Researchers at the University of Utah published a preprint on 23 October 2026 asking whether AI coding agents change how organizations choose between buying and building software. In an organizational survey, 32% of respondents said they had skipped at least one software purchase because their organization could build the functionality with coding agents.
What the preprint found
The paper, "The Customer Is the Builder" by Ripon Chandra Malo, Zimeng Yu and Tong Qiu, compares what agents can build with how the software market is moving. Since 2023, the length of benchmark tasks that leading agents complete at 50% success, measured in human working time, has doubled about every 129 days. A search found 575 potential open-source alternatives to selected commercial products, and the authors' own Craft suite produced 12 application repositories in 12 days of public development.
The authors are cautious about all of this. The survey answers are self-reports and do not show how much spending was displaced or whether the resulting tools are reliable. The new repositories are early versions with unverified completeness. Market evidence is mixed: purchased software gained US investment share, and tracked commercial installations grew. The paper has not been through journal peer review.
Was das für Engineering-Teams bedeutet
The finding describes a decision more than a trend. A team can now ask whether a workflow it would normally buy is worth building, and for some functions the answer has changed. The paper names verification, maintenance, security and integration as the costs that decide whether a build stays cheaper than a purchase.
Those costs fall on the people who own the code afterwards. A tool built in a few days still needs someone to run it, patch it and explain it a year later. Teams weighing a build can price that ownership in engineer time before they decide.
The paper also lists earlier studies that point the same way. One found that more code does not mean more finished products, and another found that a first gain in development speed came with a lasting rise in code complexity. A build that looks cheap in week one can cost more by month twelve, so the comparison with a purchase should include upkeep and not only the first version.
Was das für die Einstellung von Entwicklern bedeutet
If more buy-or-build decisions go toward building, demand rises for engineers who can specify a tool, direct an agent and judge whether the result is dependable. That is a senior skill set. Our how we vet process tests it in a real codebase, and you can read more about our AI-native engineers.
