今年只參加到 PyCon Taiwan 2026 第二天，但仍然收穫滿滿。 Max Lai 分享醫院導入 AI Coding 與團隊轉型的故事，很多台灣公司熟悉的人情世故，都很有共鳴。 Panel 裡也有個蠻喜歡的觀點：「跟人有關的事，都變成更明顯的瓶頸。」 AI 工具一直在變，但（是我無法推開生命的改變 🎶）回到社群跟人交流，依然是研討會最有價值的部分。 期待 2027！ yang.keitheis.org/blog/pycon…

Another fantasy experience with PyCon Taiwan 2026 at 臺北醫學大學教學大樓

Zen Browser @zen_browser feels like the new best Arc — and it’s still actively being developed. Even better: on Linux, just open the Omarchy menu and type “zen” to install. Arc is never coming to Linux anyway. It also gives me exactly what I need: macOS + Linux, with things synced through Firefox Sync.

The hottest Python project right now might be Hermes Agent—the #1 AI agent on OpenRouter by usage. Though on the AI timescale, “right now” may already be old news. A PyCon Taiwan ticket-sales thought: we live in an age when even pi doesn’t end in .py

On a whim, I took a look at pi-dev, the supposedly minimal agent harness. In just one week, from v0.84.4 (Aug 28, 2026) to v0.85.0 (Sep 4, 2026): • 447 commits • 688 files changed • +95,441 / −25,115 lines Now that’s some serious AI-agent velocity.

Seeing what happened with Zeabur is another reminder: the closer a platform gets to infrastructure, the more critical security becomes. From TSMC’s almost military-like operational discipline to Cloudflare’s continued investment in AI security, the lesson keeps repeating: security is never something you finish in a day or two, or by patching a few holes. It is a long-term engineering practice. After nearly 20 years across companies of very different sizes and stages, I also understand the other side. For startups and growing companies, when the team is fighting for growth—or simply survival—security is often hard to keep at #1 or #2 on the priority list. It’s not that people don’t know it matters. There’s just always something more urgent on fire. But precisely because AI is accelerating implementation and delivery so dramatically, I think Engineering Leads and Lead Developers need to use AI to buy back time—and reinvest more human attention into the things AI won’t automatically build for us: permissions, reviews, guardrails, incident response, and system resilience. AI makes implementation faster. It can also make failures—and compromises—happen faster. Human judgment doesn’t automatically speed up with it, and the leadership responsibility and planning behind software engineering haven’t become any smaller.