the log
Daily notes on what I shipped, what broke, what surprised me, and what I learned by making things with AI agents.
People keep asking me what AI means for their role. I always struggle to answer, and I finally worked out why: the question asks for a prediction, and AI quietly changed which decision strategy wins.
Every systems thinker has a panel 3 moment. It usually arrives right after the words 'so basically'.
My agent keeps a memory file for every piece of feedback I give it. There are 74. Somewhere in there is the exact shape of the boss I am.
The client publishes by talking to a coding agent. Two agents, a taste file that logs every yes and no, and a handover that taught me what AI-native actually means.
A cleaning supervisor in his 50s, no laptop, never used ChatGPT. The barrier was never his intelligence. It was the keyboard.
Half my time goes to teaching AI now, and I started to worry the building had stopped. So I made my own progress visible. The trail said otherwise.
I sat down with a fuzzy half-thought last night and asked an AI about it. The answer wasn't useful. The artifact was. Here's the loop I think I'm actually running.
I've been struggling to explain AI to people at different levels. The usual analogies fall flat. Here's the messy framework I'm working on.
I gave myself a rule without noticing: only post the good ones. That's not standards. It's a hiding place, and it just got expensive.
17 days of silence. A honeymoon, a Japan road trip, and 19,000 notes I never read. Then my AI agent found a pattern I missed.
Ported Karpathy's autoresearch to my MacBook. While explaining it to me, my AI agent pointed out that I'd been doing leverage diagnosis all along.
Connected an e-commerce client's ad data to the AI AI agent. First analysis found nearly RM 7k burning on zombie products. The CEO could finally see it. And this is just the beginning.
SEO refinements, a keyword placement map, and the realisation that fixing someone else's QA feedback is more valuable than building something new.
Bryan's birthday, one commit, and the difference between skipping a day and almost skipping a day.
My personal site was 5 days old. My company site was 3 days old. I built a third website for my husband's MC business and realised the resistance was never about skill.
Seven days of shipping code. Then I spent a Sunday filling out forms. Turns out community events don't deploy themselves.
I connected two AI tools together and accidentally built the project management system I've been avoiding for years. The trick was removing myself from the equation entirely.
I wrote six sentences on Day 1 about how building and thinking had swapped order. Five days later, I have the receipts.
I had every piece of information I needed. I was still spiralling. Turns out clarity isn't an information problem.
Some days you ship code. Some days you ship thinking. I used an AI to challenge my own assumptions about my business, and two ideas I'd been carrying separately finally clicked.
He's emceeing. I'm deploying code from the audience. This is what cognitive work looks like now.
Finally published the case study I'd been sitting on, built OG images and RSS from a bar, tried to learn SEO, and started telling people what I've been making.
I expected resistance. Instead, she generated more reports than I did, flagged a tracking failure I missed, and asked for more.
Helped the SEO team test AI on a real audit, found a gap between estimated and real search data, and named a dashboard at 3am. Day 3 was two days disguised as one.
21 commits, 3 repos, a client scoring engine, and the personal site I'd been avoiding for three years.
Noticed something weird about how I work now. The thinking happens after the shipping, not before.
Built a full company website and automated ad reporting in a single day. A year of procrastination killed in one session.