Narrative on Top AI Skills

I Let AI Agents Refactor My React Codebase — The Results Were Brutally Honest

Mon, 20 Jul 2026 00:00:00 +0000

Most AI Refactoring Advice Is Useless Without Context

I spent three weeks letting AI coding agents refactor a medium-sized React app — about 15K lines across 80 components. The results: some genuinely clever improvements, a few WTFs, and one incident that dropped our test coverage by 12%. Here is the raw breakdown.

The Setup: Why My Codebase Needed a Stranger’s Opinion

My team’s React app started life as a quick prototype. Two years and six developers later, it is a production app with 80+ components, a custom design system that nobody documented, and a test suite that started at 92% coverage but had been slowly bleeding. We knew the code needed love, but nobody had the time to audit every component.

I Let AI Handle My Team's Approval Workflow for a Week — Here's What Happened

Fri, 17 Jul 2026 00:00:00 +0000

I handed my AI agent control over our team’s approval workflow for five days. Expense reports, leave requests, purchase orders — all processed through an AI skill. Here is what went right, what broke, and whether I would do it again.

Why I Decided to Try This

Our team runs on Lark. I manage a group of fifteen engineers, a couple of product managers, and a rotating cast of interns. Every week brings a steady stream of approval requests. Leave applications, hardware purchase requests, conference travel approvals, expense reports from team lunches — the kind of administrative overhead that is simple but constant.

I Let My AI Agent Manage Lark Documentation for a Week — Here's What Happened

Sun, 05 Jul 2026 00:00:00 +0000

Why I Put Documentation on the Chopping Block

Our team lives inside Lark (Feishu). Meeting notes, technical specs, project briefs, onboarding guides — it is all in Lark documents. And like most teams that use collaborative docs heavily, we have a mess: 40+ documents scattered across spaces with inconsistent naming conventions, outdated meeting notes that nobody archived, and a growing pile of “untitled document” orphans that were created during a call and never revisited.

I Generated 100 AI Images From My Coding Assistant

Sat, 20 Jun 2026 00:00:00 +0000

Why I Went Down This Rabbit Hole

Thursday afternoon, 3 PM. I was building a landing page for a side project and needed a hero image. Not a placeholder — something that actually looked like the product. My usual workflow is: write a prompt in ChatGPT, download the image, drag it into the project folder. Then tweak the prompt, repeat. Every iteration costs a context switch.

I remembered the ai-image-generation skill sitting in my installed list from when I first set up inference.sh. I had never actually used it. The idea of generating images without leaving the terminal seemed convenient in theory, but I was skeptical about output quality compared to Midjourney or the ChatGPT web interface.