A year of shipping AI
It's been a year since the last post here. Ten projects later, almost all of them have AI somewhere in the stack. Here's what changed.
It’s been about a year since I last posted anything here. The last post was about an MCP server I’d built over a weekend for libvips, mostly because I wanted to play with the protocol and get some practice with TypeScript. At the time MCP was still new, the Cursor and Windsurf pricing wars were heating up, and I was mostly a curious observer with a side project habit.
Reading that post back now, it’s wild how much has changed. Not just in the AI world generally but in what I’ve actually been doing with it.
Since then I’ve shipped around ten projects, and almost all of them have AI somewhere in the stack. ReceptionIQ is a voice AI receptionist that answers calls and books appointments for small businesses that don’t want to hire a receptionist, and RidingDesk is a CRM for Canadian political campaigns that doesn’t ship donor data to US servers, which it turns out is a real concern for anyone running a campaign up here. DealPortal is a multi-tenant SaaS for pre-construction real estate sales, and Instant Expert turns the content already on a website into a chatbot that also captures leads.
On top of that there’s AlwaysAI, which is a cloud-backed desktop you boot from a USB stick with AI built in, plus a couple more MCP servers, one for video editing with Editly and an updated version of the libvips one building on that original weekend project. There’s also a Canadian tech talent marketplace, a children’s storybook video tool, and a WordPress media protection plugin in the mix. Each has its own story, but the through line is that I went from playing with AI to actually deploying it, in production, for things people pay for.
On the tooling side, the last post had me bouncing between Cursor and Windsurf, and I tried both for a few months each. Cursor felt like the polished one, Windsurf felt like the more interesting one, and I went back and forth depending on what I was building. Then Claude Code showed up and I more or less stopped switching, something about the terminal-first workflow, the way it actually plans work before doing it, and the fact that it hands you back a real diff at the end. It’s been my main tool for the last six or seven months and I haven’t really looked back.
The bigger change isn’t the tools though, it’s the pace at which everything is moving. A year ago an MCP server was a weekend project I’d write up for fun, and now I can stand up a working voice agent or a multi-tenant SaaS in a few weeks, end to end, alone. That used to be a team’s job, and I don’t think most people outside of this world have fully clocked how much that has shifted.
What I have noticed though is how many businesses are getting this wrong. Companies pouring money into AI pilots that never make it past a demo, or hiring expensive consultants to write deck after deck about “AI strategy” while nothing actually ships, or worse, ignoring the whole space and assuming it’ll sort itself out later. Every time I talk to someone running an actual business I find myself thinking I could probably help with this, in a way that ends with something working in production rather than a PDF in a shared drive somewhere.
Anyway, I’ll try to post here more often than once a year. We’ll see.