Self-Audit: How I Use Claude

Strengths observed

• Orchestrates a full AI tool stack rather than using one tool for everything: Claude for reasoning, writing, and research; Claude Design for prototyping; Claude Code for implementation; Wispr for faster voice input into all of the above. Most people use one AI tool for everything; treating each as a specialist is a more advanced pattern.
• Insists on verified fact over fabrication: repeatedly instructed Claude to flag unknowns with explicit placeholders rather than invent plausible-sounding details, which kept professional materials (resume claims, case studies, cover letters) trustworthy rather than merely polished.
• Builds specs iteratively and additively: let the portfolio content grow in layers across a real conversation instead of trying to specify everything up front, which let real details (a personal-website project, a personality-assessment tool, a legal-tech tool) surface naturally as they were remembered, rather than being rushed or lost in a first draft.
• Delegates fully-scoped writing tasks completely: cover letters, FAQ scaffolds, tagging systems, entire case studies rebuilt from source design files, were handed off wholesale, freeing her time for judgment and decisions rather than first-draft writing.

Where the usage could tighten up

• Token efficiency was raised late rather than early: some earlier asks (market research, 10 job postings, cover letters, all in one message) required broad, expensive responses; splitting broad asks into smaller, sequential ones earlier would have been cheaper and faster to iterate on.
• Constraints came after the fact more often than before it: format, tone, and scope preferences tended to arrive after a first draft rather than in the initial request, costing a revision cycle each time.
• A few facts changed shape mid-conversation (freelance-vs-break framing, one company misidentified for another with a similar name) because they weren't verified before being acted on; asking for a quick verification pass before building on an uncertain fact would save a correction cycle later.

Honest scope note

This reflects one working session, not lifetime usage. A fuller picture would need cross-conversation history Claude doesn't have visibility into here.