Verified, then logged.
Sun 22 Nov 2026 · 00:23 KST / 15:23 UTC한국어 | ENSearch /

02 / IT · Analysis

AI model switches take weeks as apps need reworking

Easier connection changes still leave businesses with prompts, tools and evaluation systems to adjust.

A developer-tool founder viewed from behind examines a prospective customer's repository activity and engineering task board before making contact. AI-generated image

AI model switches took weeks or longer for 70% of company respondents with switching experience, according to AI Times, a Korean AI news publication. They were moving to better-performing or cheaper models. Adapting and checking existing services were obstacles.

AI Times reported that VentureBeat published the survey findings. VentureBeat Intelligence conducted the survey in October. It covered 100 people, including AI staff and consultants at companies with at least 100 employees.

According to the report, systems designed for easy connection changes still needed further adjustments. The ability to connect another model leaves work inside the application. That makes prompts, tools and evaluation systems part of the switching decision.

App changes beyond the model connection

The report identified prompts as one area needing adjustment. Prompts are instructions given to a model. Switching therefore includes revising how a service makes requests.

AI Times also identified tools and evaluation systems as areas needing adjustment. These are parts of a service adapted to its existing model. The work extends from changing the connection to modifying and checking the application.

The limits of the survey finding

The figure reported by AI Times applies to company respondents who had switched models. That group must be distinguished from the full survey sample. Its experience cannot establish switching times for every business.

AI Times reported that companies spent more time and money than expected. Those switching costs belong in assessments of better performance or lower prices. The practical benefit of a new model depends in part on the work needed to adapt and validate the service.