How I Code in 2026

  1. #1: Introduction

    2026 is here and programming isn't what it used to be. This post kicks off a series where I document my current setup with Claude Code, from the paradigm shift that code agents represent to how to configure the CLAUDE.md file so the agent actually understands your project.

  2. #2: GitHub Integration

    In this post I show how I've set up my entire development pipeline using GitHub from the terminal: worktrees to work on multiple issues without them stepping on each other, scripts that automate branch creation, and how Claude Code reads issues, resolves conflicts in interactive rebases, and even reviews GitHub Actions errors to fix them on its own.

  3. #3: Specification Management

    In this post I explore how I changed my approach to planning before coding: from simple one-liners to full specs with Speckit for large features. I also discovered that markdown files ended up being the best interface for working with Claude Code, a kind of knowledge cache I can reuse between sessions.

  4. #4: Agents, Skills and Commands

    Commands, skills, and agents are the tools that will transform you from a casual user to a Claude Code power user. In this post I explain the differences between each one, how I use /explore to navigate large codebases, and how parallelization with subagents saves me a ton of time when fixing linter errors across multiple files.

  5. #5: Code Review

    How I organize code review when working with code agents, using a swarm of specialized agents that automatically review performance, security, and architecture, plus a manual validation layer in GitHub. Also why pre-commit hooks are key to not breaking everything before pushing code, and how to prevent Claude from being sneaky with --no-verify.

  6. #6: MCP & CLI

    MCP and CLI are the two ways to give Claude Code superpowers to do things a language model alone cannot. In this post I explain when to use each one, with practical examples like the GitHub CLI and a somewhat crazy experiment I'm working on: a game engine that exposes its own MCP tools so Claude can inspect the game while it's running.

  7. #7: Next Steps

    Now that I can explore projects I wouldn't have even started before, here's how I built a spaced repetitions app in an afternoon (something that would have taken me months), the two bigger projects I'm working on, and a reflection on what it means to stop programming directly at 40.

  8. #8: Postscript - UI, Limits, Fears

    Some things that didn't make it into the series: using ASCII to design interfaces without distractions, the real limits of code agents (security, visual, temporality), and the elephant in the room: what happens to our brains when we stop thinking.