I wrote my standard repo setup down as one spec I hand to my coding agent: pnpm pinned, oxlint, Hugo docs, Renovate, AGENTS.md, CI on Vercel. Every new project starts from the same baseline, and I don't re-decide config each time. https://t.co/f4WrURAyq0
Wanted private docs for my site repo without spinning up another repo. Ended up deploying the Hugo docs as a second Vercel project from the same repo, behind deployment protection. One codebase, two deploys, and the docs still run locally with one script.
Especially for frontend PRs. A short clip of the UI actually working tells a reviewer more than the diff does, and it catches the "wait, that's not what I meant" moments before merge.
Needed a sharper headshot for a site refresh, so I ran Real-ESRGAN locally. 4x upscale in seconds, right on my laptop. Worth checking for an open model before you reach for a subscription.
I stopped debating hero layouts and had an agent spit out five rough HTML versions. Side by side, the right one was obvious in a minute. Real renders beat polished mocks when you just need to decide. https://t.co/f4WrURA0As
@madebyheejun Exactly — task-based allowlists beat a permanent mega-list. If the short set cuts wrong turns and review time, it’s doing the job. Tool count is just a proxy.
Spent yesterday pruning Claude Code plugins. Easy to enable dozens; I capped at ~8–10 that match how I actually ship frontend — design review, Figma, Context7, Playwright + TypeScript.
More tools isn't better agent output. A short list is.
TL;DR
- Fast runners (ex Blacksmith)
- Oxlint for ESLint, bun check for tsc
- Bigger test machines, Vitest isolation off
- Fix caching, Blobless clones
- Parallel jobs, nx affected
- Pre-push hooks
What are you waiting for? Send this to your agent :)
The most useful thing I built on a https://t.co/XfyDZuMIVm map app wasn't a layer. It was a debug overlay that drew picking buffers and tile bounds right on top of the real view. WebGL bugs stay invisible until you make the GPU show its work.
Vercel CI is in beta — Turborepo-aware skipping, agent-ready via MCP/CLI, 2–6× faster push-to-pass on their own repos. This is the direction I’ve been pushing: less waiting on CI, more room for coding agents with real guardrails.
https://t.co/zkBBPQOG9p
Last week at my current full-time gig. Still looking for the right Staff frontend role, and taking select contract work on the side. Been hanging out in practitioner Slack communities lately — those conversations feel way more useful than scrolling job boards.
Ripped ESLint out of my own site repo and switched to oxlint. Deleted a pile of config and plugins, and linting went from something I waited on to something I stopped thinking about. Rust-based tooling is quietly becoming my default for new frontend projects.
Coding agents got a lot more useful for me once the repo had real guardrails: Vitest, Playwright, Storybook visual tests.
The agent writes code, the tests say no, it fixes it. Without tests it's just confident guessing.
A fractional frontend lead isn't a part-time architect who hands off diagrams. It's someone who sets direction, reviews the PRs, and still ships code, a few days a week, without you opening a full-time req.
That's the role I take on through BQC: https://t.co/f4WrURAyq0