#016 Latest 9/7–9/13 Published Sep 14

Craft, Contracts, and the Bill

This week's six pieces circle one question: is software still a craft, or has it become an assembly line? Bun's migration from Zig to Rust, and the argument it set off, forces that debate back into the open; Milkdown answers it in miniature by breaking an editor kernel into swappable, plugin-sized parts. From there the issue moves to interfaces. If agents are going to do real work, they need declared capabilities rather than guesswork: WebMCP lets a website register its actions as structured tools, NocoBase hands coding agents a mature data model, permissions and workflows to build on, and OpenClaw 2.0 turns the browser into a shared workshop. But machines that work also spend. Our closing piece traces a Claude quota that vanished in ten minutes to a forgotten background agent—91% of usage came from machines—and makes the case that meters and guardrails deserve the same design attention as features.

6 picks 3 sections ~2 hr

Section 01

Craft or Assembly Line: Two Answers on Making Software

2 / 6

newsletter.powderworks.dev · 76 min

Software Craft vs Industry: Bun's Rust Rewrite, Agents, Seat Belts软件还有手艺吗：Bun 迁 Rust、AI 代工与工业化的未来

Taking the 2026 Bun rewrite from Zig to Rust and the ugly public spat between Andrew Kelley and Jarred Sumner as its cue, this essay asks a question the author has carried through his career: is software production a craft or an industrial process? It combs through tabs-vs-spaces wars, the rise of formatters, Go's deliberate deskilling and work intensification, the Clojure community's hand-tool romance, woodworking hobby culture, and the US government's seat-belt mandates—all to build a metaphor for type systems and Rust's borrow checker. The author argues quality is subjective, 'slop' is in the eye of the beholder, and predicts coding agents will drive software further into industrialisation: language lock-in for small/medium projects disappears, hand craftsmanship becomes a hobby, and quality is increasingly quantified and automated. A thought-provoking culture essay with a clear pro-AI, industrialist viewpoint; not a technical tutorial.

github.com · 4 min

Plugin-driven WYSIWYG Markdown editor powered by ProseMirror & remark插件驱动的所见即所得 Markdown 编辑器框架

Milkdown is a plugin-driven WYSIWYG Markdown editor framework rather than a monolithic editor component. It builds on ProseMirror for document state and editing, and uses remark to parse and serialize Markdown, aiming for the seamless Typora-style writing experience. Editor features, from commands and shortcuts to themes and UI, are distributed as plugins, so teams compose only what their product needs while staying on a single underlying data model. Written in TypeScript, it fits web teams looking to embed a Markdown editor or to assemble a custom rich-text editing surface.

Section 02

Declared Interfaces: Letting Agents Into Real Systems

3 / 6

sreenathmenon.com · 25 min

WebMCP: Give AI Agents Declared Tools Instead of a DOM to Scrape把网站动作变成 Agent 可调用的接口：WebMCP 上手实测

WebMCP is an incubation-stage browser standard from the W3C Web Machine Learning Community Group, proposed by Google and Microsoft. Instead of scraping the DOM, a website registers its actions as structured tools with JSON Schema; an AI agent discovers them, calls them with typed arguments, and the page runs real JavaScript inside the user's already-open, authenticated tab. The author validates the pattern with a deployed demo, Career Copilot: it registers 13 WebMCP tools on a single page, and ChatGPT—which now supports WebMCP—used those tools in a real run, aggregating 75 live openings from GitLab, Stripe and Databricks, scoring 24 job descriptions, surfacing skill gaps, shortlisting 4 roles, then stopping for explicit human approval before submitting 3 applications. The post covers the minimal imperative API (one registerTool() call wrapping existing functions), the current Chrome 149 origin trial and flag, the call flow, and the trust model: same-origin restrictions, readOnlyHint / untrustedContentHint annotations, and human-in-the-loop confirmation for consequential actions. It also states limitations honestly: it is early, Chrome-first today, needs site adoption, and the security model is still maturing.

github.com · 9 min

AI-native no-code platform for building business systems with agentsAI 与无代码融合的开源平台，让编码 Agent 与人工协作搭建业务系统

NocoBase is an open-source, self-hostable platform that blends AI with no-code development for building business systems such as CRM, ERP, project management, and internal tools. Instead of treating AI as a black box that generates fragile code from scratch, it lets coding agents (Claude Code, Cursor, Codex, OpenCode, etc.) work on tested infrastructure while human users build and refine the same system through a WYSIWYG visual editor. The platform includes configurable AI employees that can join workflows—handling document recognition, risk monitoring, Q&A, and form filling—with role-based and field-level permissions plus audit logging. External agents can connect via MCP, HTTP APIs, CLI, or messaging channels such as Telegram, WhatsApp, Slack, and Gmail. With a microkernel and plugin architecture, business data stays in your own relational database, decoupled from UI and extensions. This makes NocoBase a pragmatic reference for engineers and teams exploring low-code systems, AI agents, and human-AI collaboration.

openclaw.ai · 5 min

OpenClaw 2.0, AccidentallyOpenClaw 2.0 的“意外”：两个月憋出 16,000 个 PR 的更新

OpenClaw 2.0 is the project's largest update yet, assembled by 933 contributors—569 first-timers—across over 16,000 pull requests, and follows an uncharacteristic seven-week release gap. The release emphasizes lowering the barrier to first use: setup reuses existing ChatGPT/Claude subscriptions, API keys, or local models already on the machine, while much configuration is deferred or eliminated. The rebuilt browser app becomes a first-class surface for setup, resuming work, and live participation. From a simple inbox-to-Telegram notification flow, the post argues, a Claw can grow into cross-app tasks like answering a brother's iMessage by finding a receipt, and finally into multiplayer via shared cloud sessions that let team members join or take over a task with context intact. Written as a product announcement, it offers community stats and feature direction but no architectural or code-level technical detail.

Section 03

The Machine's Bill: When Most Usage Isn't Yours

1 / 6

www.kelviq.com · 8 min

Forgotten Agent Ran 1,555 Sessions a Day and Ate My Claude Quota额度十分钟跑光不是计量 bug：被遗忘的 agent 一天开了 1,555 个会话

The author's Claude Max quota drained in 10 minutes, and he nearly blamed a faulty meter. Digging through Claude Code's local logs, he found a forgotten background agent from a YC batch project spawning 1,555 short-lived sessions a day, with 91% of usage generated by machines rather than by him. The post explains why the limits feel broken: the 5-hour cap is a rolling window that background sessions can already fill, and every fresh session pays a cache-warmup cost roughly 12 times higher than re-reading from cache. The author turned his two-day investigation into tare, an open-source Claude Code skill that deduplicates log entries (naive counts overstate tokens by about 86%) and answers with a cause instead of a raw spreadsheet. It runs entirely locally, makes zero network calls, and can produce a scrubbed HTML usage report that can be shared without leaking prompts or file paths. Useful for anyone whose Claude Code quota keeps disappearing and for teams auditing agent-side spending.