I procrastinated on a side project for a year. This planning loop helped me ship it in two months.

For more than a year, I kept saying I wanted to build a side project.

I had ideas, notes, and a long list of things I “should” do—but no product. The goal was so broad that every time I sat down to work, I had to decide what “work on my product” even meant.

Earlier this year, a difficult period at work made me take the side project seriously. I knew I had low energy and procrastinated easily, so I stopped looking for motivation and started looking for a process that made starting easier.

I read about GTD, the five-minute rule, Atomic Habits, and similar approaches. The pieces that made the biggest difference were surprisingly simple:

- Replace vague tasks with concrete actions. “Work on the product” became “write down three user problems for 10 minutes.”

- Put those actions on the calendar. That removed a lot of daily decision-making.

- Keep an action log. I recorded what I actually did, even if it was only five minutes.

- Make progress visible. If I took any action, I marked that day on the calendar.

- Compare the plan with reality at milestones, then adjust rather than treating missed tasks as proof that I had failed.

That loop helped me finish in two months something I had avoided for more than a year.

I turned it into a small SaaS called Actionwise: https://www.actionwise.app

It’s an AI goal planner that helps turn a vague goal into concrete daily actions, schedule them, and then adapt the plan based on what the person actually gets done.

The key product decision was not to treat AI planning as a one-shot answer. Generic AI plans often sounded sensible but were too broad to execute. So I built the product around practical constraints: specific next actions, calendar scheduling, daily action logs, and a visual insights dashboard that compares planned tasks with actual execution, identifies blockers and patterns, and suggests improvements for the next phase.

I’m building it with Next.js, React, Supabase/Postgres, and an LLM layer via OpenRouter.

The biggest lesson so far: a perfect plan is usually a form of procrastination. A small plan that can learn from reality is much more useful.

For other founders: when your plan slips, do you usually rebuild it from scratch, or do you compare it with what you actually did and adjust from there?