Cult.fit, a technology-enabled fitness company operating 708 fitness centers across 77 cities and serving nearly one million paid members, is standardizing engineering on OpenAI Codex and extending AI beyond engineering. CTO Santhosh Kumar says the goal is to become more AI-native: running operations in an AI-first way, creating new customer experiences, and transforming how the organization works.

Cult.fit’s first major shift was in engineering. Developers had been using a mix of coding agents and assistants, which created inconsistent conventions for repositories, agent instructions, skills, and development workflows. Standardizing on Codex gave the company a shared platform around which to build its software development lifecycle, with common conventions for repository structures, skills, testing, code review, and feedback loops.

From there, the team extended Codex into specialized engineering workflows. Its Site Reliability Engineering agent aggregates information from observability platforms and logs, investigates incidents, and often creates a pull request with a proposed fix for an engineer to review. Security agents identify and address vulnerabilities, while on-call agents investigate recurring production issues. Cult.fit is also moving toward a more spec-driven development model, where a specification guides work through the software development lifecycle toward a pull request. One recent example is a social feed for the cult.fit app: Kumar estimates the project would previously have taken more than a year, but using the new workflow the team built and shipped it end to end in two months.

Codex has since expanded beyond engineering. Cult.fit made its company data easier for Codex to work with, so teams can find answers and perform routine analysis themselves instead of going through a central analyst team. HR built a workflow in Codex to manage performance reviews, scheduling, feedback collection, and information aggregation. More than 20 internal applications have been built by non-engineering teams in recent months. When the cost of its internal dashboard software increased sharply, one engineer rebuilt the functionality with Codex over a weekend; cult.fit now runs the capability on its own application, avoiding more than $100,000 in annual software costs.

Adoption was supported by a large hackathon, a one-day management session where leaders across functions built something relevant to their own work with Codex, twice-weekly engineering office hours, and engineers embedded in selected functions. Cult.fit is also using the OpenAI API for its AI Trainer member experience. Around 20% of gym users currently use the AI Trainer product, and the latest version is a voice-first workout experience using GPT Realtime through an OpenAI API integration, providing coaching cues and workout logging via two-way voice interaction. The company is also applying AI to customer support workflows, such as waiving penalties for missed group classes: an AI agent checks signals like local weather and attendance patterns before deciding whether to waive the penalty. These agents are built into the existing Sprinklr support platform and connected to backend systems, moving support from a predominantly human-led chat model toward an AI-first workflow. Reported results include more than 20 internal applications built by non-engineering teams, more than $100,000 in annual software savings, and a 50%+ reduction in customer support costs.

Next, cult.fit wants to move from individual AI workflows to an organization where AI is built into how work gets done. Kumar describes the operations ambition as an “autopilot operations org,” with AI taking on more support work around frontline teams. In the member experience, the goal is a “trainer in your pocket,” extending personalized guidance beyond what a human trainer can provide at scale.