Standardizing Development Across Teams
Cult.fit has standardized its software development lifecycle on Codex, enabling teams to build applications, conduct analyses, and implement ideas without central bottlenecks. Prior to adopting Codex, developers utilized a variety of agents and assistants, leading to inconsistencies in repository conventions and workflows. By consolidating these practices around Codex, the company established shared repository structures, testing protocols, code review processes, and feedback loops.
The engineering team has developed specialized agents for Site Reliability Engineering, security, and on-call incident investigations. These agents can aggregate observability data, propose fixes, and create pull requests for engineers to review. Cult.fit is also transitioning to a specification-driven development model, which streamlines the process from specification to pull request. According to Santhosh Kumar, CTO at Cult.fit, this shift has significantly reduced delivery times; for instance, a social feed that would have previously taken over a year to develop was completed in just two months under the new workflow.
Expanding AI into Member Experience and Support
Cult.fit has made its company data accessible to Codex, allowing non-engineering teams to perform routine analyses independently, thus eliminating the need to route requests through a central analyst team. In recent months, more than 20 internal applications have been built by these non-engineering teams, showcasing the democratization of data within the organization.
On the member experience front, Cult.fit utilizes the OpenAI API to enhance its AI Trainer. With approximately 45% of members being first-time gym users, the company recognizes the importance of providing conversational guidance. Currently, around 20% of gym users engage with the AI Trainer, which has been upgraded to utilize GPT Realtime, facilitating a voice-first workout experience that allows for two-way interaction for coaching cues and workout logging.
Customer support workflows have also been restructured. AI agents now evaluate waiver requests by analyzing factors such as local weather and attendance patterns, integrating seamlessly into the company’s Sprinklr support platform. This transition has resulted in a reported reduction of over 50% in customer support costs.
Savings, Training, and Future Goals
When the cost of a commercial dashboard surged, an engineer successfully rebuilt the necessary functionality using Codex over a weekend, avoiding more than $100,000 in annual software expenses. To foster adoption of these new tools, Cult.fit organized a large hackathon, a one-day management build session, and holds engineering office hours twice a week.
Looking ahead, Santhosh Kumar outlined the company’s ambitions to create an “autopilot operations organization” and a “trainer in your pocket.” As more employees engage in building, analyzing, and automating processes with Codex, Cult.fit aims to continue eliminating operational bottlenecks and enhancing overall efficiency.
Original source: OpenAI News