Ai-Work

Niklas Gruhn’s short, sharp essay against relaying AI output verbatim in human communication. When someone asks a question in Slack, leaves PR feedback, or argues in a WhatsApp group, the worst response is “Claude said: [giant verbatim output]” — the other person can prompt Claude themselves, faster, with their own context; they don’t need a meat proxy in between. Reading AI output is extra effort: it’s verbose, full of all-too-plausible nonsense, and increasingly jargon-dense (his example: “NATS control-plane events: stream leader election / R3 quorum re-form during pod churn” — he had to look up nearly every word). The rule: prompt AI freely, but read it, understand it, validate it, then write the response in your own words — your own-words version is a “decent certificate” you actually did those steps. The sharpest edge is code review: you can ship code with near-zero effort by copy/pasting tickets into Claude Code and never reading what it wrote — but then who did the implementation? The reviewers did, using Claude Code, and you were the meat proxy. ...

David Crawshaw (Tailscale co-founder, now building exe.dev) argues the open-source-everything era of devtools has arrived not as ideology but as a consequence of AI agents. Five years ago, custom software rarely made sense — the cost of maintaining and learning a codebase dwarfed its benefit, so we amortized customization through config files, plugin systems, and extension APIs. Agents change both sides of the ROI: the setup prompt (“download the source, build it for local use, record why in version control”) makes personalizing trivial, and the maintenance prompt — a nightly cron that fetches upstream and rebases local changes on top, checking the software still works — makes staying in sync automatic. The thesis: for personal or small-team software, the source code is the extension system. He demonstrates it by wiring his own diff-minimizing tool (meat.dev) into his agent Shelley with a single prompt, and contrasts it with Claude Code, which is closed-source — “you don’t get to personalize it.” The practical pattern for agent users: record the why of local changes in version control, rebase onto upstream nightly, and codify the workflow as a skill. ...

Jeff Su’s counter to the default “jump in, pick a template, start prompting” tutorials — that path gives you generic output and burns tokens fixing unusable slides. His fix: three files prepared ahead of time, demonstrated with the actual deck he used for a paid workshop. ...

Futurepedia’s full-platform guide to Claude Design after its big upgrade — the host skipped covering it at launch because usage limits made it barely usable; that’s fixed (usage now bundles into your existing Claude credits). The overview: 15+ template types (mobile apps, slides, documents, wireframes, animations, UI mockups, resumes, 3D objects, HTML email, flyers), and the design-system workflow that stops output from looking like generic AI slop. ...