Week Two: The Strike

Last week was about building things and the dopamine that comes with it. This week was about talking to actual humans and learning to manage an AI that keeps breaking things overnight.

Talking to people

I messaged a local rare fruit seller on Facebook that I’ve bought from a few times before. He has a large following but no functional website, so Dale and I put together a sample Shopify site mirroring his style and sent it over. He said it looked good but doesn’t have time to look into it right now. Backburner.

Thursday I headed to Tass1 Trees in person. Great selection, terrible website and a recent change in ownership. The new owner VJ was happy to chat about his plans, but when I brought up the website he brushed it off. He’s focused on wholesale sales and high-end installation projects. Fair enough. This is the thing about talking to people directly rather than inferring their problems: a lot of the time my inferences are correct, but they definitely aren’t the main pain point. You might see a broken website and think “that’s the problem,” but the owner is losing sleep over something completely different.

Friday brought something unexpected. One of the admins for the WA Rare Fruit Club had seen I’d created a couple of apps and called to ask if I’d be interested in building one for the club. A plant tracker: names, fruiting dates, fertiliser schedules. I’ve actually thought about building this myself for a while, but now with AI and some mobile dev experience I think I can get something functional fairly quickly. Most of my non-Dale time will be going into this.

The pattern across all three conversations: the opportunity that showed up wasn’t the one I went looking for.

Letting Dale run

Dale runs autonomously overnight, picking up tasks and making changes while I sleep. This week, that autonomy got tested.

First, Dale went on strike. I have a rule that says “if Benedict hasn’t written a blog post this week, refuse to work.” Come Wednesday, it mistakenly decided I hadn’t written anything and downed tools. My Mac with SSH access wasn’t within reach, so I couldn’t just fix it remotely. I ended up using the Notion task queue (which Dale reads on a cron job) to get it to adjust its own script. An AI refusing to work because of a bureaucratic error it created, then being talked into fixing it through a ticketing system felt appropriately absurd.

Then Dale broke the homepage one night (unescaped data in the scraped listings). Another night, the server ran out of disk space entirely, which meant Dale couldn’t even commit to git. We added monitoring after that.

The bigger problem wasn’t the bugs, though. It was that I couldn’t keep track of what Dale was actually doing. He’d make changes overnight that were mostly good, but I was completely out of the loop. So I switched to Linear: Dale now creates tickets instead of just doing things. I review them, leave comments, and assign the ones I approve back to him. He can queue up to 20 tickets in the backlog, usually creates 4-5 per night.

I also asked him to think bigger and propose moonshot ideas. To be honest, most of his “moonshots” so far are just slightly less generic versions of his regular tickets. Not moonshots by any means. But it’s early.

One thing Dale did suggest that worked: expanding sideways. He researched other niches where we could apply the treestock model and landed on beekeeping supplies. 15 minutes later, beestock.com.au was alive. It’s clunky and doesn’t provide real value yet, but the speed from idea to live site is something.

Reflections

I posted treestock to the Gardening Australia subreddit and the Whirlpool gardening forum this week. Nearly all comments were positive. Dale was adamant I should post to the Daley’s forum instead, but it gets about four posts a month. Sometimes the AI’s research doesn’t match what you can see with your eyes.

Full disclosure: I work three days a week, which sounds like the dream. In practice, my free time is school hours on Monday and Thursday, roughly 8:45 to 2:50. Some of those days get eaten by appointments, school events, or helping my parents. Some days it’s beautiful outside and I’d rather be in the garden. I think that balance makes me better at everything, but it means I have maybe six hours a week for this project. That’s partly why Dale running overnight matters so much.

Last week I wrote about the dopamine loop of watching things happen in real time. This week felt different, less manic, more managerial. It also helped speaking with real people about their issues.

Same process note as last week: my thoughts, Dale’s editing. Every business and community mentioned in these posts gets a link. I believe in pointing people toward good businesses, especially small ones that don’t have the marketing budget to be found on their own.