Companies are using artificial intelligence tools to build their own software rather than buying it from enterprise vendors such as Workday, SAP and Salesforce, according to interviews conducted by Business Insider.
About a third of organisations decided against buying at least one software product or feature because they could build it using AI tools, according to a recent McKinsey survey of 1,719 executives and managers.
West Monroe, a Chicago-based consultancy with roughly 2,000 employees, wanted a program to check payroll for errors and another to surface insights for managers. Buying those capabilities from a vendor would have cost more than $300,000 a year, chief people officer Tanya Moore estimated; instead employees built the software earlier this year using OpenAI’s ChatGPT and Codex, she said.
“The vendor’s most dangerous competitor may now be the customer,” Silicon Valley investment firm Omega Venture Partners said in an August report on its website.
Most companies are not replacing entire enterprise platforms with in-house alternatives, but the approach could reduce the incremental revenue vendors earn from add-on features. HR-tech analyst Kyle Lagunas said the trend “is going to erode a bit of the walled garden,” with HR teams able to self-serve more.
Spotify employees last year built HR Bot, a personalised help desk that draws on the music company’s employee handbook to answer questions about vacation days and parental leave. Chief HR officer Anna Lundström said Spotify, which has about 7,000 employees, might otherwise have sought external support; she said the tool, introduced in November 2025, is widely used and has freed her team for other tasks.
Twilio, a communication software provider with around 5,500 employees, built an internal AI tool called Jarvis that listens to sales calls and coaches representatives. A spokesman said it has reached about 80% adoption across sales teams and reduced the sales-deal lifecycle by 54% for users compared with non-users.
Twilio CEO Khozema Shipchandler told Business Insider that if a tool like Jarvis exists or comes to market, Twilio “probably won’t use it, because we’ve already built our own.”
The trend is notable because investors erased hundreds of billions of dollars from software companies’ market values earlier this year over concerns that AI could make building software cheaper than buying it.
Vendors say the risks are real. Homemade tools require bug fixes, updates and maintenance, and an AI system could make costly errors, such as leaking private information. “If an agent skips a compliance step or approves pay outside your grid, you’re not dealing with a minor hallucination,” Workday chief technology officer Gabe Monroy wrote in a June blog post. “You’re dealing with an audit, a regulator, or a lawsuit.”
At a recent Goldman Sachs conference, SAP CEO Christian Klein argued that large language models are not enough to replace vendors’ years of experience and data. Gartner forecasts worldwide IT spending on software will grow nearly 16% this year to $1.47 trillion.
West Monroe’s Moore said employees have built eight tools in place of vendor offerings, each tested for several months before launch and reviewed regularly. “There’s a lot riding on that accuracy,” she said, adding that some businesses may lack the expertise or appetite for the approach.
Gray, a design and construction firm in Lexington, Kentucky, with about 2,300 employees, expects to avoid paying $1 million for a financial forecasting tool by building one later this year with Microsoft’s Copilot and Anthropic’s Claude. Finance chief Scott Parker said the company had paid that amount for comparable software a few years earlier, plus $100,000 a year in service fees. “We’re realizing the power of AI,” he said.