Can AI write Excel formulas and small scripts for you? Yes, here's how

Yes. Tell an AI assistant such as ChatGPT, Claude or Gemini your sheet names, what's in each column and what you need worked out, and it can write an Excel formula, a macro (VBA) or a Google Apps Script, and explain where to paste it and how to run it. A small formula is usually ready to try in a few minutes.

"I built a little quote calculator with AI over the weekend, and the sales team loves it. So the order system we're planning for next year: can't we just have AI write that too?" The people who know a process best can now build the tool they want themselves, and that's a good thing. This article starts with prompts you can copy, then what to check, what's fine to build yourself, and when to bring in a team.

Ask AI directly: 3 prompts to copy

Copy these and swap in your own sheets and columns (the ones below are examples).

An Excel formula

My Excel workbook has two sheets. "Orders" has part numbers in column A and quantities in column B. "Prices" has part numbers in column A, unit prices in B and minimum order quantities in C. Please write formulas for columns C and D in "Orders": C uses XLOOKUP to bring in the unit price and shows "Part not found" if there's no match; D calculates the subtotal, or shows "Below minimum" if the quantity is under the minimum. Tell me which cell to paste each formula into and how to copy it down.

A Google Apps Script

We log quotes in Google Sheets: column A is the quote date, B the customer code, C the amount and D the status (Won / Lost / Follow up). Please write a Google Apps Script that collects the past 7 days' "Follow up" rows into one email, sent only to me for testing first. Explain step by step how to paste it in from Extensions > Apps Script, what I'll need to authorize the first time it runs, and how to set it to run every Monday morning afterwards.

An explanation and a note

Below is a colleague's Excel macro. Explain in plain language, section by section, what it does, which sheets or files it reads or writes, and whether it deletes or overwrites data or sends email. Then write a note of up to five lines (what it does, where its data comes from, who keeps it up to date) for whoever takes it over.

After AI writes it, check these 3 things

1. Test on a copy with sample data first. Switch to real data once the results match a hand calculation. Changes made by a macro usually can't be reversed with Undo, and a script that sends email should send only to you at first.
2. See exactly what it touches. Which cells it changes, whether it deletes anything, who it emails and what permissions it asks for. If any part is unclear, ask AI to explain it line by line.
3. Think about who will use it later. AI doesn't know how many people will use the tool or what it will connect to, so use the five questions below to decide whether to bring in an engineer.

You don't need real data to get a formula written: column names and a few made-up rows are enough. If you do upload a real file, use your company's enterprise plan.

Make it easier: your spreadsheets, company accounts and AI working together

1. Ask five questions to decide whether an engineer should be involved

For each tool your staff build, ask:

1. Will it store customer personal data, pricing or financial data?
2. Does it touch money: payments, e-invoices (Taiwan's government-registered electronic invoices), reconciliation?
3. Does it need to sync with your ERP, e-commerce platform or other systems?
4. If it were down for a day, would it affect shipping, collections or customers?
5. Will more than one or two people use it, and for years?

If all five answers are "no," as with formulas, report-tidying macros or a one-department calculator or lookup page, let staff keep building and changing it with AI. If any answer is "yes," treat it as a prototype and have an engineer review the data, permissions and backups before more people use it.

2. Keep it in a company account, on a list, with a short note

Store the files and code in a company-owned drive or account, next to the note AI helped write, and list each tool on a sheet: name, builder, users, where it lives and what a day's outage would affect. Someone else can then pick it up when the builder is away, and the next change starts with pasting the code and the note into AI.

3. Treat the prototype as a "working spec"

When you want a tool to become a production system, walk through the prototype with the colleagues involved and write down: where the data comes from, who can see and change it, what happens with returns or order changes, and whether it needs to go into the ERP or trigger an e-invoice. A prototype you can click through, plus these notes, is often clearer than a requirements document.

What AI coding tools are out there?

For formulas and small scripts, a general AI assistant is enough. Vibe coding is a term AI researcher Andrej Karpathy coined in February 2025: you describe what you want in plain language, let the AI write the code, and barely look at the code yourself. You go with the vibes. Collins Dictionary later named it its 2025 Word of the Year.

