Claude Code Use Cases for Business: What Our 66 Commands Actually Do

The Claude Code use cases that pay off for a business are the ones where documents go in and documents come out. These are the ones that earn their keep at our company:

None of them ask the person running them to write code.

Our company runs out of one git repository I call our company brain. On October 2 its .claude/commands/ folder held 66 command files. A command is a markdown file of plain-English instructions that you run by typing its name, and Anthropic’s skills documentation explains how a file in a subfolder like sales/ becomes /sales:build-proposal. The folders follow our departments, so the inventory doubles as a map of where the work is.

Sales has 21, marketing 13, strategy 12, finance 6, project management 4, our Notion helpers 3 and ops 1. Six more sit at the top level for setup and housekeeping. Next to them are 69 skills, the background knowledge commands load, and 24 of those are generic (Word files, PDFs, writing checks). Anthropic’s own overview lists use cases like writing tests and resolving merge conflicts. Most of ours have nothing to do with software. Whether people who don’t code can run them is its own question, and Claude Code for non-developers answers it with our setup and what we took off their plate.

Sales: building a proposal from the client folder

Eight of the 21 sales commands form one chain. /sales:start-project creates a client folder and asks you to drop in raw notes and transcripts. /sales:synthesize turns those into organized docs. Then come a solution document, a gap check, a t-shirt-sized estimate, the proposal, the engagement letter and the invoice details. Each step reads the files the previous one wrote.

/sales:build-proposal is the middle of that chain. It refuses to start if the solution document or the estimate is missing, and tells you which command to run first. It fills our proposal template from the folder and aims for four to six pages. Before it writes a word about pricing, it reads our log of active business rulings, so a proposal can’t offer a payment structure we’ve already decided against. (Why we keep rulings in a file the AI reads is in Write Down the Rulings.)

On October 2 the repo held 40 proposals across 38 engagement folders. The command file has been edited 26 times since January 3. My co-founder and I run it, since we own sales.

The limit: the instructions are 993 lines. That’s what it took to get proposals we’d send without rewriting, and it means nobody can hold the whole thing in their head. A change to one section can quietly contradict another.

Project management: the weekly client update

/pm:weekly-update reads our project list in Notion, finds every active client project and groups them so each client gets one email. Then it stops and waits. Whoever runs it (our client-success lead owns this one) pastes in each engineer’s raw update. It keeps their words close to verbatim and ties each bullet to the pasted text it came from. If nobody sent an update for a project, it falls back to gathering evidence itself from tasks, commits, meeting transcripts and email, and every bullet still has to cite a source.

Our client folders hold 63 weekly update emails since May 1, across 16 clients. The command has been edited 26 times since February. It’s the one project management command that survived when we deleted 52 skills and commands, which is also where the stuff that didn’t work is buried. If you manage projects for a living, Claude Code for project managers goes deeper.

The limit: it’s only as good as what engineers hand over. The automated gather was meant to replace pasting, and it’s still the fallback because it can’t tell what mattered this week.

Strategy: the end of the day

/strategy:day-end-process is the most-edited command we have: 39 commits between March 31 and October 1, and 1,092 lines today. Each founder runs it at the end of the working day. It reads the day’s meetings and email, files what matters into client folders, creates and updates tasks, and lays out tomorrow. I’ve written about the tasks it created and two founders sharing one brain.

The limit: it’s the command most likely to break, because it touches email, calendar, Notion and the repo in one run. Being the most-edited command is not a compliment.

Finance and ops: reconciliation and timesheets

/finance:reconcile takes a month, pulls the bank ledger plus our invoice, contractor payment and time records from Notion, and checks three things: are the books complete, are we paying contractors right, and are we getting paid right. It writes a report and a file of proposed fixes, then changes nothing until someone approves each one. Our operations lead owns finance. The bookkeeping commands that sat next to it got retired in August for a bank feed and off-the-shelf software (that story).

The limit: the only reconciliation report on our main branch is for May. A monthly command that ran once on record isn’t monthly yet.

/notion:timesheet arrived on September 18. It takes the week’s time tracker suggestions, Claude Code sessions and recorded meetings, and proposes time entries in Notion. Any teammate can run it for themselves. Two rules from the file: “Never drop time, never invent a project.” A block it can’t place gets written without a project, so Notion flags it as incomplete.

The limit: it was two weeks old on October 2. Ask me in a quarter.

Marketing and outreach: LinkedIn in three steps, and warm lead follow-up

/marketing:heyreach runs LinkedIn campaigns through HeyReach in three separate steps. build takes your copy and a lead list and creates a draft campaign. It can’t start one. start re-reads everything from the API and tells you how many people will be contacted over how many working days before you authorize it. reply triages responses, drafts answers for review, and suppresses opt-outs immediately. Splitting it means one slip can’t take a campaign from idea to a stranger’s inbox.

The limit: the command file forbids Claude from writing campaign copy. Claude does the plumbing and the reply drafts, and a human still writes the pitch. It was five days old on October 2.

/sales:lead-followup takes a warm list, like event attendees or inbound form fills, ranks it, researches the top tier and writes Gmail drafts. A person approves where the list gets cut, and my co-founder edits or strikes each email before any draft exists. It refuses cold or purchased lists, and in its own words “there is no send path.”

The limit: everything ends as a draft, so the last mile is still a person hitting send on each one.

How to pick your first one

Look for a weekly task that starts with documents you already have and ends with one document someone reads. The weekly client update is the cleanest example: the inputs exist, the output has a fixed shape, and a person reviews it before it goes anywhere.

Then build the boring version. Our weekly update was 22 lines on February 18 and is 378 lines now. The rules that make it trustworthy came from running it on real weeks and fixing what went wrong. Commands written for imagined work are the ones we ended up deleting.

If you want help choosing, bring that one weekly task to a coaching session. Forty-five minutes is $99 over screen share: you do the typing, we build the first version of the command on your own files, and you get a recap afterward so you can keep improving it.