---
title: "Is It Cheaper to Build Your Own Software With AI Than Buy It?"
url: "https://goremy.ai/blog/articles/build-your-own-software-vs-buying-ai-costs"
author: Luis Chavez-Mattos
editor: Luis Chavez-Mattos
editor_title: Director, Product
published: "2026-11-06T18:00:10.605Z"
updated: "2026-11-09T18:52:30.370Z"
topic: Build vs Buy Software
description: "AI agents can clone Excel for about $10 in tokens. Here is how the build-versus-subscribe math compares, and where DIY software still falls short."
---

# Is It Cheaper to Build Your Own Software With AI Than Buy It?

> AI agents can clone Excel for about $10 in tokens. Here is how the build-versus-subscribe math compares, and where DIY software still falls short.

Edited by Luis Chavez-Mattos, Director, Product · Published 2026-11-06 · Updated 2026-11-09

## Is it now cheaper to have an AI build your own software than to buy it?

Sometimes, yes, for the build step. One developer reported spending roughly $10 in tokens to have an AI agent rebuild Excel in a browser in about 90 minutes to a couple of hours. Excel costs $180 as a standalone license or $7 to $10 per month in a subscription. But token cost is only one line in the budget. Your time, the AI plan you already pay for, testing, security and ongoing maintenance all count, and those numbers are less clear.

## TL;DR

- A creator reported cloning Excel end to end for about **$10 in tokens**, compared with $180 for a standalone license or $7 to $10 a month for a Microsoft subscription.
- Three conditions had to line up: **models good enough, tokens cheap enough, and agents autonomous enough** that building is easy.
- Small single-purpose tools like a PDF reader (Folio) and a video stats overlay (Outlier Tracker, about 30 minutes to build) are the strongest case for DIY.
- The biggest open question is **maintenance and security**, which the cheap build price does not cover.
- You can skip the build entirely by using **open source projects** that others build and maintain, such as Artcraft, an open source Adobe-style suite.
- The $10 figure excludes the cost of the AI subscription (the creator cites $200 a month plans) and your own time, so the real comparison is murkier than the headline.

## Why is building software suddenly viable?

The creator behind these examples argues three things had to converge, and that they did within the last month or so.

1. **The models got good enough.** He tracks progress with a personal test ladder. It started with Snake in Python, moved to a Rubik's Cube simulation, and then to cloning whole applications. Earlier models (he names the GPT-5 generation) got close to Excel but left rough edges and missing features. He says the latest generation clones it end to end. He names Opus 5.5 as the turning point.
2. **The models got cheap enough.** Burning tens of millions of tokens to recreate something you can buy for $10 to $30 makes no sense. At roughly $10 for a full clone, it starts to.
3. **The agents got easy enough.** Wrapped in an agent, the model works autonomously. You give it a short prompt like "clone this software" and it iterates until it matches.

These are one person's observations, not a controlled benchmark. Still, the pattern matches what many builders report: tasks that took days of supervision now run as a single agent session.

## How does the cost compare, line by line?

| Option | Upfront | Ongoing | What you get |
|---|---|---|---|
| Excel, standalone license | $180 | None (lifetime license) | Excel only |
| Microsoft subscription | None | $7 to $10 per month | Excel, Word, PowerPoint and more |
| AI-built Excel clone | About $10 in tokens (reported) | Your maintenance cost | Features you choose, in the browser |
| Existing open source project | Free | Depends on the project | Whatever the maintainers ship |

Read the table with care. The $10 covers tokens for one build. It does not cover the monthly AI plan that many people use to run agents (the creator mentions $200 a month Claude or ChatGPT subscriptions), the hours spent steering and testing, or what happens when the file format spec changes or a vulnerability appears.

If you only need one app, a $7 monthly subscription is $84 a year. For one person who already pays for an AI plan, the marginal build cost may be small. For someone who would subscribe just to build one spreadsheet, it is not.

## What did people actually build?

Three examples show the range.

**Excel clone.** A browser-based spreadsheet the creator says has complete feature parity with Excel. He describes it as built in a single run, and says he plans to open source it. I could not verify the parity claim from the material, and Excel is a huge product, so treat "every feature" as a claim to test yourself.

**Folio.** This is a PDF reader and editor he built after getting annoyed with Adobe Acrobat on a new PC. He describes Acrobat as bloated, pushing other products and nagging for a login. Folio lets you draw on a PDF, add notes, sign documents and ask an AI questions about the contents. He calls it lightweight, fast, free and open source.

