Today I want to share a guide I wrote about building forms in React. Forms look simple until you actually build one. Then you're dealing with validation rules, error states, and edge cases everywhere.

The guide starts with a basic form and progressively adds features. You'll learn:

• When to use uncontrolled inputs versus controlled ones
• How React Hook Form can eliminate most of your boilerplate
• Why Zod makes complex validation easy to manage
• How to set up proper client and server validation for production apps

Each section builds on the last so you can see exactly what changes and why. There are also interactive code examples you can play with directly in the browser.

Check out the blog with interactive examples here: https://akrom.dev/blog/react-forms-guide

With vibe coding and AI agents writing more code, bugs are multiplying faster than we can squash them.

One of the most common culprits? Wrong types and handling data incorrectly.

TypeScript catches type mismatches, but Zod takes it further with precise validation errors.

Put together a practical guide if you want to level up your validation game.

🔗 https://akrom.dev/blog/practical-zod-guide

🎯 Stay tuned on @akromdotdev for upcoming articles

Haven't finished yet, but excellent read so far. Highly recommend it

https://blog.mikeswanson.com/backseat-software/

Claude code added a new command called /insights.

When you run it, it will read your message history from the past month. It'll summarize your projects, how you use Claude Code, and give suggestions on how to improve your workflow.

https://x.com/trq212/status/2019173731042750509?s=46

I used to think Claude Code was a worse alternative to Cursor AI.

I was wrong.

Cursor AI is amazing. Polished UX. Great inline edits.
But with newer models like Claude Opus 4.5, my monthly bill quietly climbed to $500.

Worse than the cost was the mindset.
I kept thinking about tokens instead of code.

After using Claude Code daily for a few weeks, my opinion changed completely.
I am mostly using Claude Opus, which Claude Code provides out of the box.

What stood out:

• Slash commands are genuinely useful
• Skills feel like real workflow upgrades
• Long-context reasoning is 🔥
• No constant “am I burning money?” anxiety

Result:
👉 I became about 2x more productive.

Today, I am on Claude’s $100 plan, and it covers everything I need.

Cursor still has the best IDE experience.
But Claude Code became my default for real work.

Full write-up here:
👉 https://akrom.dev/blog/cursor-to-claude-code

Follow for more posts:
👉 https://t.me/akromdotdev

Which tool are you using right now, and why?

🏰 Morning Briefing — Friday, March 6, 2026 · 5:11 AM

━━━━━━━━━━━━━━━━━━━━━━
🤖 TOP 5 AI/DEV NEWS
━━━━━━━━━━━━━━━━━━━━━━

1. OpenAI open-sources Symphony — autonomous coding agent framework
Symphony connects issue trackers (like Linear) to LLM agents that execute full "implementation runs" — entire dev tasks, no developer middleman required. Still a preview, but already viral.
🔗 https://www.marktechpost.com/2026/03/05/openai-releases-symphony-an-open-source-agentic-framework-for-orchestrating-autonomous-ai-agents-through-structured-scalable-implementation-runs/

2. Critical Claude Code vulnerability: AI executes attacker commands just from opening a project
A prompt-injection flaw let malicious repo files trigger commands on claude open — zero user interaction. Patch is out; update immediately if you use it.
🔗 https://cybernews.com/security/

3. LexisNexis breached via "React2Shell" — 2.04GB leaked
A hacker exploited an unpatched React frontend flaw to pivot into LexisNexis's cloud infrastructure and dumped 2GB of data publicly. Frontend security → real infra risk.
🔗 https://www.prismnews.com/news/hacker-posts-204gb-of-lexisnexis-lp-files-after-claimed-cloud-intrusion

4. China's Five-Year Plan makes AI its national operating system
The new plan embeds AI across manufacturing, healthcare, logistics, and education — paired with a "tech self-reliance" push (quantum, 6G, robotics). Industrial-scale deployment, not model benchmarks.
🔗 https://www.reuters.com/world/asia-pacific/china-vows-accelerate-technological-self-reliance-ai-push-2026-03-05/

5. Broadcom forecasts $100B+ in AI chip sales by 2027
Bespoke AI silicon demand is ahead of schedule. Broadcom's custom chip bet is paying off as hyperscalers shift away from commodity GPUs.
🔗 https://www.reuters.com/technology/artificial-intelligence/

━━━━━━━━━━━━━━━━━━━━━━
🔥 GITHUB TRENDING TODAY
━━━━━━━━━━━━━━━━━━━━━━

🥇 moeru-ai/airi — 3,006 ⭐ today
Self-hosted AI companion with realtime voice chat, Minecraft/Factorio support. TypeScript.
https://github.com/moeru-ai/airi

🥈 KeygraphHQ/shannon — 2,930 ⭐ today
Fully autonomous AI pentester for web apps & APIs. 96% on XBOW benchmark. TypeScript.
https://github.com/KeygraphHQ/shannon

🥉 aquasecurity/trivy — 298 ⭐ today
Container/K8s/code vulnerability scanner. Go. Industry standard, worth knowing.
https://github.com/aquasecurity/trivy

