If you can no longer tell whether ChatGPT or a human wrote a piece of text, what does that do to authenticity and trust in what we read?

Microsoft wired ChatGPT into Bing and used the waitlist to push Bing onto millions of devices — pure marketing genius in my book.

When DeepSeek dropped R1, my first reaction was not hype — it was spreadsheet panic. The comfortable pricing story suddenly had a counterexample.

Claude 3.7 Sonnet was the first time I caught myself delegating context, not lines. AI coding assistants stopped pretending to be autocomplete.

Agents fill the merge queue faster than humans can read. The constraint moved from typing to verification — and most teams still review like nothing changed.

Three years of AI-assisted development later, vibe coding is real. The thinking is yours, the judgment is yours — the typing was never the point.

The single-agent model has a ceiling. Claude Code's parallel agents let you split a feature into independent pieces and work like a focused team.

How we merged three Ionic/Angular/Capacitor codebases into a single Nx workspace so a small team could stop maintaining the same code in three places.

You cannot prompt AI-generated code into being reliable. What works is what always worked with new hires — guardrails that say no.

I stopped pointing the frontier model at my backlog and started using it to write the Skills and guidelines my cheaper models read.

Put shared project rules in AGENTS.md, keep a thin CLAUDE.md that imports it, and leave Skills for verbs — so Cursor, Claude Code, and Codex stop drifting apart.

I finally built an Obsidian second brain that stuck — by writing down what my AI agents are and aren't allowed to put in it.