Main pages: SpecKitty (SDD), Spec Driven Development exploration
3 problems I’ve encountered with SDD:
- It feels slow, and less active. Stuff like: what columns did we just create, can also take like 5 seconds.
- You don’t develop muscle memory - since you’re mostly reviewing, retaining what files got created and exact names, is hard.
- You need to hold high level context for longer - sometimes the AI makes mistakes, and during the re-review and re-spec cycle, you might need to hold the context for a large task for more than one day.
Fixes:
- Slow - run multiple agents in parallel. Keeps you busy. Feels more active. Also a GUI could here, as opposed to a CLI like Claude Code.
- Write pseudocode for your story (set of related tasks) on a single page. Everything is a local function be it API calls, cross service calls, webhooks, pubsub. Whatever.
- This helps you quickly know where things are. Like after the lunch break.
- You can count the number of major chunks of logic, and during final review time, see that they match the generated code.
- Holding context longer - break down the task by who they touch, not lines of code. For example - if you’re using the Payments service, ensure all tasks related to it are in one huge task. Also, don’t postpone review, and quickly review each task as its done - this frees you up to use your personal/human context for other tasks.
Still learning, will keep things updated. Follow this Notion page for updates.
SDD