#Coding-Agents
11 notas · última:
Goodbye to the Hard Parts That Never Mattered
The author argues that coding agents remove toil, not engineering. They shift work toward judgment, verification, responsibility, and broader creative exploration, while raising real concerns about access, learning, and scaled mistakes.
Among the Agents
The author argues that coding agents have become capable enough to do substantial real work, but the real shift is using them well. He sketches practical limits, risks, and likely effects on software, data, jobs, policy, and growth.
The Mythical Agent-Month — Wes McKinney
Wes McKinney argues that coding agents reduce accidental complexity and speed up code generation, but they also amplify scope creep, coordination problems, and bloated codebases. Design, taste, and expert human judgment remain the bottlenecks.
Your job is to deliver code you have proven to work
Simon Willison argues that AI-assisted development still requires proving changes work through manual testing and automated tests, with humans accountable for the result and coding agents helping only if they can verify their own output.
How I run 4–8 parallel coding agents with tmux and Markdown specs
The author describes a manual system for coordinating 4–8 coding agents with tmux, Markdown feature designs, and slash commands that track planning, implementation, verification, and archive stages.
Implementing a secure sandbox for local agents
Cursor describes a sandbox for coding agents that cuts approval interruptions, preserves security, and adapts to macOS, Linux, and Windows limits while teaching agents when escalation is needed.
Software is about to eat the world much faster
Marc Andreessen argues that software is moving from human-speed creation to compute-speed creation through coding agents, which is raising engineer leverage and accelerating automation across industries.
Engineering Management in the Age of Agents
The article argues that even if coding agents reach mid-level engineering ability, engineering managers will still be needed to connect business and technical needs, define solutions, and handle people issues.
LLM predictions for 2026, shared with Oxide and Friends
Simon Willison shares predictions about coding agents, sandboxing, security, a browser built with AI help, and code typing fading as software work shifts away from manual syntax entry.
The disappearing middle of software work
The post argues that coding agents are shrinking the manual middle of software work. The main craft shifts toward forming clear intent, shaping context, and improving review and release workflows.
Superpowers: How I'm using coding agents in October 2025
The post describes Superpowers, a Claude Code plugin system built from skills, subagent workflows, TDD, worktrees, and pressure-tested instructions that teach agents to search and use skills.