Crosspost 1.0.4 has been released! 🔧 Updated LinkedIn API strategy 🔧 Upgraded MCP server SDK for better performance Details: https://github.com/human...

I noticed the LinkedIn strategy stopped working on Crosspost so I asked Copilot to investigate. It tried to post, evaluated the error that happened, then rewrote the code to use the new API. The future is so much fun.

In case you missed, my most recent blog post explains how GitHub merge queues can save you time: https://humanwhocodes.co...

Have you ever accidentally `git reset`? Quick tip to get back to where you were: git reset --mixed ORIG_HEAD

In case you missed it, my latest newsletter talks about the very real cost of burnout to your health and career along with some tips to avoid it. https://newsletter.human...

Reading @humanwhocodes.com's newsletter today made me smile when I realized I'm already doing the things he recommends to avoid burnout. Great minds think alike and all that I guess. 😅

tailwind-csstree 0.3.1 has been released! 🐞 Fixed Tailwind 4 @custom-variant parsing for ESLint CSS plugin and expanded syntax coverage Details: https://github.com/human...

I asked Gemini to redesign the Bredbox homepage. I'm pretty happy with the result. What do you think? https://bredbox.app

I'm getting scattered reports that yesterday's newsletter has ended up in people's spam folders. If you're a subscriber, please double-check your spam folder and mark as not spam. Thanks!

Burnout is a real problem for software engineers and can affect your health in addition to your career. In my latest newsletter, I share my story and some tips on how to avoid burnout: https://newsletter.human...

Merge queues are one of those sleeper GitHub features that can make a huge difference in your productivity. In my latest post, I explain what a merge queue is and how to use it: https://humanwhocodes.co...

Just re-sharing this blog post for no particular reason. https://humanwhocodes.co...

I'm getting tired of reviewing AI-generated pull requests from folks who don't understand the code they're submitting. If you can't answer questions about the code in the PR, please don't send the PR.

I updated my AI development pipeline so it outputs deployment instructions at the end. It's shocking how often I test something and think it's broken when it's just a service that hasn't deployed an update.

There are a lot of companies creating open source projects, which is great. I'd also like to encourage everyone to support and use independent open source projects. We don't want an ecosystem where all important packages are under the complete control of companies.

I think the next levelup for AI has to be enhanced memory. Larger context windows sound good but isn't as helpful if only 40% is usable. Layering in a memory solution that allows models to continue learning useful knowledge over the long term is key.

tailwind-csstree 0.3.0 has been released! ✨ Added @media screen() function support for Tailwind 3 🐞 Fixed !important support with @apply in Tailwind 3 Details: https://github.com/human...

Rewired my AI code generation approach to incorporate an orchestrator based on Burke Holland's Ultralight Orchestration. 🤯 Yes, this is the way. Will try to get it all cleaned up and shared at some point in the near future.

Pretty excited that both the Chrome and Firefox Bredbox extensions are now live! I'm now at 90% of how I used to use Pocket and there's still more to come. Check it out, and if you miss Pocket, sign up for an invite. https://bredbox.app/down...

tailwind-csstree 0.2.0 is out! ✨ Supports Tailwind 4 @custom-variant & [@slot] at-rules 🔧 Handles Tailwind v4 @theme wildcard colors 🐞 Decouples runtime tokenTypes from @eslint/css-tree Details: https://github.com/human...

Having spent the past couple days both in the Azure dashboard and Edge extension store, it's clear Microsoft *loves* taxonomies. Every page is a deep swim through multiple levels. It all makes sense at a global level, but the UX is such a slog. Big company UI at its worst.

My experience publishing browser extensions: 1. Firefox - super easy, barely an inconvenience. 2. Chrome - fill out this multi-page questionnaire, pay a fee, give us your first born. 3. Edge - dig through endless menus just to get to a screen that doesn't even work. 4. Safari - YOU MUST BUY A MAC!

tailwind-csstree 0.1.5 is out! 🔧 Made eslint/css an optional peer dependency 🔧 Improved pnpm compatibility with function-based syntax extensions Details: https://github.com/human...

I tried uploading a scanned PDF to Google Drive and the OCR to a doc was awful. I asked Claude Desktop how to get a clean conversion. It said, "give it to me." Five minutes later, I have a damn near perfect doc file. 🤯

If you have AI writing tech specs for you, be sure to also have AI update the tech spec after implementation to ensure it's up to date with the final approach.

