I’ve been looking for something that sits between heavy specification frameworks and just diving straight into code with coding agents. My experience with tools like SpecKit has been useful, but I haven’t naturally gone back to using it again. When Thariq from Anthropic posted about using Claude Code’s “ask user question” tool to build out specifications, it seemed worth exploring.
The approach is straightforward: start with a minimal spec, sometimes just a single sentence describing the tool you want to build, and let Claude ask clarifying questions. You answer, it refines, and you end up with a detailed specification covering sync strategies, file formats, deletion handling, and all the other considerations you’d normally think through yourself if you had the time.
For this session, my spec was simply “a command line tool to mirror your Microsoft 365 mailbox to local files.” What followed was dozens of questions about incremental sync strategies, what to do when Microsoft’s delta sync tokens expire, how to handle initial full synchronization, folder structure decisions, file naming conventions, and more. The depth of questioning matched what you’d normally work through in your head, but the difference is you wouldn’t typically do it upfront at value proposition stage due to time (and management 😉) pressure.
This was my first time using the Claude Windows app rather than the Claude Code CLI, and the lack of familiarity shows in the video. The timed multiple-choice interface caught me off guard. Questions would rotate through options, and I’d find myself thinking “come back, I wasn’t finished reading that”. The timeout pressure felt counterproductive when the whole point is thoughtful consideration of design decisions.
I’ve subsequently found a work around by clicking on the Type something else... option. The timer for selection is stopped and I can choose a multiple choice option or type something else at my leisure. Its unclear if Anthropic will change the UI in the future to be less pressure inducing.
What worked well was how the approach made early design thinking feel worthwhile. Disk space exhaustion mid-sync, timezone handling, stripping external image tags for security, shared mailbox support—these are considerations that normally surface during implementation rather than design. Having them prompted upfront, by something other than my own recall, made the case for thorough specification more compelling than it usually feels under delivery pressure.
The session produced a detailed specification file covering message storage format, sync state management, configuration options, and command-line interface design. Brief specifications, but comprehensive in coverage. The work tree feature automatically branched the changes, which was convenient for reviewing before merging to main.
This isn’t a video for AI skeptics. It’s for curious professionals who want to understand how these tools work in practice, business and management folks wondering about practical AI applications, people who don’t use coding agents daily but want to see the concepts demonstrated. The video is long, boring and unpolished because that’s the authentic developer experience. Fast-forward at 2x (or more!) if you want the gist.
The ask user question approach sits in a useful middle ground. It’s lighter than full specification frameworks but more structured than ad-hoc development. For building out design thinking quickly, it’s a valuable addition to my thinking toolkit.