I took a break from blogging, then came back and wrote two posts back-to-back and didn’t get a newsletter out for the first one. So this one is a double feature:

Vibe-coding is not just for websites! You can build your own Chrome extensions and install them directly from a local folder- bypassing the Chrome web store. This lets you customize a certain website, change the new-tab page, or add a toolbar that works on all sites.

Ever had a spreadsheet that you wished was a web app? Or an Excel tracker that breaks every time someone inserts a row and doesn’t copy the formulas?

Many critical business operations are trapped in spreadsheets because of difficult to replicate features and formulas. But it’s only difficult if you rebuild them from scratch. HyperFormula is compatible with over 400+ Google Sheets and Excel formulas, and Handsontable provides a rich spreadsheet-like interface, making it easy to migrate business logic to a custom web application.

There are a LOT of ways to integrate different tools and services with Claude Code: MCP servers, Skills, Connectors in the marketplace, writing custom REST API integrations, etc. The best option depends on what service you are integrating with, and what they offer. But in general, I’ve found the most powerful option to be a CLI when one is available for the service you are integrating.

A robust, well-documented CLI gives Claude full access to all the methods and endpoints that MCPs and marketplace connectors tend to leave out, and they usually have wrappers that simplify usage vs the REST API.

One good example is Pinecone’s Claude Plugin and CLI. This gives Claude the tools to build RAG pipelines and assistants into your app, and configure everything from the terminal- without any manual setup in the Pinecone dashboard.

The AI brain-drain is real, and the longer you go without coding, the worse it gets. So how can you use AI to code, and still learn something? The trick is to add in more of the friction and muscle memory that AI coding tools remove!

You get a rush from vibe coding and seeing results quickly, but learning happens during the friction and muscle memory- and vibe coding skips that part. To learn, YOU have to be the one searching the docs, debugging the code, typing the syntax repeatedly. When AI does that for you, it’s like going to the gym and using a forklift to lift the weights for you.

Here’s how I approach vibe coding, to help me learn more in the process, with 𝘮𝘪𝘯𝘪𝘮𝘢𝘭 slow down:

Those cleaned up note cards become topics for me to explore later on YouTube, or voice chat with ChatGPT. Once I’ve done that extra research, I toss them out. They aren’t for reading again later. It’s the act of writing them that helps.

By the time I go to actually type any of this new code, I’ve written it by hand several times, watched videos, and chatted about it, with minimal slowdown during the vibe coding itself. I just make quick notes to build muscle memory, and then research those things later so that the vibes can soak into my brain instead of evaporating.

It has been quite a while since I recorded a new video, but I’m back! This one is short- but that’s the point.

They say a good quick start guide should be doable in 5-10 minutes. At Handsontable, we have 2 libraries: the Handsontable data grid UI library, and HyperFormula, the calculation engine for Excel-like formulas. They can be used together, but they each have their own quick start. Then there’s connecting your data and formulas.

I’m going to try doing all four, in only 5 minutes!

This one is from fellow DevRel, drummer, and metalhead, Ryan Blunden from Queensland, Australia! 🤘🥁

Ryan is starting a 1:1 coaching program to help DevRels, Developers, and Technical Marketers learn to create their own video content. No prior video experience needed.

He’s offering 10 people a free 60 minute coaching session. I met with him last week and it helped me with an idea I had been stuck on for months. I now have a clear picture of what I want to create, and chatting with Ryan helped me figure it out!

Until recently, running an LLM locally was more of a hobby or fun experiment, but not really useful for serious use cases. However, the latest generation of open source models are catching up with foundation models on coding benchmarks. Qwen 3.5 and Gemma 4 can both be run locally, enabling free and offline coding assistants with comparable code quality to commercial models.

In this guide, I’ll show you how to integrate a local model with your own app, allowing you to build AI-powered websites that run locally and without relying on an external service for the AI.

If you enjoy this newsletter, please share it with others in your network. Got a question or comment? Login at news.greenflux.us to share your thoughts.