How one consultant cut 20 hours of weekly admin with Claude
Grace Clarke says she now runs her service business on tools she built herself: an hourly pipeline, a proposal maker and a replacement inbox. The method is copyable; the price tag is not stated.
Grace Clarke is an AI educator and a former marketing consultant. She taught herself Claude Code earlier this year and turned the process into a curriculum.
On her own account, she has replaced about 20 hours of weekly admin — proposals, client tracking and email — with tools she built rather than bought.
The first is what she calls a pipeline operator. It runs every hour and moves clients from one stage of her process to the next without her opening anything.
The second is a proposal maker. She abandoned conventional proposal documents in favour of password-protected interactive HTML pages generated in Claude.
The third is the one that will make most readers stop: a custom inbox that replaced Gmail entirely. She says she built it in under 30 minutes.
The mechanism holding it together is a set of skill files. One is a voice guide, which she describes as teaching the model how she thinks rather than how she writes, so the output does not read as generic AI text. Others hold proposal rules and versioning.
She started the build in Claude Code, then handed the work to Cowork by passing a Markdown session file between them. The supporting parts are ordinary: Google Forms and Sheets for intake and records, Google Cloud service accounts for the connectors, Netlify to host what gets published.
For clients who are not technical, she teaches a two-step forcing function. The point is not the output. The point is building the habit of opening the tool at all.
She also rejects prompt engineering as the useful skill and teaches what she calls intent engineering — being precise about what you actually want before you ask for it.
The habit extends past work. She uses Claude on her phone on walks, to track workouts and to manage plants, which is either trivial or the whole point, depending on how much you believe repetition matters.
Two caveats a founder should hold onto. The 20-hour saving is her own figure, measured against her own previous working week, and no independent check on it exists. And nothing here puts a number on what the stack costs to run each month — not the model subscription, not the Google Cloud usage, not the hosting.
The transferable part is the sequence, and it is cheap to test. Pick the single most repetitive administrative task in your week, write down the rules you already follow by instinct, and see whether a tool can be made to follow them for an hour at a time. That is a weekend, not a transformation programme.