A few things became very clear to me while building dougantin.ai.
First, a lot of the intimidation around technical work is fake. I used GitHub, Vercel, Cloudflare, APIs, cron jobs, Redis, environment variables, and production debugging for the first time on this project. None of it was effortless, but very little of it was actually inaccessible. Most of the friction came from unfamiliarity, not impossibility.
Second, building something live forces a different kind of thinking than just having opinions about it. I had to work through broken deployments, DNS confusion, API plan limitations, stale cache issues, runtime errors, schema changes, and copy decisions in real time. That process was messy, but it made the site better and made my thinking better. The work stopped being abstract the moment it had to survive contact with production.
Third, iteration and speed to market beat overplanning. The tracker did not arrive fully formed and is not in its final form. It started as an MVP, then got better through actual use: better data, better structure, better framing, better taxonomy, better metrics, and a better understanding of what was useful versus what only looked sophisticated. Importantly, I learned what worked and what didn't through the act of doing it live. That loop mattered more than getting it perfect up front.
The broader takeaway is the same conviction I keep returning to, and the same one I just published: You Really Can Just Do Things. AI does not remove the need for judgment, but it radically lowers the cost of turning intent into execution. The barrier is less often capability than willingness: willingness to start, to break things, to learn publicly, and to keep moving.