What Codex CLI actually does for a small NZ business

Codex CLI is a developer tool, and you should never touch it. That's the point. The work it does, cleaning a messy customer export, emailing a Xero report every Monday, drafting a contract from your template, you commission the way you'd commission a one-off script from a developer. The difference is it lands in a few hours for a few hundred dollars, not a week for a few thousand.

That gap is the whole story for a small business. A job that used to need a quote and a week of someone's time is now a conversation and an afternoon. This post is what Codex CLI is actually doing inside the engagements I run around Nelson, and where it doesn't earn its keep.

It runs on a developer's laptop, not yours

Codex CLI is OpenAI's terminal coding agent. It reads and writes files, runs commands, and calls the GPT models. Through 2026 OpenAI has pushed it well past writing code: it remembers context between sessions (a native Memories system), and it can drive other apps on the computer by seeing, clicking, and typing (computer use, which OpenAI shipped on Mac and then Windows during 2026). Pair it with a scheduler and it runs recurring work on its own.

For you, none of that matters except the result. A developer describes the job, Codex CLI does it, the developer checks it, and you get a clean file or an email that lands on Monday. The interesting question is which jobs are worth it.

Where it's paying for itself

Skip the theoretical list. Here's what's earning its fee in live engagements.

The most common job is cleaning up a CSV. A client exports customers, suppliers, or transactions from one system and needs it formatted for another: inconsistent capitalisation, duplicates, dates in the wrong format, one address column that should be three. That's half a day of someone's time turned into a short Codex CLI run plus a few minutes of the client checking the file. They never see the tool. They see a clean file.

Close behind is automating a Xero export. Most of my clients live in Xero and want the same reports pulled every week or month, usually Xero data joined to something else, a jobs spreadsheet, a CRM export, a supplier list. A script that pulls it, joins it, formats it, and emails the owner on a schedule used to be a multi-day build. Now it's an afternoon.

The rest rhyme with those two. Drafting a scope-of-work from a template, so the admin lead starts from 80% done instead of a blank page. Migrating data when a client switches CRM or bookkeeping tools and the formats never quite line up. Standing checks nobody enjoys running by hand: flag any quote from the last fortnight with no follow-up, email a one-line sales comparison every Friday. All of it small, custom, recurring work that never justified a developer's day rate before.

Too small for a developer, too custom for Zapier

For years the choice was binary. You used a no-code tool like Zapier or Make for what it could handle, or you paid developer rates for everything else. The middle was thin.

Codex CLI fills it. A one-off script that does exactly the thing your business needs, deployed once and running on a schedule, is now cheap enough to be worth building. In the jobs I quote, that kind of script lands in the low hundreds to low thousands, where a bespoke developer build used to start at several thousand. That changes which problems you can afford to solve.

Iteration got cheap too. When the report format or the data source or the schedule needs to change, it's a short conversation, not a fresh project. The systems you build feel less like fragile contractor work and more like something the business actually owns.

The class of work that used to be too small for a developer but too custom for Zapier is now economically viable. That's the change owners need to understand.

Where it doesn't fit

The use case is genuinely good, which is exactly why the edges are worth being honest about.

It still needs a developer to set up and review. Codex CLI is a terminal tool. Articles that suggest a non-technical owner can drive it directly are wrong in any practical sense, and learning it from scratch is a poor use of an owner's time.

It's the wrong tool for simple plumbing. If the job is "when a form is submitted, email sales and add a spreadsheet row", that's a Zapier or Make job at around $20 a month, not $500 of setup. Codex CLI earns its keep on the custom end, the work too specific for a templated automation. Picking the right tool for the right job is most of what the first conversation is about.

Anything customer-facing still needs a human to read it before it goes out. A script that generates a report for the owner is fine. A script that emails customers with nobody looking is a risk that needs explicit thought, same as any agent. I go deeper on that in AI agent risk and governance for NZ small businesses.

And it won't fix a broken process. This is the big one. If your data is messy, ownership is unclear, or the process lives in one person's head, automating the symptoms won't fix the cause. Codex CLI is fast at work that has a clear shape. It can't invent the shape for you. That's a consulting problem, not a tooling one.

What commissioning it looks like

A job usually runs like this. We spend 30 minutes on what you want done. I quote a fixed price. I do the work, usually half a day to a day, and walk you through the output. You use it for a week or two and we adjust anything that's off. Most of these close inside a fortnight.

The output is rarely glamorous. A CSV that arrives by email. A report in the inbox on Monday. A folder of contract drafts that are most of the way done before the admin lead opens them. The absence of a weekly chore nobody enjoyed.

Before you commission one, get four things clear:

• The exact job, in business terms. Not "automate our admin", but "a list of overdue invoices every Friday, sorted by customer, with the last contact date beside each one".
• Who the output goes to and what they do with it.
• What data the script will see, and whether any of it is sensitive enough to need a conversation about where it's processed.
• Who owns the script once it's built, meaning who notices when it breaks and who decides when it changes. This is the one that gets skipped.

Get those right and you get a system that earns its keep. Get them wrong and you get a tool that runs for a month, breaks quietly, and goes unused. The technology isn't the constraint anymore. The clarity of the brief is.

Which first job you pick matters most. I've written separately on what to automate first in a small NZ business, and the same thinking applies here, with Codex CLI suited to the more bespoke end.

If you've got a workflow you've been meaning to automate and want to know whether Codex CLI is the right fit, or whether something simpler would do the job, start at Codex CLI consulting or get in touch.