Classic software development in its familiar form risks losing almost all of its value under pressure from artificial intelligence. According to an estimate by Andrew Wilkinson, founder of the investment company Tiny, up to 98% of the software industry may not survive in its current format.

Tiny controls the studio Metalab, the services Serato and Letterboxd, the AeroPress brand, and more than 35 other companies. Wilkinson compared developers to portrait artists at the moment the camera appeared: the technology did not destroy the profession instantly, but it radically changed its economics.

Excel in eight seconds

The argument cites the ability of the AI model Fable 5.2 to reproduce the functionality of Microsoft Excel in roughly 50 hours. Extrapolating the current pace of model improvement several years forward, the timeframes for completing a similar task look like this:

  • 2027 — 7 hours;
  • 2028 — 57 minutes;
  • 2029 — 8 minutes;
  • 2030 — 1 minute;
  • 2031 — 8 seconds.

The second argument was a new ChatGPT feature that began independently choosing the form of its response — for example, a map or a data selection. According to the forecast, within just three to five years AI will create both the interface and the program itself for a specific task, spending less time on it than is required to heat up food in a microwave.

Who will be hit

If the scenario materializes, those at risk will include:

  • Microsoft, whose Excel was cited as the example;
  • Salesforce, Adobe, and Atlassian, which sell work applications by subscription;
  • Notion, Monday.com, and other services with a narrow set of functions;
  • EPAM, Globant, and other custom development studios.

What software companies' revenue rests on

According to Wilkinson, the industry will retain residual value. Its sources:

  • "zombie subscriptions" — payment by inertia;
  • elderly users who are not ready to switch to something new;
  • corporations paying dearly for what should be free;
  • programs for mainframes and large corporate computers that no one finds it worthwhile to update.

The key takeaway for investors: the software market will earn from clients' habits, not from the product itself. A program that can be generated in minutes ceases to protect a company from competitors. To the objection that music stores are still operating, the answer follows: a large store owner used to fly on a private jet, but now drives a battered Ford Taurus. The business may not disappear, but it will no longer bring its owners their former income.

Cryptalist analysis: The 98% thesis looks radical, but it reflects a real shift in the value structure of software companies — from product to retention and ecosystem switching costs. For investors, this is a signal to reconsider the valuation of issuers whose revenue is built on functionality that is easily reproduced by generative models, in favor of platforms with strong network protection and deep corporate integration.