKit is a structured framework for turning vibe-coded ideas into production-grade Next.js applications. Instead of prompting blindly and praying, you follow a repeatable system. How it works (in 3 moves): Plan — Paste the CLAUDE_PROMPT.md into Claude, describe your idea, and answer 6–10 clarifying questions. You get back three files: a project description, a phased build blueprint, and a ready-to-run Claude Code prompt. Standardize — Drop master_prompt.md into your repo so Claude Code writes to strict coding standards, not AI slop. Build phase by phase — Claude Code executes one phase at a time and pauses for your approval before continuing. The opinionated stack: Next.js 16, TypeScript, Prisma v7 on Neon Postgres, Better Auth, React Query, Zod, Tailwind v4 + shadcn/ui, Stripe, Resend, and Vercel — chosen because AI models handle these patterns reliably. It also ships reference guides for deployment, environment variables, database design, monetization, and troubleshooting — the boring-but-critical pieces most AI-generated projects skip. TL;DR: VibeKit is the planning layer that makes vibe coding actually ship. 🔗 https://lnkd.in/dmWDipq3

Building a Full-Stack Agentic Research Engine with .NET 9, React, and Ollama 🚀: I’ve officially kicked off Project 1—AI Research Engine that bridges the gap between local data and LLMs. As a Full-Stack Developer, my goal is to build a system that is not just smart, but also scalable and user-friendly. Today’s focus was the Backend Architecture. Here’s a look at my setup (see screenshot): 🔹 The .NET 9 Powerhouse: I'm using the .NET Dependency Injection (DI) container to manage my AI services. By registering SemanticKernelService and ChatService properly, I'm ensuring the application remains loosely coupled. Today it's Ollama; tomorrow I can switch to Azure OpenAI with zero friction. 🔹 The React Frontend: The UI is being built with React & Tailwind CSS. Why? Because handling real-time AI streaming responses and managing complex file upload states (PDFs/YouTube URLs) requires a robust frontend framework. 🔹 Privacy-First with Ollama: By running Llama 3.2 locally, I’m ensuring that no sensitive research data leaves the machine. Privacy is a feature, not an afterthought. Current Tech Stack: 💻 Frontend: React,Tailwind CSS, Vite ⚙️ Backend: .NET 9, Semantic Kernel 🧠 LLM: Ollama (Llama 3.2 & Nomic-Embed) 🗄️ Vector DB: Pinecone & SQLite Next up: Connecting the React frontend to my .NET Ingestion API to start processing real-world documents. The journey to becoming an AI-Orchestrator continues! 🛠️ #FullStack #ReactJS #DotNet #Ollama #SemanticKernel #AIAgents #WebDev #LocalAI #SoftwareArchitecture

Async queues are the difference between “it works locally” and “it survives production.” ⚙️🚦 In JavaScript, async doesn’t mean “parallel.” It means “scheduled.” Without a queue, bursts of work turn into API rate-limit storms, memory spikes, and unpredictable latency. A solid async queue / scheduler usually needs: • Concurrency limits (e.g., 5 tasks at a time) 🧵 • Backpressure (stop accepting work when the system is saturated) 🛑 • Retries with jitter + exponential backoff (avoid thundering herds) 🔁 • Timeouts + cancellation (AbortController) ⏱️✋ • Priorities + fairness (don’t starve low-priority tasks forever) 🎚️ • Observability: queue depth, wait time, in-flight, error rate 📈 Where it pays off: • Node.js workers pulling jobs from Redis/SQS/Kafka • Next.js/React apps batching expensive client calls to protect UX • AI pipelines: embedding generation, document parsing, model calls—rate limits are real 🤖 • Healthcare/HR/enterprise automation: “event storms” after imports or syncs need controlled throughput 🏥🏢 Practical rule: design for bursts, not averages. If you can bound concurrency and measure backlog, you can predict behavior under load. What’s your go-to pattern: in-memory queue, BullMQ, custom scheduler, or serverless orchestration? 👇 #javascript #nodejs #frontend #distributed-systems #ai

New frontend developers be like: "Why write logic when AI can code it for me?" 🧐 But here’s the truth nobody says out loud — AI can generate the component. It can’t give you the understanding behind it. When you skip the basics, you skip: ❌ Semantic HTML ❌ CSS fundamentals (Flexbox, Grid, specificity) ❌ JavaScript & DOM understanding ❌ Framework fundamentals (React/Vue/Angular) ❌ Component architecture Then you jump into AI tools or generators like Vercel v0… build something fast… and suddenly: 🚫 It doesn’t scale 🚫 Performance is poor 🚫 Debugging is painful 🚫 Accessibility is ignored ___________________________________________ AI is an accelerator — not a foundation. Copilot won’t make you an architect. AI tools won’t make you a developer. Your real edge? 👉 Problem-solving 👉 System thinking 👉 Strong fundamentals Because when AI gives everyone the same output… only you decide how good it becomes. ___________________________________________ Learn the fundamentals first. Then let AI multiply your speed. The stairs exist for a reason. 🧱 Where are you on the staircase right now? 👇 #FrontendDeveloper #WebDevelopment #JavaScript #DeveloperMindset #CleanCode #LearnInPublic #BuildInPublic

