It feels later in the week than Tuesday. But Google Cloud Next is officially sold out, and much of our team is working towards delivering the best tech event possible for you all. And I had a few minutes to code up an agent today, so no complaints.
[article] Components of A Coding Agent. Fantastic. We suddenly start throwing terms like “harness” around and assume we’re all talking about the same thing. Sebastian offers up a terrific definition and deep dive into coding harnesses.
[article] Nobody Is Coming to Save Your Career. No one ever was. If you have a strong manager and a supportive org, fantastic. But you still own your career plan.
[article] Burnout Looks Different Across the Org Chart. Watch for These Signs. I haven’t seen it framed like this before. Early-career signals are different than mid or late career. Most of us probably recognize these!
[blog] Get more done with new vertical tabs and immersive reading mode in Chrome. I’m not sold on switching to vertical tabs. But maybe!
[article] Microsoft’s Agent Stack Confuses Developers While Rivals Simplify. Platforms are hard. Hyperscalers don’t always find the right abstractions, or focus.
[blog] I Still Prefer MCP Over Skills. These points resonate with me. I don’t personally want to write giant skill files that I have to store, share, and maintain. MCP does a lot of good things for me.
[blog] AI Code Reviews with Gemini CLI on GitHub Enterprise Server. A lightweight CLI can be added to a lot of environments. Here, Karl uses it to do code reviews with GitHub Actions.
[paper] Effective Strategies for Asynchronous Software Engineering Agents. Teams of coding agents can step on each other’s code and cause merge issues. This work proposes a model of central control and isolation using git worktrees.
[article] Google Open Sources Experimental Multi-Agent Orchestration Testbed Scion. I got this up and running over the weekend. It’s an interesting take on harness-agnostic orchestration.
[article] 27 questions to ask when choosing an LLM. Solid list. Tailor to whatever you actually care about, however.
[blog] TorchTPU: Running PyTorch Natively on TPUs at Google Scale. TPUs are hot, and we’re making the PyTorch experience on TPUs easier.
[article] Anthropic says its most powerful AI cyber model is too dangerous to release publicly — so it built Project Glasswing. This is that model rumored to get released. Here’s Anthropic’s take. We’ve also got it available in private preview on Google Cloud.
[blog] Good APIs Age Slowly. Useful reminders about how to shape an API, keep it decoupled from a frontend, and identify boundaries.
Want to get this update sent to you every day? Subscribe to my RSS feed or subscribe via email below:
Richard Seroter's Architecture Musings 
Leave a comment