Claude Code just changed its defaults

Give Your Coding Agent Senior-Engineer Instincts - Addy Osmani packaged production-grade engineering skills your agent can load so it stops writing junior-dev spaghetti and starts thinking about edge cases, testing, and perf. Drop these into Claude Code or Cursor this week and watch your review comments drop, it's basically a staff engineer's brain as a config file, from the guy who literally wrote Google's performance playbook.

Let Your Agent Actually Debug in the Browser - Official Chrome MCP that hands your agent real DevTools access to inspect DOM, network, and console, no more guessing why the button's broken.

Make Claude Build Your n8n Workflows - Describe the automation in plain English and this MCP wires up the actual n8n nodes for you across Claude, Cursor, or Windsurf.

Stop Vibe-Coding, Start Spec-Driven - GitHub's toolkit forces you to write the spec first so your agent builds the right thing instead of confidently shipping the wrong one.

Code Reviews at Alibaba Scale - Hybrid tool mixing deterministic pipelines with an LLM agent for precise line-level comments and multi-language NPE/thread-safety catches.

Firecrawl Turns the Web Into an LLM-Ready Firehose - Search, scrape, and crawl any site into clean markdown your model actually understands, with 50k+ GitHub stars backing it up. If you've ever glued together Puppeteer, retry logic, and prayer to feed a RAG pipeline, this deletes that whole mess.

n8n Lets You Automate Workflows Without Selling Your Soul to Zapier - Fair-code automation with 400+ integrations, native AI nodes, and self-hosting so your data stays yours.

Langfuse Is the Observability Layer Your LLM App Is Secretly Begging For - Open-source evals, tracing, and prompt management that plugs into OpenTelemetry, LangChain, and the OpenAI SDK.

Cloudflare Open-Sources Its Vibe-Coding Platform , Cloudflare dropped an open-source app builder aimed at non-coders, and the real play is lock-in: build on their tools, deploy on their edge. Free scaffolding today, your infra bill tomorrow.

Replit's CEO Wants a Company That Runs Itself , Amjad Masad on running lean with agents. Ambitious vision, but "self-running company" is still 90% marketing, 10% reality.

AI Ad Startup Raises $30.5M to Plant Brands Inside Chatbots , Ads inside AI answers are coming for Best Buy and Target. If you're building on LLM outputs, plan for a sponsored layer you don't control.

OpenAI Ships GPT-5.6 Sol Upgrades, Wider Luna Access , Better Sol reasoning plus

33 Users From Talking to Strangers on the Street - No growth hack here, just brutal hustle: uni groups, cold street conversations, and months of grinding got this builder to 33 users with real paying customers. The copyable move? Distribution can start with your literal neighbors before you ever touch an ad dashboard, go where warm bodies already trust you.

TikTok Ads for SaaS Without Ever Dancing - Faceless screen-recording demos and UGC-style clips can carry your SaaS ads, no cringe choreography required.

10 Days In, Reality Bites the Kickstarter Plan - Three months of organic content across four platforms before launch, proof that "build an audience first" is easier said than converted.

Ditch AI Permission Prompts for 134 Standing Rules - Ashley spent four and a half months treating AI like a report instead of a chatbot, replacing click-to-approve theater with actual guardrails. This is the rare "I do things differently" post with receipts, and any builder drowning in permission fatigue should steal her playbook.

Coordinate AI Agents Across Sessions With Kiro Crew - AWS open-sourced a persistent workspace that wrangles coding agents across repos and schedules, worth a look before you build your own.

Steal From 100+ Open-Source AI Agents and RAG Apps - A free, growing repo of working agents and RAG builds; the fastest way to learn by copying real code.

An agentic IDE that catches your AI's bugs before prod does , Coldtea.ai pairs coding agents with visual QA agents that flag regressions before they hit production. If you're shipping fast with AI and quietly terrified of what breaks at 2am, this is the rare launch actually solving the mess AI coding creates instead of adding to it.

AdAnt AI , A team of creative agents that plans, builds, and iterates on social ads; built by folks claiming 50M+ organic views, so decent for solo marketers.

Wispr Flow Notetaker , Meeting notetaker that nails speaker attribution before the call starts, so your recaps and follow-ups don't hallucinate who said what. Handy if your calendar is a warzone.

The "I Can Build Anything But Can't Get Anyone to Care" Trap - A founder admits their product works (real users confirmed it unprompted), yet distribution is a black hole. The hardest lesson every technical builder eventually learns: shipping is 10% of the job, and the comments are a masterclass in why "if you build it, they will come" is a lie.

Docker's Disposable Sandboxes for AI Agents - 149 comments debating whether isolated throwaway environments finally make it safe to let agents run wild.

I Left Google to Build Startups, 2 Years, No Pretty Ending - Ex-Googler learns the hard way that pedigree buys you nothing when you're begging for your first ten users.

Your AI Coding Bill Is Out of Control - 264-comment HN thread on taming AI coding spend at scale before it eats your margins.

Build vs. Buy, Now With a Spreadsheet - A Wall Street firm built a tool to answer software's oldest question, so you don't have to guess.

AMD Buys Its Way Into Hardwired AI Silicon , AMD is baking AI models straight into chips to chase Nvidia's fat inference margins. If it works, hosted inference gets cheaper and your per-token API costs finally get a second bidder, competition is the only thing that drops your infra bill.

Google's AI Reporting Tools Hit Ads & Analytics in 2026 , Free AI reporting eats a chunk of the dashboard-SaaS market; if that's your product, differentiate now.

Caveman-Speak Your Way to Cheaper Claude Bills - This is either genius or a fever dream: a Claude Code skill that strips your prompts down to grunt-speak and claims a 65% token cut. Absurd premise, real savings, if it holds up, that's your API bill nearly halved for the price of sounding like a Neanderthal.

Give Your Agent an Actual IDE - MCP toolkit adding semantic code retrieval and editing so your agent stops grepping blindly and starts navigating like a senior dev.

Memory That Survives /clear - Captures, compresses, and re-injects session context so your agent finally remembers what it did yesterday.

Crash-Proof Planning for Long Tasks - File-based markdown plans that recover after compaction and fight context rot on marathon coding runs.

AI Pentester You Can Actually Run - Open-source AI agent that hunts and helps fix your app's vulnerabilities before the bad guys do.

Browser Sandbox for Agents, Zero Infra - Batteries-included open-source browser API that lets AI agents automate the web without you babysitting Chrome instances.

Any Webpage → Desktop App in One Command - Rust-powered wrapper that turns URLs into lightweight native apps, no Electron bloat required.