CodebuffAI/codebuff — 275 ⭐ today
Generate code straight from the terminal. TypeScript.
https://github.com/CodebuffAI/codebuff

microsoft/mcp-for-beginners — 137 ⭐ today
MCP curriculum with examples in TS, JS, Python, Rust, .NET, Java.
https://github.com/microsoft/mcp-for-beginners

━━━━━━━━━━━━━━━━━━━━━━
🚀 TOOL RELEASES
━━━━━━━━━━━━━━━━━━━━━━

Next.js 16.1.6 ✅
React Compiler 1.0 is now stable and baked in. New DevTools MCP server lets AI coding agents automate upgrades via natural language. Upgrade path: npx @next/codemod@canary upgrade latest
🔗 https://nextjs.org/docs/app/guides/upgrading/version-16

Astro 6 (beta) 🔶
Post-Cloudflare acquisition (Jan 2026), Astro 6 brings a workerd-based dev server, Live Collections, and improved CSP. The Islands framework now has enterprise muscle behind it.
🔗 https://alexbobes.com/programming/a-deep-dive-into-astro-build/

React, TypeScript, Zod — No new releases in the last 24h. React Compiler 1.0 was the headline this cycle, now stable in Next.js 16.

━━━━━━━━━━━━━━━━━━━━━━
💬 HACKER NEWS HOT
━━━━━━━━━━━━━━━━━━━━━━

🔹 "Labor market impacts of AI: A new measure and early evidence" — First real empirical data on which job categories are getting displaced. Worth reading if you want numbers over narratives.
🔗 https://news.ycombinator.com/item?id=47268391

🔹 "A standard protocol to handle and discard low-effort AI-generated PRs" — Someone built 406.fail: a spec for rejecting AI slop PRs with a standardized HTTP 406 response. Discussion is lively.

🏰 Daily Briefing — Tue Mar 24, 2026

───

🔥 Top story: iPhone 17 Pro runs a 400B LLM
Someone demoed a 400-billion parameter model running locally on iPhone 17 Pro. HN is losing it (600+ points). Edge AI just entered a new era. → Twitter

───

🤖 AI/Dev News

1. Claude can now control your Mac (yesterday)
Anthropic launched a research preview of full computer use — Claude opens your apps, navigates browsers, fills spreadsheets. Paid Claude Cowork + Claude Code users, macOS only for now. Think of it as a remote employee in your laptop. → 9to5Mac

2. Xiaomi MiMo-V2-Pro "Hunter Alpha" revealed
Mystery model appeared on OpenRouter 2 weeks ago — everyone assumed DeepSeek V4. Turned out to be Xiaomi. Benchmarks: 61.5 on ClawEval vs Claude Opus 4.6's 66.3, at a fraction of the cost. China's consumer hardware players are serious about foundation models now. → VentureBeat

3. FCC bans foreign-made consumer routers
Added to national security covered list. Big implications for home/office networking — TP-Link, Huawei gear etc. affected. (316 pts on HN)

4. Nudge Security: 80% of orgs have rogue AI agent problem
New tools to discover AI agents employees spin up without authorization. 4 in 5 organizations already seeing data-access risk from unsanctioned agents. Worth tracking if you build B2B.

───

📦 GitHub Trending Today

| Repo                              | Stars | +Today | Notes                                               |
| --------------------------------- | ----- | ------ | --------------------------------------------------- |
| bytedance/deer-flow               | 41.2k | +3.5k  | SuperAgent harness (research, code, create)         |
| Crosstalk-Solutions/project-nomad | 14.1k | +4.1k  | Offline survival AI computer (TypeScript)           |
| browser-use/browser-use           | 84k   | +1.2k  | Still pumping                                       |
| TauricResearch/TradingAgents      | 40k   | +2.5k  | Multi-agent LLM finance framework                   |
| czlonkowski/n8n-mcp               | 16.2k | —      | MCP for building n8n workflows via Claude           |
| affaan-m/everything-claude-code   | —     | —      | Claude Code optimization — skills, memory, security |
| hesreallyhim/awesome-claude-code  | —     | —      | Curated Claude Code skills, hooks, slash commands   |

───

💬 HN Buzz

• ProofShot — CLI that gives your AI coding agent browser eyes. Agent runs your app, takes screenshots + collects console errors, bundles into a self-contained HTML report. Packaged as a skill for Claude Code/Cursor/Codex. proofshot.argil.io
• Autoresearch on an old research idea — dev used AI agents to auto-research a 3-year-old idea of theirs, found it actually held up. Interesting experiment in using agents for scientific self-review. (359 pts)

I added a chatbot to my portfolio site that answers questions about my work and projects. It searches my actual content by meaning, finds the relevant pieces, and sends them to an LLM to write an answer.

The whole thing runs for $0/month. Embeddings use a local 22MB model, search is just cosine similarity over a JSON file, and the LLM is on Groq's free tier.

I wrote up the full approach:
- splitting content into searchable chunks
- semantic search with local embeddings
- getting a local AI model to actually run on Vercel serverless

👉 https://akrom.dev/blog/building-rag-chatbot

There's a chat button on the site if you want to try it.

Monitor model is samsung g9 oddsey 49 if anyone is interested. It supports PiP (pincture-in-picture) split.