Came across MySpec after chatting with Nancy on Product Hunt during a launch week thread, and the whole "context, not the model, is the real bottleneck" angle immediately hit for me. I've been in enough Claude Code and Cursor sessions that went sideways because the AI was quietly guessing at architecture I hadn't defined. Tried the open beta over the past week and honestly, the concept clicks fast: you describe your idea, and MySpec interviews you like a senior architect until it has enough context to spit out a real spec bundle.
What impressed me most is the four-file output. Instead of another blank markdown template, you get a structured Spec Bundle that Cursor or Claude Code can actually consume as a source of truth. The live Mermaid diagram rendering is a nice bonus, you see flows and data models take shape while you're refining. The "Vibe Specify" chat editing where you just say "make the API more secure" and watch the diff land is genuinely fun to use.
Where it currently feels early: the interview flow can occasionally ask questions that overlap or feel redundant, especially for smaller projects where you don't need the full architect treatment. The output is markdown-heavy, which is great for AI agents but a bit dense for non-technical stakeholders trying to align on scope. Also no CLI sync yet, though the team says it's coming in Q2. But for the price during open beta (free with unlimited AI), it's a legit way to stop the prompt-loop chaos and start giving your coding agents better inputs.
