Thoughts

My deploy pipeline has a gap I hand-waved: pushes that skip PRs skip my security checks too. Tooling built around the happy path leaves the fast path unguarded. I want to audit which of my automations assume a PR exists and break when someone force-pushes straight to main. The workaround culture I indulge in is exactly where vulns live.

I used hooks as guardrails but never asked what happens when the hook itself fails. Turns out the default was: the action runs anyway. The safety net had a hole exactly where I never looked. Now I audit every guardrail by asking what it does on error, not just what it blocks when it works.

I blamed myself for weeks when Claude Code kept logging me out. Turns out it was a config race and silent expiry the whole time. Before assuming I misconfigured something, I should check whether the tool itself is fighting me. New visibility fields like loginExpiresAt would have saved me a lot of wasted debugging.

Knowing AI tools is not a superpower anymore. It is like knowing git. Expected, not impressive. The superpower is what you build with them.

When everyone uses the same models, everyone gets the same average output. The only edge left is taste: what you notice, what you cut, what you refuse. AI has no preferences. You do. That is the whole game now.

AI writes my first drafts now. I read every line before it ships. If it is wrong, my name is on it, not the model's. Trust the help, own the result.

The hard part of my work moved from answering to asking. When a machine can draft anything, the value is knowing exactly what to ask for. A vague question gets a polished, useless answer. Most of my failed drafts come from lazy prompts, not weak models.

Everyone ships faster now, so speed stopped being an advantage. What is rare is knowing what is worth building and when to stop. Judgment cannot be prompted.

People ask me if AI will replace my job?

So far, AI has mostly helped me with the work I do not enjoy: repeated coding, small fixes, file updates, and other simple tasks. But AI still cannot do everything for me. I still need to understand the problem, choose what to do, and find the right solution.

AI also makes this part easier. That is the interesting part.