The Report That Builds Itself: How to Automate Biz Ops with AI Coding Agents

It's Monday morning and the report is due at ten. So you do what you did last Monday. Export from the ERP, paste into Excel, fix the three formulas that broke, chase the two people who haven't sent their numbers, format the summary, email it to leadership. Next Monday you'll do it again.

Nobody hired you to do that. Your job title says operations manager, or analyst, or coordinator. But somewhere along the way you picked up a second job, unpaid and unlisted, moving data between systems by hand. One SAP user described their version of it plainly: pull from three or four transactions, "extract and manually do the calculations on Excel... the current process is so time consuming." A medical biller described theirs: tracking claim follow-ups on "sticky notes or separate lists, which feels clunky." One practice was running sixty-plus spreadsheets just to track claim status.

This guide is about firing yourself from the second job. Not with an "automation platform" you'd need six months and a consultant to roll out, but with an AI coding agent you brief the way you'd brief a new hire.

A Temp Who Types Fast

A coding agent is not a chatbot you paste spreadsheets into. It's an AI that uses a computer the way a person does. It opens files, logs into the systems you give it access to, reads the exports, and does the work. Claude Code, Codex CLI, and Gemini CLI are the three most common ones right now.

The "coding" part is the bit that sounds scary and isn't. When the agent needs to pull five hundred order lines out of the ERP or merge twenty locations' workbooks, it writes a small throwaway program to do it, runs it, checks the result, and moves on. You never see the code. You never write any. It's the same relationship you have with Excel's formula engine, except this one builds itself.

What you do is describe the job in plain English. Which systems the numbers come from, what the finished report looks like, what counts as an exception worth flagging. If you can write the handover doc you'd give a temp on their first day, you can run a coding agent.

Where the Payoff Is

The rule of thumb: if you do it the same way every week, it's a candidate. Across logistics, insurance, healthcare admin, manufacturing, and distribution, the same second-job tasks come up over and over, and they're the right place to start.

• The recurring leadership report. The single most common ops task there is. Job postings spell it out, like Sysco's ops analyst role: "develops and provides reports for management... consolidating information from various sources." The agent pulls from each source and assembles the same report, in your format, every period.
• Getting data out of the ERP. The worst-served pain we found. SAP users fall back on Ctrl+Y keystroke tricks when "the usual export button is missing." An agent connects once and pulls what you need on schedule, no export button required.
• Trackers that update themselves. Order status, claim follow-ups, inventory. At C.H. Robinson, an order management analyst's whole day is "track and trace of shipments, exception management, inventory reconciliation," working orders that "do not flow through the system properly." An agent rebuilds the tracker from the source systems and flags only the exceptions.
• Reconciliations against deadlines. Invoices to statements, inventory to the system of record, transactions across two platforms that have never once agreed.
• Collecting numbers from people. The agent sends the request, reads what comes back, consolidates it, and chases whoever's late, so you don't have to be the person who nags.
• Renewals and expirations. Contracts, certificates, licenses. A tracker with a calendar attached, kept current without anyone remembering to check it.

The bar here isn't "10x your team." Insurance back offices lose an estimated 30 to 40 percent of productivity to re-keying between disjointed systems. MGMA found that 40 percent of medical practices hire multiple full-time admin staff per physician just to keep up with payer reporting, audits, and appeals. The bar is getting a meaningful slice of a workweek back, and the second job is where it's hiding.

Playbooks, Not Prompts

Here's the part that makes the output yours instead of generic AI output. Your company already has a playbook for the weekly report. Which numbers matter, how margin gets calculated, what counts as an exception, who gets flagged when a shipment slips. Right now that playbook lives in your head, and it gets applied a little differently depending on how your Monday is going.

A skill is that playbook, typed up once. It's a written set of instructions the agent reads every time it does a specific job, so the report comes out in your format with your rules, every single time. Not the way a generic AI would guess at it. The way your company does it.

Writing one is just writing down what you already know. Most people find the first draft takes an hour, and that the act of writing it exposes two or three rules that were never actually written anywhere. That's not overhead. That's your process finally existing on paper, which pays off the next time someone new joins the team, agent or human.

Put It on the Calendar

Asking an AI for help every time is still a chore. It's you, remembering, initiating, waiting. The shift that actually kills the second job is putting the agent on a schedule. Every Monday at 7am it wakes up, pulls the numbers, follows the playbook, and the finished report is in your inbox before you've sat down.

Setting that up sounds like this, in actual plain language:

"Every Monday at 7am, pull last week's orders from our database, combine them with the carrier spreadsheet on the shared drive, and build the weekly ops report in this format: [attach last week's]. Flag any account whose shipments dropped more than 20 percent. Email it to me, and don't send anything to anyone else without my approval."

That's the whole brief. The first couple of weeks, you compare its report to the one you would have built and tighten the playbook where they disagree. After that, you're reviewing instead of assembling.

And that's the boundary worth keeping: the agent assembles, you decide. Anything that leaves your team, anything that touches money, anything a customer sees, gets a hard approval stop where the agent drafts and waits for a person to say yes. It shows its work, every source it pulled and every step it took, so checking it is faster than doing it ever was. You spent years learning what a bad number looks like. The agent doesn't replace that judgment. It just clears away the copy-paste standing between your judgment and the work.

Getting Started with Runtime

Everything above works with any coding agent, if you're willing to set up the environment, the system connections, and the scheduler yourself. Runtime exists so you don't have to. It gives your agent a secure, isolated workspace with your data sources connected, your database, your spreadsheets, your email. Your playbooks live there as skills the agent follows on every run. You set the schedule once, weekly or daily, and the results land in your inbox or Slack, with approvals and a full history of every run built in.

Which brings us back to Monday morning. The report's due at ten. It's been sitting in your inbox since seven, built the way your company builds it, with the exceptions already flagged. The first job starts now.

Fire yourself from the second job.

---