I’ve decided to lean into the AI Daily Brief New Year’s Resolution, an accountability framework designed to keep us shipping one AI-centric project a week for ten weeks. As someone who spends my days building Local by WP Engine, I’m usually deep in the flow of the tech-builder’s craft. For Week 1, the goal was simple: build a working web app to track my progress through this 10-weekend sprint.
But as any engineer knows, the “how” is just as important as the “what.” I wanted to see if I could find a workflow that felt like actual engineering—not just a glorified chat interface. I tried two different platforms (Replit, Base44) and a CLI agent (OpenCode).
Friction of no-code Builders
I started where most people do: the high-abstraction platforms. I tried using the same prompts across the board to see how they’d handle the logic of a resolution tracker.
Both eventually hit a wall where they wanted me to “pony up” some cash before I had a usable tool. For a hobbyist project where the goal is *play*, hitting a paywall five minutes in is a massive vibe-killer.
Replit
I set up a project at passionsplay/AI-Progress-Tracker, but the honeymoon phase ended fast. It only gave me about two prompts before the output stalled, and the final product wasn’t exactly something I’d want to ship.
Base44
This one was a bit more generous with the prompt count and even dangled a 30% onboarding coupon in front of me. It did a great job of abstracting away the technical decisions, which is cool for a MVP, but as a dev, I kept wanting to peek under the hood.
OpenCode: Where the Real Work (and Play) Happens
I eventually pivoted to OpenCode, and this is where things got interesting. It required more setup, but the environment felt familiar—more like my actual day-to-day workbench.
Instead of just asking for a “tracker app,” I collaborated with an agent to handle the heavy lifting while I made the “dev-y” decisions. We walked through a legitimate engineering workflow:
- Initializing the Git repository.
- Building out the initial prototype.
- Setting up Playwright-cli so the agent could actually see and interact with the project it was building.
You can check out the final Week 1 result here:
- Repo: https://github.com/bgturner/aidb-new-year–wk01-resolution-tracker
- Live MVP: https://bgturner.github.io/aidb-new-year–wk01-resolution-tracker/
The Takeaway
Was OpenCode + Github more work?
Yep.
It’s possible that I still need to learn to let go of caring about the craft of building and just let the AI figure it out.
On the other hand, the output is a real repo that I actually own and can improve upon. Plus, it was effectively free—though I’m sure my interactions are helping train whatever upstream model they’re using.
Week 1 is in the books. I have a tracker, a repository, and a workflow that doesn’t feel like a toy. Onward to Week 2.