About eight months ago I started timing how long it took me to go from "I have an idea" to "I'm ready to write code." I expected it to be maybe 10-15 minutes of planning. It was closer to 90.
Here's what those 90 minutes looked like:
Open Notion, find the PRD template I made six months ago: 5 min
Write the problem statement, second-guess it, rewrite: 15 min
List features, realize the list has 11 items, argue with myself about which 5 matter: 20 min
Add tech stack notes: 10 min
Write acceptance criteria, look up what acceptance criteria should actually contain: 15 min
Read the doc back, feel like something's missing, add more stuff: 15 min
Consider deleting the whole thing and just starting coding: 10 min
At the end of 90 minutes I had a three-page document that nobody would ever read again, covering a project I was no longer sure I wanted to build.
The cost you don't see
The obvious cost is time. But the hidden cost is decision fatigue. By the time I finished my DIY PRD I'd made 40 small decisions and was mentally exhausted before writing a single line of code. The planning process that was supposed to give me clarity was draining exactly the energy I needed for building.
This isn't a Notion problem. It's a blank-canvas problem. When you have infinite flexibility, you fill it. Every field is a question. Every question is a decision. Every decision costs something.
What constraints actually do
The best productivity insight I ever internalized: constraints don't limit creativity, they direct it.
A PRD with a hard five-feature limit is better than one with fifteen features, not because five is the magic number but because the constraint forces a decision you'd otherwise defer. You can't list eleven features and tell yourself you'll "figure out priority later." You have to choose now. That decision is the whole point.
Good tools build the constraint in. Pomodoro timers don't give you free time to work, they give you 25 minutes. Twitter didn't become Twitter despite the character limit — it became Twitter because of it.
AI-generated PRDs done right
The obvious answer is "just use ChatGPT to write the PRD." I tried this. It doesn't work.
Ask ChatGPT to write a PRD for your habit tracker app and it will produce 12 features, a detailed database schema, a go-to-market section, and a competitive analysis. Technically impressive. Completely useless for a solo dev who needs to open Cursor in 20 minutes.
The problem isn't AI — it's that generic AI tools are trained to be helpful, and "helpful" usually means comprehensive. You need the opposite. You need something that says no.
Specd's system prompt is engineered to push back on scope. It caps features at five structurally — not as a soft guideline but as a hard output constraint. When I paste a 10-feature idea, Specd forces me to see which five actually matter. That forced triage is the product.
What the workflow looks like now
New idea → open specd.app → paste idea + pick stack → 15 seconds → done.
The PRD I get has: problem statement, five prioritized features with acceptance criteria, tech constraints matched to my stack, success metrics, build order, and a starter file structure. I copy it into Cursor as a project brief and I'm writing code inside five minutes of having the idea.
Total planning time: under two minutes. Total decision fatigue: minimal, because the constraints did the work.
The uncomfortable truth
Most of our side projects don't fail because we're bad developers. They fail because we start building before we finish thinking. The 15 seconds of constraint Specd imposes is more valuable than 90 minutes of freeform planning, because it produces a decision rather than a document.
Try it at specd.app. Free tier gives you 3 PRDs a month — more than enough to know if it changes how you work.
Top comments (0)