grep -l "#ai" posts/*
#ai
75 posts · all posts
- Herdr, seven months in
- Stack Overflow, under a thousand
- The Programmer Competency Matrix, 2026 edition
- I rebuilt this site with an agent
- What 10x actually looks like
- Book review: Kill It with Fire
- The cron job that ran every minute
- Sandboxing agents
- My Obsidian vault as an agent's memory
- Teaching juniors when the agent knows more
- Evals for my own agents
- The agent loop, visually
- Our model provider had a security incident
- SICP, reread
- Don't be a meat proxy
- Bespoke software is cheap now
- Writing specs for agents
- Mixture of experts, visually
- Hiring engineers when agents do the typing
- Show human effort
- My personal agent
- Neovim in 2026: rebuilt with an agent
- Book review: Co-Intelligence
- The prompt injection in a support ticket
- Local models are good now
- KV caches and why context is expensive
- Code review when agents write the code
- Speculative decoding, visually
- A team of agents in my terminal
- Book review: AI Engineering
- Agents need accountability
- Agents talking to agents
- Should my agents get their own computer?
- Where did Stack Overflow's questions go?
- What I'm betting on in 2026
- 2025: the year my team stopped writing boilerplate
- How LLM tool calling actually works
- Picking a model per task
- Sub-agents and parallel work
- Why I'm still hiring juniors
- GPT-5 and choosing models on evals, not launches
- Pointing an agent at my notes
- Agentic IDEs compared, from the terminal side
- Software in the era of AI, for teams
- To my AI-skeptic friends
- Giving an agent an hour and a test suite
- I'd rather read the prompt
- The support bot that invented a policy
- A week with Claude Code
- DeepSeek-R1, and what a cheap open reasoning model changes
- The Model Context Protocol
- Computer use and the first real agents
- Reasoning models
- Is open-weights AI the path forward?
- Vector embeddings, visually
- An agent that files my expenses
- Diffusion models, visually
- 2023: how AI changed my team's year
- Retrieval-augmented generation, visually
- The OpenAI weekend and AI vendor risk
- Llama 2 on my laptop
- "No moat": open models catch up
- Writing our AI usage policy
- GPT-4 and my moving skepticism
- Chatbots with a persona problem
- The transformer, visually: building a tiny GPT
- A week with ChatGPT
- Copilot, a year in
- Whisper: transcription that works
- DALL·E 2 and images from nowhere
- GitHub Copilot, first week with an AI pair programmer
- GPT-3 and few-shot learning
- GPT-2 and "too dangerous to release"
- Monte Carlo tree search, visually
- Book review: Superintelligence