Non-Developers Built 20+ Internal Apps: How cult.fit Uses Codex
OpenAI has published a case study on Indian fitness company cult.fit, which uses Codex for engineering and internal operations, and has added voice conversation to its member-facing AI Trainer.
On November 5 local time, OpenAI published a case study on how the Indian fitness company cult.fit uses Codex. The company relies on Codex for everything from its engineering team’s coding work to internal app development by employees who aren’t developers. Its AI Trainer, which helps members through their workouts, was built by plugging OpenAI models into the company’s own service.
More Than 20 Apps Built by Non-Developers
cult.fit says that teams outside of engineering have built more than 20 internal apps over the past few months. Codex is also used for data analysis and for building work tools. AI use has expanded beyond developers writing code into individual teams building the tools they need themselves.
The company also rebuilt features of software it had already been paying for. cult.fit says it recreated an internal dashboard in-house and cut more than $100,000 from its annual software spending. The social feed in its app was expected to take more than a year to build, but shipped in two months.
Humans Still Review the AI’s Bug Fixes
The engineering team standardized its repository structure, testing, and code review practices around Codex. When investigating the cause of an incident and drafting a fix, the team uses agents, AI that handles tasks on its own. Fixes proposed by the AI are then reviewed by an engineer.
In customer support, AI now evaluates requests to waive no-show fees, weighing factors such as the weather and attendance patterns for the class in question. The feature is connected to the company’s existing Sprinklr customer support system as well as internal systems. cult.fit says customer support costs have dropped by more than 50%.
An AI Trainer That Talks With Members
The member-facing feature is the AI Trainer, which guides workouts and logs them. The latest version is built on GPT Realtime, a model designed for voice conversation, integrated into the service. Members and the AI can speak back and forth, handling both workout guidance and logging through conversation.
According to cult.fit, more than 45% of its members are gym beginners. About 20% of gym users use the AI Trainer. With so many members who have little training experience, the AI Trainer serves as a channel for getting spoken guidance and recording what they did.
cult.fit has applied AI in three distinct places: internal tool building, customer support, and workout guidance for members. For employees, the way work tools get made has changed. For members, the way they receive workout guidance and submit support requests has changed. The open question is whether the lower costs and shorter build times also translate into better tool quality and higher member satisfaction.