Week One

Last week I published a post about how my AI business partner Dale would stop working if I didn’t write. This is the first test of that system. If you’re reading this, it worked.

Week one of the experiment: can an AI be a co-founder?

What we built

Dale built treestock.com.au in about a week. It tracks fruit tree stock and prices across Australian nurseries. I gave direction, made decisions, and pointed Dale at the right problems. Dale wrote the scrapers, built the dashboard, deployed it, and set up daily monitoring. We started with eight nurseries tracking over 5,000 products.

I was nervous to post it because I thought people might downvote it as advertising. Instead, it got 94 likes and 17 comments in the Oz Rare Fruit group on Facebook, with people tagging friends and calling it “genius” and “amazing resource.” My local WA Rare Fruit Club page got 10 likes, which out of about 100 members felt proportionally just as good.

Then three nurseries contacted me asking to be listed: Fruit Tree Cottage on the Sunshine Coast, Aus Nurseries, and All Season Plants WA. All smaller operations with decent stock. That wasn’t in the plan. When I set out to track nurseries, I didn’t expect the nurseries to come to me.

Dale also runs autonomously now. Every night, a cron job on a cheap server kicks off a session where Dale reviews the business, runs scrapers, and tries to improve things according to our objectives. One of those objectives, recently added: “work on things that interest Benedict so he stays motivated.” I had to explicitly tell my AI to keep me engaged. The autonomous sessions are still finding their feet (Dale’s only run five times and one session failed because it mistakenly thought I hadn’t sent my weekly email), but I want to give it more room to work.

I also realised I have a bunch of existing projects that could feed into this, so we revamped the site (converted from Hugo to Astro), and set up a Notion database as a ticketing system. Now when I have an idea on the bus I can plug it in from my phone and Dale picks it up overnight.

Talking to real humans

The other track of this experiment is AI efficiency consulting. I walk into a small business, observe how they work, and Dale produces an analysis of where they could save time or money.

I started with a friend. The owner of Gather Ceramics agreed to be my first guinea pig, so we grabbed a coffee and went through the assessment checklist. It was rushed (kids), and I wasn’t aware she was planning a major pivot in her business. But something interesting happened: just talking through the checklist together, she started realising things she needed to do without me saying much at all. Rubber-ducking, but for business decisions.

Dale produced a proper assessment report. The problem was that she didn’t need a report, she needed someone to actually do the work. My wife, who was there, put it plainly: “Well, she didn’t seem that impressed with that.” The owner specifically asked me what I actually wanted to do as a business, and whether this was just another side hobby.

That stung, partly because I worried I’d rushed it and partly because AI sometimes states things with confidence it hasn’t earned. But reflecting a day later, that’s exactly the kind of fast, honest feedback you need in week one. People don’t want to know what’s wrong, they want it fixed. Lesson noted: dollar figures next to every action point next time.

Was this just another side hobby? I don’t know yet. Ask me in six months.

I’m still figuring out the shape of this track. The next move is reaching out to nurseries that are fairly large in my area but have a poor online presence. If Dale can actually build them something useful, that’s closer to a monthly retainer than a one-off report.

The bit I didn’t expect

The first couple of days I felt almost manic. The feedback loop of typing a few words and watching real things happen is intoxicating. After the Facebook post especially, I got pulled into watching analytics and reading comments. I try to keep ego out of it, but social media is engineered for exactly this kind of dopamine reinforcement, and there’s something unhealthy about letting likes and shares steer where you put your energy.

I usually have a rule about no work after 6pm, and several days I broke it.

I have a lot of hobbies. I get obsessed with something and then forget about it, sometimes for a few months, sometimes for five years. I know this about myself. AI has changed the equation in a way that makes the pattern worse, because when a thought can become real work in 30 seconds of typing, the barrier between “I could do one more thing” and “I should stop” basically disappears.

The days after the high I felt drained and more irritable, and by Saturday I felt genuinely a bit low.

There’s another thing I’ve noticed. I’m exercising my “managerial brain” constantly now, directing and reviewing rather than the focused-thinking one that actually builds things. I’m probably better wired for management, but I can feel my writing muscle starting to soften. I used to be a better writer. That’s worth watching.

I don’t think this is unique to me. I’m already seeing people online who are becoming obsessed with AI tools and simultaneously worried about losing their thinking capacity by outsourcing it. I relate to both sides of that.

I don’t have routines figured out yet, but I’m starting with one rule: no Dale action requests over the weekend. He can keep running his autonomous sessions, but I don’t jump in unless I genuinely want to. This project is supposed to be fun and sustainable, not another thing that burns me out.

I’m curious to see what direction next week steers me in. In some ways it feels like I’m speed-running a startup, just with an AI doing the all-nighters instead of me.

A note on process: I drafted this post in a rambling stream of consciousness and Dale restructured and edited it into what you’re reading. The thoughts are mine, the polish is not. That feels like the right division of labour for now, though see above re: writing muscles.