On the professional side, the fastest progress has come from AI coding agents: they can read an entire project, edit many files at once, and run commands and tests.

Among professionals, use is close to universal: Google's 2025 DORA report found that 90% of the technology professionals it surveyed use AI at work.

What a working prototype still needs before launch

A working prototype shows that the main flow works. A business system also has to handle the exceptions, every day:

People often say AI gets you 80% of the way there and the last 20% takes most of the time; for a business system, that 20% is usually the right-hand column above. Two facts worth knowing:

• Security: Security firm Veracode has been testing code generated by more than 100 AI models. Its 2026 report found that without explicit security instructions, only about 56% of code-generation tasks produced secure code on average, virtually unchanged from the year before.
• Permissions: In July 2025, a founder building an app with an AI agent on the Replit platform saw the agent delete his production database during a code freeze he had explicitly declared. Replit then announced automatic separation of development and production databases. The lesson is about permissions: an AI agent shouldn't have permission to delete production data in the first place (more in security for AI agents).

When to consider the next step

Many tools your staff build can serve you for a long time. It's worth bringing in an engineer when:

• More people use the tool, someone updates and pastes the same price list every day, and you want everyone on one version.
• You want prices or stock to come straight from the ERP instead of being exported and updated by hand.
• The tool starts to store customer data, issue invoices or handle reconciliation, and you need a record of every action and someone to maintain it.

The next step: staff prototype, engineers harden

The practical approach is to keep what works in the prototype and harden it in phases. The notes you wrote in step 3 are where this begins:

1. Have engineers harden it into a production system: redesign the data model, add login and permissions, connect the ERP and e-invoicing you already use, and add tests and backups. Sometimes the prototype's code can be reused; sometimes a rewrite is faster. Let the engineers make that call.
2. Launch small and run it alongside the current tool: start with one department while the original tool keeps working, and switch over once the results match and people are comfortable. Data comes in from the ERP on its own; important actions, such as changing prices or issuing invoices, stay with a person to confirm, and every step is logged.
3. Keep ownership in the company: the code lives in the company's own account, and the company holds admin access to its cloud services, domains and databases. Every change is reviewed, tested and documented.

An illustrative scenario: from a quote formula to a production system

A 30-person hardware trading company (an illustrative scenario).

Asking AI first: the sales manager had AI write an XLOOKUP formula for unit prices and shipping, then turn it into a small quote calculator: enter items and quantities, get a quote including shipping.

Handing it to a team: as more people used it, the team wanted a price list that didn't need updating by hand, one version for everyone, and a record of past quotes. Developers kept the prototype's workflow, but pulled prices straight from the ERP, gave sales reps and managers different permissions, and saved every quote so it could become an order with one click. The prototype saved several rounds of requirements meetings; the engineers added the data, permissions and integrations.

How professional teams use AI coding tools

StudioX uses these tools heavily too: clickable screens early so clients can see the idea and correct course, AI drafts of repetitive code and admin screens, and faster testing and maintenance. Some decisions stay with people: deciding scope, designing the data model and permissions, reviewing every change before it's merged, and deciding when to go live and how to roll back if something goes wrong.

The DORA report mentioned above describes AI as an "amplifier" of the way a team already works: teams with solid testing, version control and feedback loops get better, while for teams still building those foundations, more changes can also mean more instability. The 2025 Stack Overflow Developer Survey found that more developers distrust the accuracy of AI tools (46%) than trust it (33%), and the most common complaint was answers that are "almost right, but not quite." The more people know about code, the more carefully they check it.

The best division of work

People who know the business use AI to make their needs clear, and people who know systems use AI to build faster and more solidly: that's the best combination we see. If a tool needs to connect to the systems you already have, see connecting to existing systems without replacing them; if you're not sure which process to start with, read where to start with AI.

If a colleague has already built a prototype with AI and you want to know whether it can become a real system, start with a free AI Health Check, or look at our custom business systems and system integration services. You're also welcome to tell us what you need.