𝗙𝗮𝘀𝘁𝗔𝗣𝗜 – 𝗧𝗵𝗲 𝗠𝗼𝗱𝗲𝗿𝗻 𝗛𝗶𝗴𝗵-𝗣𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲 𝗔𝗣𝗜 𝗙𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸 In today’s world of scalable systems and real-time applications, choosing the right backend framework can make all the difference. That’s where 𝗙𝗮𝘀𝘁𝗔𝗣𝗜 stands out. Built with modern Python features and designed for speed, FastAPI enables developers to create robust APIs with minimal effort and maximum performance ⚡ 𝗪𝗵𝘆 𝗙𝗮𝘀𝘁𝗔𝗣𝗜? 𝗕𝗹𝗮𝘇𝗶𝗻𝗴 𝗙𝗮𝘀𝘁 𝗣𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲 Powered by ASGI and Starlette, FastAPI delivers performance comparable to Node.js and Go, making it ideal for high-throughput systems. 𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗿-𝗙𝗿𝗶𝗲𝗻𝗱𝗹𝘆 Automatic generation of interactive API documentation (Swagger UI & ReDoc) helps teams test and collaborate efficiently without extra setup. 𝗧𝘆𝗽𝗲 𝗦𝗮𝗳𝗲𝘁𝘆 & 𝗩𝗮𝗹𝗶𝗱𝗮𝘁𝗶𝗼𝗻 Leverages Python type hints and Pydantic for automatic request validation, serialization, and clear error handling—reducing bugs significantly. 𝗔𝘀𝘆𝗻𝗰-𝗙𝗶𝗿𝘀𝘁 𝗔𝗽𝗽𝗿𝗼𝗮𝗰𝗵 Native support for async/await allows handling thousands of concurrent requests, perfect for modern cloud-native apps. 𝗣𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝗼𝗻 𝗥𝗲𝗮𝗱𝘆 From dependency injection to security features, FastAPI provides everything needed to build scalable, maintainable services. 𝗪𝗵𝗲𝗿𝗲 𝗙𝗮𝘀𝘁𝗔𝗣𝗜 𝘀𝗵𝗶𝗻𝗲𝘀: • High-performance RESTful APIs • Microservices architectures • AI/ML model serving • Real-time systems (chat, streaming, notifications) Whether you're building a startup MVP or scaling enterprise systems, FastAPI offers the perfect balance of speed, simplicity, and power. Have you used FastAPI in production? What’s been your experience? Let’s discuss #FastAPI #Python #BackendDevelopment #SoftwareEngineering #APIs #Microservices #Async #TechLeadership

The Anatomy of a Better Dev Workflow 🛠️ I just came across this breakdown of the .claude/ folder structure, and it’s a masterclass in AI orchestration. I’m always looking for ways to streamline the "setup" phase of a task. This modular approach allows you to separate: Global Project Rules (committed to Git for the team) Personal Overrides (kept in .local files and gitignored) Automated Workflows (via the skills/ and agents/ folders) It’s essentially Infrastructure as Code (IaC), but for your AI assistant. If you aren't version-controlling your AI instructions yet, you're missing out on serious local development gains. #PHP #NodeJS #Javascript #AIforDevs #CleanCode #MahiITServices

“Turn text into React components using AI 🔥” What if you could generate React components just by describing them? 🤯 I built a project that converts text prompts into UI components using AI. 💡 Features: 🔸Generate React components from simple prompts 🔸Live preview for basic UI 🔸Clean code output for complex components 🔸Loading & error handling implemented 🛠️ Tech Stack: React.js, JavaScript, Vite, AI API ⚠️ Note: Due to free API limitations, usage may be restricted after a few requests. 🔗 Live Demo: https://lnkd.in/da3vrB4A 💻 GitHub Repo: https://lnkd.in/dSqy-WSu 💬 Try prompts like: “Loading spinner”, “Login form”, “Card UI” Would love your feedback! #React #WebDevelopment #AI #Frontend #JavaScript #Projects #Learning

Just shipped a Node.js AI backend from scratch! Built a production-style LLM server with: Custom system prompt + personality design Free LLM API integration (no keys hardcoded, .env based) Conversation memory (context-aware replies, long-term-ready) Clean REST endpoints, tested via Postman This project forced me to think like a backend engineer and a prompt engineer at the same time – not just “call the model”, but design how it thinks, remembers, and responds. Repo is live on GitHub – open to feedback, suggestions, and collaboration on smarter AI agents 🤝 #NodeJS #BackendDevelopment #LLM #GenerativeAI #APIDevelopment #JavaScript #OpenSource #StudentDeveloper #AIProjects