**Outlier Tracker.** He had used a third-party tool that showed performance stats for YouTube videos and channels. It then added features he did not need, required authentication and, in his words, took up about 30% of the screen. He rebuilt just the parts he used (per-video stats, tracking, channel analysis) in around 30 minutes.

The pattern is clear. The wins come from narrow tools where you know exactly what you need and a vendor's extra features are noise. The Excel clone is a stunt that proves capability more than it proves most people should run their own spreadsheet.

## What is Artcraft and why does it matter?

Artcraft is an open source project that attempts to recreate the Adobe creative suite. The creator shows PhotoCraft (Photoshop-style), Vectorcraft (vector graphics) and Filmcraft (a Premiere-style video editor). According to his account, the author pointed an AI at existing software and told it to iterate until it matched.

This matters because it changes the build-versus-buy question into a third option. You do not have to build or buy. If someone else builds an open source clone and keeps feeding it tokens, you can download it, fork it and change what you want. That is how open source has always worked, only now the cost of producing the first working version has dropped sharply.

The creator takes this further and predicts that all software will become open source by default. That is a forecast, not a fact. Commercial vendors still have data, support, compliance certifications and distribution that a clone does not replicate.

## Where does DIY software fall short?

The creator names the standard objection himself: building is not the hard part, owning it is.

- **Maintenance.** Software needs updates for operating systems, file formats and dependencies. Someone has to run the agent that does it.
- **Security.** Hastily generated, "vibe coded" apps can contain holes. The creator expects models to write more hardened code over time and argues that running an agent continuously to patch and scan costs less than buying software. That is plausible but unproven in the material.
- **Scope creep and testing.** A clone that looks right can still fail on edge cases, such as unusual spreadsheet formulas or malformed PDFs. You only find out when it matters.
- **Collaboration and compliance.** Teams need shared formats, permissions, audit trails and vendor accountability. A personal tool does not give you those.
- **Real total cost.** Token spend is easy to measure. Your attention is not.

A reasonable rule: build when the tool is small, personal, low-risk and annoying to buy. Buy or adopt a maintained open source project when the software handles sensitive data, needs to interoperate with other people, or has to be right every time.

## Is it worth building your own software?

It is worth trying if you have a specific irritation with a tool you pay for or tolerate. A PDF viewer, a stats overlay or a small internal utility can reach a usable state in under an hour of agent time. The downside is limited, and you end up with exactly the features you want, even ones no vendor would ship because nobody else wants them.

It is less compelling for broad suites where you use a fraction of the features anyway. In that case, a cheap subscription or a maintained open source alternative gives you most of the benefit without taking on upkeep.

A practical test before you start:

1. List the three features you actually use.
2. Check whether a maintained open source tool already covers them.
3. If not, estimate how often the tool will need changes and who will make them.
4. Start with a prototype and run it next to the paid product for a few weeks.

## Frequently Asked Questions

### Can AI really clone Excel for $10?

One developer reported it, using a short prompt and about 90 minutes to a couple of hours of agent time. That figure covers tokens only. It does not include the AI subscription you may already pay for, your time or later maintenance, and the claim of full feature parity is the creator's own.

### How much does Excel cost compared with building a clone?

Per the source, standalone Excel is $180 for a lifetime single license. A Microsoft subscription runs $7 to $10 per month and includes Word, PowerPoint and other apps. A one-off clone at about $10 in tokens is cheaper upfront, but only makes sense if you accept the maintenance burden.

### Is there an open source alternative to Adobe?

Artcraft is one example. It includes PhotoCraft, Vectorcraft and Filmcraft, which mirror Photoshop, vector illustration and Premiere-style editing. Its quality and long-term maintenance are for you to evaluate before relying on it for professional work.

### Who maintains AI-built software?

You do, unless you adopt a project that others maintain. The creator argues that agents can keep software updated and scan for security issues at a lower cost than a subscription. That depends on your setup and risk tolerance, and it is not yet a solved, hands-off process.

### Is vibe-coded software secure?

Not automatically. Generated code can contain vulnerabilities, and nothing guarantees an AI-built clone has been audited. The creator expects better models to produce more secure code, but for anything handling sensitive data, review the code and keep dependencies updated.