Build an AI Agent That Writes Your Weekly Report

You are not too busy to grow your business. You are too busy building the report that tells you whether it grew.

It is 7 a.m. On a Monday. Marcus opens four spreadsheets.

One for sales by branch. One for refunds. One for stock. One where he pastes it all together, colours a few cells, and turns it into the slide he sends his two managers.

Forty minutes later he has a number he already half knew. Then the week starts. He does not look at that number again until next Monday, when he builds the whole thing from scratch.

Marcus is not lazy. He is doing data plumbing by hand. And that plumbing is the only thing standing between him and the ten minutes of actual thinking the report is supposed to trigger.

You know this Monday. You have lived some version of it. So let us kill it.

How do I automate my weekly business report with AI?

Point an AI coding agent at the place your numbers already live, tell it in plain English what the report should show, and let it write a small script that pulls the data, cleans it, does the maths and draws the charts. Save that as a one-click button or a scheduled job. From then on, the report builds itself.

Read that again, because the shift is subtle and it is the whole point. You are not asking the AI a question about your data. You are asking it to build the thing that answers the question every week, without you in the loop.

Here is the shape of it before we build it step by step.

The report was never the job

Here is the trap. You think the weekly report is work. It is not. It is the tax you pay before the work.

The work is the decision. Which branch is quietly bleeding. Which product line is carrying you. Whether cash is tighter this month than last. That is the ten minutes that actually change what you do on Tuesday.

Everything before that ten minutes is copying, pasting, formatting and checking you did not fat-finger a cell. It is not thinking. It is plumbing.

And plumbing is exactly what a computer should do. You have simply never had a cheap way to hand it over. Now you do.

What changed: AI stopped reading your numbers and started building the report

For two years, AI and your data meant one thing: you paste a table into ChatGPT and ask it a question. Useful, but it forgets by Monday, and you are still the one who assembled the table.

That is not what is happening now. The newer tools do not stop at reading a dashboard. They write it.

One post making the rounds among builders this month put it in one line: your AI can build your dashboards now, not query them, build them. It was talking about Rill, a business intelligence tool where every dashboard, metric and data model is defined as SQL and YAML in a code project, so an agent like Claude or Cursor can write the whole thing from a prompt. That is one post from one builder, so treat the hype with the usual salt. But the capability underneath it is real, and it is not limited to Rill.

On YouTube, the same idea keeps showing up because it keeps working. One tutorial walking through a self-updating finance dashboard built with Claude has drawn close to 30,000 views. A newer one, an "AI CFO" that watches receivables and cash flow and writes a leadership-ready review on demand, has passed 65,000 views in a couple of weeks. People are not watching because it is clever. They are watching because they have a Monday like Marcus.

The tools have names, and you should know them: Claude Code and Cursor for the building, Anthropic's Claude models underneath, Rill and Quadratic on the dashboard side, plain old Microsoft Excel and Google Sheets and QuickBooks as the data. You do not need all of them. You need one agent and one honest data source.

Isn't this just a chatbot that reads my numbers?

No, and the difference is the reason this is worth your afternoon.

A chatbot answers once, inside the chat, then forgets. You still fetch the data, you still paste it, you still do it again next week.

A reporting agent builds a machine. It pulls the data from where it lives, cleans the messy bits, runs the same calculations every time, draws the charts, and leaves you a file or a page you can open whenever you want. Ask it to refresh, and it re-pulls today's numbers into the same shape.

Picture Marcus again. Instead of opening four spreadsheets, he double-clicks one button. A script reads this week's sales export, matches it against last week, flags the branch that dropped, and opens a clean page with the three charts he cares about. He spends his forty minutes on the drop, not on the pasting. That is the whole win.

Build it: the report that refreshes with one click

You can do this today, on the report you already make by hand. You do not need to code. You need to describe what you want clearly, the way you would brief a new hire on their first morning.

1. Pick one report you build by hand every week. Just one. The Monday sales slide, the cash summary, the stock reorder list. Small and real beats ambitious and vague.
2. Put the data where the agent can reach it. A folder with your CSV or Excel exports is enough to start. Later you can connect Google Sheets, QuickBooks, or a database like Snowflake or BigQuery. Begin with a file you trust.
3. Open an AI coding agent and brief it like an intern. In Claude Code or Cursor, describe the end goal, the style you want, and the exact data source. As one builder put it, talk to it like a person in your company who does not yet know what they are doing: give it the goal and the boundaries, not jargon.
4. Let it write the pull-and-clean script, then check its numbers. It will produce a first version in minutes. Do not trust it blind. Compare the totals against a report you already believe. This step is not optional. Your brand is your numbers.
5. Ask for a one-click refresh. Have it create a button, or a small scheduled job, that re-pulls fresh data and rebuilds the report. This is the moment the weekly chore disappears.
6. Turn the output into a simple dashboard. Ask it to render the result as a filterable page you can open any time and hand to a manager. Interactive beats a static screenshot, and it costs you nothing extra to ask.
7. Only when it is right, wire it to live data and put it on a schedule. Full hands-off automation, the report that lands in your inbox at 7 a.m., is worth doing, but do it last. Get the one-click version trustworthy first.

Notice the order. The gap between "I do this by hand" and "I press one button" is an afternoon. The gap between one button and fully automatic is a bit more, and it is fine to walk it later. Most of the pain leaves at the button.

If you have read our piece on putting your AI in a loop, this is the same instinct applied to reporting: build the process once, then let it run.

Manual Monday report vs an AI reporting agent

Which model should you use, and what will it cost?

Inside these agents you pick a model, and the names sound like poetry classes: Haiku, Sonnet, Opus. That is roughly the size order too.

Haiku is the smallest, fastest and cheapest. Sonnet sits in the middle and handles most everyday automation well. Opus, currently Opus 4.8, is the most capable and the most expensive. The cheaper tiers cost a fraction of the top one, so the honest advice is to start on the cheapest model that gets the job done and only reach for Opus when a build is genuinely fiddly.

On cost, keep your feet on the ground. The builder who filmed the 23-minute report said the whole thing ran to a couple of dollars in tokens. That is his figure for a small job, not a promise for yours, and a heavier daily report will cost more. But the shape is clear: this is a few dollars and an afternoon, not a software project. That is the part worth sitting with.

This is also where being deliberate pays off. An agent that remembers how your business works, the kind we described in the agent that remembers your business, gets your report right faster and wastes fewer tokens doing it.

What you have after a week, a month, six months

After a week, one report builds itself. Your Monday starts with a decision instead of a chore.

After a month, you have three of them, because once you have seen it work once you cannot unsee it. The cash summary, the stock list, the branch comparison, all one click each.

After six months, your business quietly reports on itself. You stop being the person who assembles the numbers and start being the person who reads them and moves. New staff inherit the reports instead of inheriting your Monday ritual.

Same numbers. Same you. A business that reports on itself while you run it.

That is the light bulb, and it is not far away. Your first move is small: pick the one report you dread most, and this afternoon, describe it to an agent instead of building it. Chasing money owed is another good first target, and we walk through it in the AI agent that chases your late invoices.