I've been experimenting with a bunch of AI coding tools lately, and I keep running into the same lesson: the tool doesn't matter nearly as much as the systems you build around it.

If you give Claude or Codex a vague prompt, you're going to get mediocre output. Generic apps, generic code, generic everything. But if you build systems around that agent, things like skills that define how it handles specific tasks, memory that captures your preferences across sessions, processes that enforce your standards on every run, the output gets dramatically better.

Systems have always mattered in development. I've been using a WordPress launch checklist at work for over a decade. It still catches things like making sure search engine indexing is on for production. The checklist isn't glamorous, but it works because it's a system.

With coding agents, systems matter even more. These tools can go off the rails fast if you're not guiding them with defined processes. Skills, sub-agents, onboarding docs for new codebases. That's where engineering happens now. Not in writing the code yourself, but in building the harnesses that make agent-written code consistently good.

There's this thread in tech culture that says you should always be chasing the newest tool, the hottest framework. But I keep coming back to the same question: are we building the systems that make any of these tools actually useful? Because a developer who understands how to engineer agent workflows, build repeatable processes, and onboard others into those systems is going to be way more valuable than someone who just knows which prompt to type.

🎥 Watch
Real-Time UI with Design Systems & AI
Brad Frost
This is a very cool proof of concept on using a design system along with AI tools and the Web Speech API to generate user interfaces in real-time. Very cool to see how you could bring a tool like this into a concept meeting and be able to leave with a prototype using your own design system.

💻 Tool
Perplexity Personal Computer
Perplexity
Perplexity opened up the waitlist for their Personal Computer product that builds on their Perplexity Computer product. Personal Computer looks to be the system that brings your AI tools together to allow you to handle tasks from an always on, always accessible machine. It feels like Perplexity's pass at an OpenClaw for a wider market. No setup and no configuring your own VPS or machine for OpenClaw to live on.

🔧 Tool
Why Google Workspace CLI is Such a Big Deal
The AI Daily Brief
CLI's are so hot right now. The latest is the Google Workspace CLI now available on GitHub. The AI Daily Brief does a good job breaking down how this is useful for agents. It gives access to Drive, Gmail, Calendar, etc. all under a single CLI. The tool is "built for humans and AI agents" with 40+ skills. Which are worth taking a look at themselves for ideas.

🧠 Guide
The Complete Guide to Building Skills for Claude
Anthropic
Speaking of skills, Anthropic released a PDF guide to building skills for use in Claude. Given skills stick to a defined spec it is useful information regardless of your agent of choice and a nice simple breakdown to share with your teammates.

📰 Article
Crawl entire websites with a single API call using Browser Rendering
Cloudflare
Cloudflare's latest release brings the capability to crawl entire sites from a single API call with rendering in a headless browser. You can also get back content as markdown. Great for using with agentic tools to monitor and train on content.

Thanks for reading,
Jason