yama | software developer 🇯🇵

Why AI-generated code lives less than a week. Not because it's wrong. ecause nobody remembers why it exists. AI writes fast. Humans forget faster. Only naming and structure bridge the gap between them. #SoftwareEngineering

The teams shipping fastest with AI aren't the ones who prompt best. They're the ones who delete generated code without hesitation. Attachment to output slows you down. Speed comes from treating AI as a rough draft machine, not a solution provider.

AI is a speed boost, not a brain upgrade. Anthropic found comprehension dropped 17% with AI. Debugging got hit hardest But those who used AI to verify their thinking, not replace it, maintained strong learning outcomes. The tool stays sharp. The question is whether you do too.

【Good News】 Claude Code's biggest challenge, The inability to carry over memory across sessions, has finally been resolved. By compressing persistent memory across sessions and feeding it to Claude, it can retain knowledge about the project even after termination or reconnection. OSS available here: github.com/thedotmack/cla…

Go 1.26's new GC "Green Tea" cuts overhead by 10-40% with one key shift: scanning pages instead of individual objects. The old bottleneck? 35% of marking time was just waiting for memory. Batching objects per page and processing sequentially makes cache misses drop. Upgrading and profiling my services next. #Go #GarbageCollection

Patterns where I feel I'm effectively utilizing AI coding tools within my team They don't optimize for "faster code generation." They optimize for "clearer problem definitions." The tool just happens to work better when you know exactly what you're building.

The more I use AI coding tools, the more I find myself going back to manual for certain things. Not because AI can't do it, but because some decisions need friction. Friction forces thinking. Speed isn't always the goal. #AI #SoftwareEngineering

MCP Apps is standardizing how AI agents render interactive UIs inside chats. "here's the data" → "here's a chart you can interact with" Claude, ChatGPT, VS Code speaking the same UI language📝 stream-tech.dev/en/media/techn…

Seeing more requests like: “AI already got it mostly done.” Now that “mostly” is easy, the real value shifts to the last 10%: architecture decisions, ops, infra, edge cases. AI can get you to “mostly built.” Engineers still get it to “actually works.”

📅 Today's tech digest MCP Apps now let AI operate dynamic UIs directly - could be a turning point for agent interfaces. WhatsApp completed a massive Rust migration. Meanwhile, Japan reported record-high security incidents last year. #developers #AI

Clawdbot renamed to Moltbot overnight. Anthropic's trademark request. 67k GitHub stars, 309 contributors—and suddenly your project name is a legal risk. Building on someone else's platform is one thing. Building with their name in yours? A ticking clock. #opensource #AI

📅 Today's tech digest Windows 11 KB5074109 causing boot failures - might want to hold off on that update. Interesting discussions on autonomous agents and ZEH metrics for LLM reliability.

"vibe coding is just declarative programming in English." ...which means we've spent decades building abstractions, only to realize the ultimate abstraction was natural language all along. The shift isn't about tools. It's about how we describe intent. #VibeCoding

GPT-5.2 Pro nearly autonomously solved math problem (Erdős #728). I absolutely love this kind of news. stream-tech.dev/en/media/techn…

Your indie hacker window isn't infinite. There's a sweet spot where energy, risk tolerance, and learning speed collide perfectly. Before mortgages and life complexity stack up. The goal isn't grinding for 40 years. It's going all in during your highest-leverage years.

The shift from "AI copilot" to "AI that learns skills"👀 Google Antigravity, Supabase, Remotion... all publishing "Skills" - packaged know-how AI can apply autonomously. Not prompts. Structured expertise.

📅 Today's digest AI agents are automating development, but friction is growing. Popular OSS projects closing external PRs due to AI-generated spam. The shift: implementation matters less, "guardrails" design matters more. #DevTools #AI stream-tech.dev/en/articles/01…

I caught up on the “The Darwin Incident” at a manga cafe🫡

📅 Today's tech digest GitHub Copilot SDK enters technical preview, Supabase publishes Postgres best practices for AI agents, and Qwen3-TTS goes open source. #BuildInPublic Ref: gihyo.jp/article/2026/0… supabase.com/blog/postgres-…