Matthew Berman on Building Personal Software With AI

Matthew Berman says a frustrating experience installing a PDF reader on a new PC led him to build his own. He wanted a lightweight way to read and edit documents without features and account prompts he did not need.

That project led him to build other tools and consider a broader possibility: consumers may increasingly make or customize their own software. He also points to other developers creating open-source alternatives to established applications.

In his view, three conditions had to converge: models needed to be capable enough, cheap enough and easy enough to use through autonomous agents. He says recent improvements in both capability and token cost have changed his personal calculation.

The decision, he argues, is the cost of creating and maintaining a custom application versus buying or subscribing to one. A custom tool can also include a feature that matters to its owner even if a commercial vendor would not prioritize it.

Berman describes how his model tests evolved from making Snake to simulating a Rubik’s Cube and then rebuilding more complex software. He says he prompted an agent to create a browser-based spreadsheet application modeled on Excel.

He claims a recent model built his spreadsheet clone in about 90 minutes to two hours for roughly $10 in tokens, and claims complete feature parity. The video shows a sample workbook; it does not independently establish that every Excel feature is present or correct.

He compares that claimed build cost with the price of a standalone license or subscription, and says open-sourcing a base application lets other people adapt it. He predicts software will become much cheaper and increasingly open source; these are his forecasts, not demonstrated universal economics.

Berman points to Artcraft as an example of an open-source creative software suite and shows interfaces for image, vector and video work. He presents it as another example of developers using AI to build alternatives to established products.

He revisits an earlier thesis that enterprise software might collapse into agents operating on databases. He says he previously did not expect individual consumers to build their own applications economically, but now thinks the balance has shifted.

He demonstrates Folio, his PDF reader and editor. The interface lets him view a document, draw, add text and signatures, and ask questions. He says he can add functionality as needed and has made the project available for others to use or fork.

He also describes Outlier Tracker, a tool for inspecting YouTube video and channel statistics. He built it after finding an existing service increasingly cluttered with features he did not need; he says the initial version took about 30 minutes.

Berman shows a spreadsheet interface with tables and charts and reiterates his claim that it reproduces Excel functionality. The demonstration illustrates a working interface, but does not test all features, compatibility or reliability.

A common objection is that building an application is only the beginning: it must be maintained, secured and kept current. Berman acknowledges those concerns, then argues that improving models and cheaper agents could make ongoing maintenance economical.

He suggests users can also adopt open-source projects maintained by others. He expects falling token costs and developers’ interest in building to support a wider ecosystem, although the future cost and quality of maintenance remain uncertain.

Finally, he mentions Omarchy, an operating system project associated with Ruby on Rails creator David Heinemeier Hansson, as an example of customizable computing. He encourages viewers to try open-source tools and build a small application themselves.