They did it...

Main Argument

The video argues that advances in AI coding models and agents are making it practical for individuals to build software tailored to their own needs. The speaker says three factors have aligned: models are capable enough, token costs have fallen, and agent-based tools have made development easier.

In his view, building and maintaining a custom app can now cost less than buying commercial software, especially subscription products. AI-generated projects can also be shared or adapted as open-source software.

Examples of Software

• Folio: A lightweight, free, open-source PDF reader and editor. It supports drawing, adding notes, and asking AI questions about a PDF. The speaker built it after finding Adobe Acrobat unnecessarily bloated for basic reading and editing.
• Outlier Tracker: A personal replacement for a YouTube analytics tool, focused on the video and channel statistics the speaker wanted without extra features or screen clutter.
• Excel clone: The speaker says recent models generated a browser-based spreadsheet with extensive Excel-like functionality in about 90 minutes to a few hours, for roughly $10 in tokens. He presents it as an example of the falling cost of recreating complex software.
• Artcraft: An open-source project described as recreating parts of Adobe’s software suite, including Photoshop-like image editing, vector tools, and a Premiere-like video editor. The speaker says it was built by directing AI to iteratively clone existing software.

These examples support the video’s broader prediction that software may become “de facto open-source”: people can create, share, and customize tools rather than relying exclusively on commercial products.

Development, Maintenance, and Security

The speaker acknowledges concerns that AI-generated, or “vibe-coded,” software may have security flaws and still require maintenance. He argues that improving models could produce more reliable code, and that running agents to update software and check for vulnerabilities may become affordable. He also notes that users can rely on open-source projects maintained by other developers.

He revisits an earlier idea that software might collapse into AI agents plus traditional databases, but says recent progress has changed his view: consumers may now have a practical reason to build their own applications.

Product Impressions: NVIDIA RTX Spark Laptop

The speaker gives a positive hands-on impression of a laptop he says is powered by NVIDIA RTX Spark hardware. He describes a Grace CPU, Blackwell RTX GPU, up to 128 GB of unified memory, and native CUDA support.

He reports using it for local AI models, gaming, and video editing, and praises its premium materials, touchscreen, relatively quiet fan, and low heat under load. The segment is presented as a personal evaluation alongside a recommendation to check out the product.

Other Developments and Recommendations

The speaker highlights Omari, an operating system from the creator of Ruby on Rails, saying AI is built into its design so users can customize the OS. He also recommends experimenting with open-source software and trying to build a small application.

DeepSeek V4.1 Flash is mentioned as a notable open-source model, with the speaker pointing viewers to a separate video about it.

Caveats and Sources

The claims about software-cloning speed, feature completeness, costs, and future security are the speaker’s assessments rather than independently verified tests. The video includes demonstrations of projects but does not provide a step-by-step coding tutorial.

• Main speaker/source: Matthew Berman
• Other sources mentioned: The developers behind Artcraft and Omari, and the creator of Ruby on Rails; they do not speak directly in the subtitles.