First, the Cowork setup. We'll install and configure the desktop app, point it at a real working folder so Claude reads and writes files directly instead of taking uploads, connect the tools the work already lives in, install the skills and plugins that match the job, tune permissions so a task can run to completion without a confirmation at every step, and give Claude persistent memory so it stops relearning the firm's context every morning. Then we'll run one multi-step task end to end and watch it produce actual files rather than a wall of chat text.

Second, the Terminal setup. We'll get Claude Code installed and authenticated, walk through what the permission modes actually do and which one to work in day to day, put a project instructions file in place so Claude arrives with your context already loaded, and run a comparable task from the command line. We'll close on the decision rule: which surface handles which kind of work, and the specific signals that tell you you're in the wrong one.

Claude subscribers who have gotten comfortable with chat and hit its ceiling. The ones re-uploading the same rent roll every session, copying output into documents by hand, and wondering why the assistant can't see the folder it just wrote about.

The Cowork half assumes no technical background at all. The Terminal half assumes willingness, not experience.