So He Asked Claude Cowork to Build His Dream App

A personal “second brain” knowledge system can be turned into a standalone CBT (cognitive behavioral therapy) logging app surprisingly fast—using Claude Cowork to convert saved mental-health materials into step-by-step daily workflows, then generating a PWA-style desktop app that runs offline. The payoff is practical: instead of manually searching Notion, extracting book highlights, and calculating scores, the app guides mood logging, captures automatic thoughts, links them to cognitive distortions, and tracks BDI (Beck Depression Inventory) results over time.

The workflow starts with a structured knowledge base in Notion and mirrored folders on a computer and cloud drive. The creator focuses on mental well-being, pulling in a PDF of David D. Burns’ Feeling Good and exporting only the highlighted, distilled parts into a project folder as Markdown. That distilled material still contains too much raw information for daily use, so Claude Cowork is asked to “get familiar” with the sources and then generate a step-by-step daily guide for CBT tasks—specifically for registering mood, running the cognitive distortions process, and completing the BDI assessment.

The generated guide becomes the blueprint for the app’s features. The CBT workflow includes a mood check-in, a BDI assessment with 21 questions and scoring, and additional tools for procrastination and thought-challenging. It also emphasizes that thoughts are what can be challenged and changed, while feelings follow; it suggests weekly tracking of BDI trends and daily routines like catching and logging negative thoughts, using the “Tik Tok” technique, and reviewing activity schedules and “triple column” entries.

Next comes app design for a non-programmer. The requirements are clear: a dedicated desktop app with a dashboard, export options (Markdown and table formats), and automatic prompts to log mood and walk through CBT steps. The creator initially mentions MacOS and later wants phone support, but clarifies they use Google Pixel rather than iPhone. Claude Cowork proposes a progressive web app approach: installable via Chrome, offline-capable through service workers, and storing data locally in an indexed database.

A local server test brings up the app in Chrome with an install option. The first usability pass reveals friction points: the interface doesn’t adapt well to window size, the mood logging button is oversized, and the “setup reminders” control is missing inside the app (it’s instead a separate command/file in the project folder that opens a terminal to set reminder intervals). Functional testing also surfaces feature bugs—like the “Tik Tok” activity log opening a blank page and the need to relaunch the app.

The creator records screen feedback via Zoom Clips, exports the transcript, and feeds it back into Claude Cowork for targeted fixes. After Claude rewrites the app (including CSS changes), the mood logging flow improves: automatic thoughts and cognitive distortions become visible and selectable, entries save correctly, and history/export reflects detailed records. The result is a usable CBT mood-logging app built in roughly 2.5 hours, with reminders and some secondary features still untested—but the core loop (log mood → capture thoughts → apply distortions → complete BDI → review trends) works well enough to call the experiment successful.