Seven More Things I Built in Claude Code That Run Without Me

I built seven more things in Claude Code this week that an agency would normally need a developer for.

1. Delivery margin dashboard. An agency hired me to improve delivery, and that starts with where the delivery hours go. I pulled six months of project management history, matched every finished task to their services menu, and built a dashboard of each service’s hours against budget, what revisions add, and each account’s effective bill rate. Now we can see where delivery has to change to improve both performance and profitability.

2. My own effective bill rate. I run two AI time trackers, and every morning Claude compares them and moves anything filed under the wrong client. A dashboard combines those hours with my invoices and contractor costs into my effective bill rate by client, by week and by month.

3. Client voice guides. Since July I’ve been building Slate, an SEO and content platform for agencies. AI drafts sound like AI when the model has no samples of how the client writes, so it now builds a voice guide from an embedded scan of the client’s site and their call transcripts. Every rule points to a line they wrote or said.

4. Rankings into Databox. Slate has a keyword rank tracker built on DataForSEO, and now it sends organic rankings through Databox’s API into the client’s reporting dashboard. That saves 5 to 10 hours a month on reports that used to be built by hand.

5. Board order scraper. One of my clients serves state licensing boards, so every Monday I automatically pull disciplinary orders from 116 of them, read the scanned ones with OCR, and sort each order by what the board requires. Over 9,000 orders a year become market trends they use to focus sales, marketing, and product development.

6. Automatic skill builder. Every week it reads my Claude sessions for work I repeat and builds a skill for it. So far it’s built /debrief, which turns a call into next steps and a recap; /rundown, which writes the note that goes out with a finished deliverable; /reply, which drafts my answer to a tricky message; /recall, which finds what someone said on a past call, with the quote and the date; /source-check, which checks every number and claim in a deliverable against where it came from and flags anything it can’t back up; /work-review, which opens what a teammate built and drafts my feedback; and /hours, which tells me how long I spent on a client, deliverable, or task.

7. Overnight product builder. I set and approve the roadmap for my products, and every night Claude deploys agents to build against it, up to six items a night. It runs the full tests, and the work is waiting for my review in the morning. Nothing goes live until I approve it.

And it’s not all work. Claude also writes my Tuesday fantasy football waiver rundown and Saturday lineup prep for both my leagues. I had it sort every waiver bid from last season and this one by owner, so it knows which owners pay up for running backs and how cheap receivers go.

Most of these run without me starting them

When I wrote the first version of this list in August, most of it was a tool I opened when I needed it. A brief generator. A mockup. A canvas for a pitch. I sat down, ran it, and closed it.

Look at this list again and count how many of them wait for me. The time trackers get reconciled every morning before I’m at my desk. The board orders get pulled every Monday. The skill builder reads my week every week. The product builder runs every night. I don’t kick any of those off.

That’s the change I care about more than any single item. A tool saves you time when you remember to use it. A routine saves you time whether you remember or not, and the stuff that falls through the cracks at an agency is almost always the stuff that depended on somebody remembering.

The skill builder is the clearest example. I didn’t know I was writing the same handoff note over and over until it told me. Since it built /rundown in mid-September I’ve used it 13 times. I never would have scoped that as a project, because no single handoff note felt like a problem worth solving.

I measure my own hours the same way I measure a client’s

Two of the seven are the same build pointed in different directions. One shows an agency where its delivery hours go by service and by account. The other shows me where mine go by client.

Every agency owner I work with has a hunch about which accounts are worth what they charge and which ones aren’t. I have the same hunch about my own clients. The hunch is usually right about which ones hurt and usually off on how much, and the only way to find out is to put the hours next to the invoices.

It would be strange to sit across from an owner and ask them to look hard at their effective bill rate by account when I hadn’t done it for myself. Running it on my own numbers every week also keeps me honest about how messy the inputs are. Time tracking is never as clean as the dashboard makes it look, which is why I run two trackers and have one check the other.

“So who’s checking all of this?”

This is the question I get most when I share a list like this, and it’s a fair one. If a scraper is reading 116 boards and an agent is writing code overnight, something is going to come out wrong.

It does. That’s why /source-check exists. Anything with a number in it that’s going to a client gets checked against where the number came from before it leaves, and anything it can’t back up gets flagged for me to fix or cut.

The overnight builder works the same way. It only picks up work I’ve already decided on, so it can’t wander off and build something I never asked for. Some nights there’s nothing in the queue and it doesn’t run at all. When it does, the work sits in its own copy with the tests run, waiting for me. Nothing ships until I’ve looked at it.

So the review didn’t go away. It became the job. I spend a lot less time making the first version of things and a lot more time deciding whether the first version is right, which is the part I should have been spending my time on anyway.

What this means for an agency

None of the seven needed a developer, and none of them needed me to be technical. What they needed was somebody who knew the work well enough to describe it. I’ve run delivery for a long time, so I know what an effective bill rate is supposed to tell you and what a good handoff note has in it.

Every agency has people like that on the team. The account lead who knows exactly which report takes four hours and why. The strategist who rewrites the same brief section on every account. They’re the ones who should be describing these builds, because they’re the ones who can tell when the output is wrong.

The builds are the easy part now. Knowing which part of the week to hand over, and who’s going to check it, is the part that still takes someone who’s done the work.