Never thought buying a new oven would involve and extensive internet search for "disable wifi on oven."

One of my goals with Bredbox was to make the dev environment run 100% locally. That came in handy yesterday when I had three hours without internet.

My internet was out for three hours this morning. During that time, I had to revert to artisanal software engineering (no AI). And you know what? It was kind of fun. I got less done but I was able to get deeper into the code. I also felt calmer without all the inline suggestions popping up.

I find myself using GPT-5.3-Codex a lot more now in VS Code because it's the only one that consistently reads nested AGENTS.md files. This makes the experience so much smoother.

Using AI to write code has made me realize there are two types of code I've traditionally written: 1. Code that I want to know how it works 2. Code that I want to exist but don't care how it works I delegated 100% of the second category to AI now. Maybe 50% of the first.

The Supabase MCP server is really nice. I use Bredbox's production MCP server in read-only mode which enables the agent to find problems more easily. It just debugged an OAuth error in production by pulling the logs. Saved a ton of time.

The urge to multi-task is strong when using AI for coding. Single-tasking, however, still makes you more productive. In my latest newsletter, I discuss strategies for staying focused while AI does the heavy lifting: https://newsletter.human...

Here's a tip if, like me, you set up AI to do something in your Node.js project and walk away. Add this to your AGENTS.md > If changes to package.json are necessary, do those changes last. Prevents coming back an hour later to see nothing done because it was waiting for a confirmation.

I spent some time digging into proxying fetch requests in server-side JavaScript. It turns out each runtime does things somewhat differently. I wrote it all up here: https://humanwhocodes.co...

One of the benefits of using a PRD-driven approach to AI-assisted programming, is you can ask a model to produce marketing material based on the functionality described in the PRDs.

Finally figured out how to speed up Bredbox page loads. 🎉 Outside of the cold start penalty, the page-to-page navigation is much faster now. (I was hitting Supabase to verify the user on every request, but turns out that's not required with the new asymmetric token signing keys.)

Chrome Web Store requires you to publish a phone number along with your extension? Making me think twice about a Bredbox Chrome extension. The last thing I need is *more* spam phone calls.

Claude Desktop has been broken for me since Friday. The process starts but no UI is displayed.

My latest creation for Bredbox (with Copilot's help), a zero-dependency npm package that can extract the images from a .ico file (and also convert the BMPs to PNGs if you so desire): https://www.npmjs.com/pa...

One thing I've been surprised to learn while building Bredbox is just how poor HTML structure is in the wild. I've seen <title> tags inside <body>, incorrect og:type, missing favicon <link> on just the homepage. It's a jungle out there.

The Bredbox OAuth authorization service is now live! 😅 This is a case where AI not only implemented the code but taught me each step of the way. Just successfully tested the soon-to-come Bredbox Firefox extension with it.

I remember when apps started showing the weather when you signed in to let you know they were location-aware. I feel like summarization is the equivalent for AI.

It's pretty sweet that local Supabase now ships with a built-in MCP server.

Claude 4.6 Sonnet has been pretty impressive in my recent testing. I'm getting Opus-like performance on long-running, complex tasks.

Really disappointed that the VS Code - Gemini 3.1 Pro integration is still so flaky. I've been itching to give it a try.

One of the unfortunate side effects of company-backed open source projects is that people seem to expect all popular projects to be staffed full-time and just as responsive. Many maintainers work in their spare time on projects so this is an unfair expectation.

I really wish security vulnerability reports had more discernment over whether something represents an actual risk. Every report, we get flooded with ESLint issues demanding immediate upgrades even though there's no real exploit for ESLint. It's such a waste of everyone's energy.

It's tempting to use Claude 4.6 Opus for everything because it's so capable, but that's not cost efficient. The goal of AI-assisted programming is to get the greatest value for the lowest cost. That's why I think that budgeting is quickly becoming a necessary skill for software engineers.

It lives!

The pay-as-you-go X API looks enticing, but it won't return you full articles. You get the raw article text with all images, code blocks, and links removed. Then you get arrays of each with no location info. There's no way to reconstruct the whole article.