Part 3: Your First Real Data Task with opencode

In Part 1 you saw the difference between asking a chatbot and handing work to an agent. In Part 2 you got opencode installed. This is Part 3: let’s actually do a real piece of work together, the kind you’d recognise from your day-to-day.

No tl;dr here. Just follow along and you’ll have something real done by the end.

The task

Imagine you’ve been handed a messy CSV of sales (the kind that arrives with inconsistent columns, blank rows, mixed date formats, and a stray “NA” or two). You want two things:

1. A cleaned version of the file.
2. An answer to a question: who are the top 5 customers by region?

Any spreadsheet wonk has done this by hand dozens of times. It’s fiddly. Let’s have opencode do it.

Step 1: Put your file where opencode can see it

Create a folder on your computer for this little project (call it something like sales-task) and drop your CSV inside it. Then:

1. Open a terminal in that folder (in Windows Terminal / Git Bash, cd into it).
2. Run opencode and press Enter.

When opencode is running from that folder, it can see and edit the files in it. Done. The setup is just that.

Step 2: Hand it the work (the prompt)

Copy this, open opencode, and paste it in:

Here's a messy sales CSV in this folder. Clean it up: fix the date formats,
drop fully-empty rows, and handle the "NA" values. Save the cleaned version
as sales_clean.csv. Then look at the cleaned data and tell me who the top 5
customers are by region.

Notice what you didn’t do:

• You didn’t explain how to clean it (it can read the file and figure it out).
• You didn’t say which tool to use (Python vs formulas, you don’t care).
• You did say what done looks like: a saved sales_clean.csv plus a top-5-by-region answer.

That’s the agent-style prompt from Part 1, in action.

Step 3: Watch, then check its work

OpenCode will probably ask a clarifying question or two, pick an approach, and get to it. Depending on the file, it may have to stop and check something with you; that’s normal and fine.

When it says it’s done, don’t just say “okay.” Verify:

• Did a file called sales_clean.csv actually appear? Peek at it.
• Does the top-5 answer make sense against the data you know? A wrong top-customer should set off the same alarm bell it would if a colleague told you.
• If anything looks off, tell it: “the dates still look mixed, fix that” or “I expected Acme at the top, why isn’t it here?”

This check-you’ve-always-done-yourself is the part you keep. The agent did the slog; you did the judging.

Step 4: Ask one follow-up

Now lean on it a bit more; this is where the collaboration gets fun. Try:

“Good. Now turn the top-5 result into an Excel formula I can paste into a PivotTable.”

Or:

“Explain in plain terms how you found the top 5, step by step.”

You now have a repeatable pattern: state the goal → let it work → check the result → iterate. That pattern is the whole skill, and it transfers to far bigger tasks than a CSV.

What you just learned

• The agent did hands-on file work, not just talk.
• You directed the outcome and verified it; your judgment mattered.
• You can iterate and refine without starting over.

Where to next? The opencode docs have much more depth. But you’ve already crossed the threshold that matters: you used an AI agent to do actual work, your way.