AI agents are quietly breaking open source

From React patterns to AI agents, plus lessons from RollerCoaster Tycoon and modern cloud infrastructure.

March 23, 2026

Hey there.

Welcome to Full Stack Focus. 👋

This week, we’re looking at the reality behind modern development, from bloated 49MB web pages and JavaScript quirks to the growing influence (and risks) of AI coding agents. Along the way, we’ve got lessons in performance, open-source signal vs noise, and the kind of engineering insights that actually stick with you.

Free email without sacrificing your privacy

Gmail tracks you. Proton doesn’t. Get private email that puts your data — and your privacy — first.

Front-End

💔 The 49MB Web Page
Someone audited a major news website and found it's shipping a staggering 49 megabytes of data to your browser. This deep dive breaks down exactly where all that weight is coming from - spoiler: it's not just images. If you care about web performance and want a real-world example of what not to do, this is a sobering read you'll want to bookmark.

⚠️ JavaScript's date parser is out of control and needs to be stopped
Turns out JavaScript will try to parse almost anything as a date - and succeed in the most unhinged ways possible. This post documents the weird, wild, and genuinely dangerous edge cases hiding inside the language's date parsing logic. If you've ever had a bug you couldn't explain, there's a good chance this rabbit hole has your answer.

🐭 I prompt injected my CONTRIBUTING.md – 50% of PRs are bots
A maintainer got so tired of AI-generated spam PRs that they embedded a hidden prompt injection inside their CONTRIBUTING.md file - and the results were revealing. Half of all incoming pull requests failed the test, confirming what many open source maintainers already suspect. This is a fascinating look at how AI coding agents are quietly changing the contribution landscape in ways nobody really planned for.

🛠 Start naming your useEffects
If your React codebase is littered with anonymous useEffect calls, this short but punchy post makes a compelling case for giving them names. Named effects show up better in stack traces, make your intent clearer to teammates, and cost you almost nothing to implement. It's one of those tiny habits that separates clean codebases from the ones that make people groan during onboarding.

🎨 I built a landing page to preview every Tailwind color
Picking the right Tailwind CSS color palette just got a lot less guesswork-heavy. Someone built a landing page that lets you visually browse and preview every color Tailwind ships with, all in one place. If you've ever spent way too long toggling between slate-600 and zinc-600 wondering which one looks right, this tool is going to save you a surprising amount of time.

Programming

👀 The gold standard of optimization: A look under the hood of RollerCoaster Tycoon
RollerCoaster Tycoon was famously written almost entirely in hand-tuned assembly by a single developer, and the result was a masterclass in squeezing every last drop out of hardware. This post pulls back the curtain on the techniques Chris Sawyer used, and why they still hold up as a benchmark for thoughtful optimization. Whether you write games or enterprise software, there's something genuinely inspiring about code this intentional.

✒️ Some things just take time
The creator of Flask reflects on why some problems in software simply can't be rushed, no matter how talented your team or how much pressure you're under. It's a grounded, honest take on patience, compounding progress, and why slow and deliberate work often produces better outcomes than shipping fast and fixing later. A great read for anyone feeling the heat of unrealistic deadlines.

📈 Slug Algorithm released into public domain
After over a decade of use, the Slug Algorithm - a method for generating URL-friendly slugs from arbitrary strings - has been released into the public domain. If you've ever rolled your own slug logic and ran into edge cases with unicode, punctuation, or whitespace, this is worth a look. It's a small but well-considered piece of infrastructure that a lot of projects quietly need.

🐭 Storing 2 bytes of data in your Logitech mouse
This delightfully weird post explores what happens when you treat your Logitech mouse's onboard memory as a tiny storage device. It's not practical, it's not something you should do in production, and it's absolutely worth reading anyway. A fun reminder that low-level hardware hacking is alive and well, and that curiosity is still one of the best traits a developer can have.

☁️ How AWS S3 serves 1 petabyte per second on top of slow HDDs
Amazon S3 moves an almost incomprehensible amount of data every second, and the hardware underneath is mostly just regular spinning hard drives. This breakdown explains the layered architecture, caching strategies, and distributed systems tricks that make those numbers possible. If you work with cloud storage or distributed systems at any scale, this is the kind of engineering deep dive that genuinely changes how you think about infrastructure.

88% resolved. 22% loyal. Your stack has a problem.

Those numbers aren't a CX issue — they're a design issue. Gladly's 2026 Customer Expectations Report breaks down exactly where AI-powered service loses customers, and what the architecture of loyalty-driven CX actually looks like.

AI

⚠️ OpenClaw is a security nightmare dressed up as a daydream
OpenClaw has been getting attention in the AI agent community, but this post argues that underneath the promising surface lies a serious set of security vulnerabilities. The analysis covers how the tool handles permissions, data access, and trust boundaries in ways that should make any security-conscious developer pause. Before you integrate it into your stack, this is required reading.

🤖 OpenCode – Open source AI coding agent
OpenCode is a new fully open-source AI coding agent that aims to bring the kind of functionality you'd find in Copilot or Cursor to a codebase you can actually inspect, modify, and self-host. It's early days, but the project is already generating serious interest from developers who want AI-assisted coding without handing over their code to a third party. Worth keeping an eye on as it matures.

📈 Introducing Unsloth Studio: A new open-source web UI to train and run LLMs
Unsloth has been one of the most exciting projects in the open-source LLM space for making fine-tuning faster and cheaper, and now they've shipped a full web UI to go with it. Unsloth Studio lets you train and run large language models without writing a line of setup code, which dramatically lowers the barrier to entry for custom model training. If you've been curious about fine-tuning but intimidated by the tooling, this could be your entry point.

More Tools & Stories

Have a great week, see you the same time next week!

Full Stack Focus team :)