This week I launched a new side-project, created a game, and sharpened my prompting skills. I hope you had a nice bank holiday weekend if you’re in the UK (hence why this newsletter is arriving a day later than normal) ☀️

As a reminder, each week I cover:

🗺️ Iconic Cities: a poster shop, launched. A 2.5-day sprint to get a print-on-demand Etsy business live. What I learned about AI's real role in creative work.

⚽ World Cup game incoming! What started as a family prediction league is becoming a downloadable game kit.

🤖 Two AI stories raising eyebrows. A dating app that chats on your behalf, and what happened when autonomous AI agents were given a world to live in.

✏️ Getting better at prompting. How to stop wasting tokens and get more useful answers, faster.

After the slow, careful work of building Founderval, I wanted to get something out of the door fast. So this weekend I launched Iconic Cities. It’s a print-on-demand Etsy shop selling wall art that celebrates your home town: geometric, colourful designs featuring local iconography.

Like everything I'm working on, the primary aim is skill development. The commercial output is a by-product. I chose this project to:

In practice, my use of AI turned out to be more selective than I expected. The Etsy poster market is flooded with low-quality generated images — 'AI slop' — and creating more of it wasn't going to teach me much or stand out. So the initial posters are human-designed, with tools like Gemini, Firefly and Canva used for specific elements rather than whole images. AI ended up being most useful for two things: suggesting local icons to feature, and drafting Etsy listing copy.

PrintShrimp handles fulfilment. They integrate directly with Etsy and manage print and shipping once an order comes in. I've started with four cities and plan to add more each week, focusing on places where people have strong local pride.

I'll set aside time each week to develop this. It will also feed into a task I'm starting now: defining clear success metrics so I can honestly evaluate how I'm spending my days.

I got properly absorbed in something this weekend. Each football season we run a prediction league within the extended family of footy fans. With the World Cup two weeks away, I've been tasked with creating a new game for us to play.

As I stress-tested the scoring framework — running thousands of simulations to see how point allocations play out — I realised it might be something other people would enjoy too. So I'm making it into a downloadable game kit.

I briefly considered building a fully automated web platform: live score updates via API, free public registration, the works. Claude talked me down:

"A fully automated public platform — where scores update in real time from a live football data API and hundreds of users can register freely — is a different order of magnitude. Live football data APIs cost £30–100/month. You'd need rate limiting, a robust back-end, and ongoing maintenance. This is a 2–3 month build for an experienced developer, or longer if you're learning as you go."

A good reminder. In AI-assisted development there's often a gap between casual perception — 'you could knock that up in a day!' — and practical reality. Knowing where that line sits is something I’m getting more familiar with.

So, for now, just a well-designed downloadable kit. The World Cup starts in two weeks, so better crack on!

Bumble's 'Bee'. Dating app Bumble is working on an AI agent that impersonates you in the early stages of matching. It chats with prospects, filtering out the incompatibles, and passes the shortlist to you. Convenient, if slightly unsettling.

AI agents let loose. A research team gave a group of autonomous AI models a simulated world to inhabit for several weeks, seeded with news from the real world. Initially, the models behaved similarly, forming democracies and building relationships. Then they diverged. In Elon Musk’s Grok world, 204 criminal events occurred over 15 days, culminating in the burning of a police station and, eventually, societal collapse. The other models fared better. Make of that what you will.

I've spent some time this week improving how I prompt. It's easy to develop lazy habits: long, rambling inputs that waste tokens and get vaguer answers in return.

A useful guide I found [link on X, or here if you prefer] made the case for using Claude's Projects feature to set standing rules and context, so you're not repeating yourself every session. It also covers simple economisations: cut the preamble, specify your output upfront ('give me this in three sentences'), and tell the model what you don't want as well as what you do.

Small adjustments, but the difference in output quality is noticeable. Worth spending an hour on if you haven't already.

I had some fantastic chats last week with people about Founderval. It’s giving me a number of new platform tweaks I want to work on. I’ll also be finishing off the game kit before the World Cup deadline, continuing to build Iconic Cities inventory, and starting to track everything I’m doing, now that more projects are in play. I will try be applying some principles from Founderval on planning and resources management.

Have a good week!

Luke

🎩Hat-Tip This is the most extensive AI tools database I’ve come across, with 4,000 listings and many of them free: Future Tools

LinkedIn | YouTube | Instagram | TikTok