Letters from Silicon Valley

kept by eddie sep 18

Splitting monorepos into small, bounded services speeds agent-driven development by reducing test wait times and improving code legibility, despite added coordination overhead.

Letters from Silicon Valley: Not Everyone Needs Superintelligence
Most consumer AI agents don't need frontier models; the author's agent runs reliably on cheap "daily-driver" models, with fabrication errors stemming from poor harness design rather than model size.
also on Pepper, Vycari, #software-engineering, #developer-tools
There is an AI code review bubble
The Hacker News discussion argues that the AI code review market is a bubble, with many tools failing to differentiate or provide meaningful value beyond linters.
also on Greptile, Claude, #ai-coding, #software-engineering, #developer-tools
The State of AI Coding 2025
The 2025 AI coding report found developer output up 3.5x and PR size up 79%, with Anthropic SDK closing the gap on OpenAI.
also on Greptile, #ai-coding, #software-engineering, #developer-tools
Codex-maxxing - Jason Liu
The author details how Codex's durable threads, voice input, steering, memory vaults, heartbeats, and goals transform ephemeral AI chats into persistent, self-running workstreams for coding, knowledge work, and even customer service.
also on Obsidian, #automation, #ai-coding, #productivity, #developer-tools
It's Time to Walk
Dictation and agentic AI allow knowledge workers to walk and rest, shifting focus from production to direction.
also on Obsidian, Claude, #productivity