Today at a glance
The big headline is GPT-5.6, but the more useful pattern is broader: teams are moving from one-off AI usage to managed workflows. Agent skills, memory settings, AI-assisted migrations, and vendor governance all matter once these tools touch real production work. On the infrastructure side, Rust, Postgres experiments, Cloudflare Drop, and Turbopack memory work show the usual constraint still applies: good systems need strong feedback loops.
Items
GPT-5.6 · Hacker News / OpenAI
GPT-5.6 landed with three sizes: Luna, Terra, and Sol. The launch emphasizes long-context work, agentic performance, and better cost/performance tradeoffs. The practical question for engineering teams is no longer “which model is best,” but which class of task deserves which model tier and evaluation budget.
addyosmani/agent-skills · GitHub Trending
This repository packages production engineering practices as reusable skills for coding agents. It is a useful sign of where agent work is headed: away from private prompt snippets and toward reviewable, versioned operational knowledge. Teams adopting coding agents should treat these skills like any other shared engineering standard.
Postgres rewritten in Rust, now passing 100% of the Postgres regression tests · Hacker News
pgrust is an experiment, not a drop-in database choice, but passing the Postgres regression tests makes it worth watching. It highlights a serious long-term question for infrastructure: how much of the existing C systems world can be reimplemented or wrapped with stronger memory-safety guarantees. Even if Postgres itself is not rewritten, extensions and adjacent tools may move in that direction.
外部大脑越来越强，原生大脑逐渐成为瓶颈 · V2EX
This V2EX thread argues that as external AI tools improve, the human ability to frame problems becomes the bottleneck. That is a useful reminder for global teams too. Better agents amplify clear intent, acceptance criteria, and review discipline; they do not remove the need for them.
你们会开启 Codex 的 memories 开关，用来讲记忆带入到新对话吗？好用吗？ · V2EX
Codex memories triggered a practical discussion about whether long-term context should follow users across sessions. The upside is obvious: less repeated setup and better continuity. The risk is also real: stale preferences, accidental leakage, and project boundaries that are hard to reason about.
HTML即公開！Cloudflare Drop が面白そう · Zenn
Cloudflare Drop lets users publish static sites by dropping in HTML assets, folders, or zip files. It fits neatly into the AI-generated prototype loop: generate a page, upload it, share it, and iterate. For small tools, demos, and campaign pages, this trims away a lot of deployment overhead.
Next.js 16.3でTurbopackのメモリ使用量を劇的に減らした話 · Zenn
This Zenn post walks through a real-world Turbopack memory improvement after moving to Next.js 16.3. Frontend build performance often looks like a local annoyance until it hits Docker, CI, and large teams. The useful takeaway is to measure development tooling like production infrastructure, because it directly affects throughput.
JavaScriptランタイムのBun、Claude Fable 5を11日間稼働させてZigからRustへの移植を実現。Claudeをどう使ったのか？ · Publickey
Publickey covers Bun’s AI-assisted migration from Zig to Rust using Claude Fable 5. The important part is not just the volume of generated code, but the surrounding process: tests, parallel agent loops, adversarial review, and human supervision. Large migrations may become more feasible, but only when the validation system is strong enough.
Inviting hard questions · Anthropic News
Anthropic published a governance-oriented post inviting harder scrutiny of the company. It is not an API feature, but it matters for enterprise adoption. As AI vendors move deeper into regulated and critical workflows, trust, accountability, and public answers become part of the technical evaluation.
Introducing a way to reflect on how you use Claude · Anthropic News
Anthropic also introduced a way for users to reflect on how they use Claude. That points to a product layer beyond raw assistance: helping users understand habits, dependence, and boundaries. Builders of AI tools should watch this space, because usage review and preference management may become table stakes.
Editor’s note
Today’s mix is 3 English-source items, 2 Chinese-community items, 3 Japanese-source items, and 2 official Anthropic posts. All seven requested sources were reachable. Dev Digest 编辑 would start with GPT-5.6, agent-skills, and the Bun migration story: together they show that the next phase of AI engineering is about repeatable process, not just stronger outputs.