Cult.fit operates 708 fitness centers across 77 cities and serves nearly one million paid members. As the enterprise expanded its physical footprint alongside a consumer products business of fitness equipment, accessories, workout footwear, and activewear, operational overhead began to collide with SaaS licensing expenses. To curb recurring overhead, the company turned to internal code generation, using Codex to replace third-party software workflows across its business units.
Internal code generation through Codex allows businesses to abandon costly third-party software subscriptions and reduce baseline operating costs. For leadership teams evaluating custom development versus commercial software, the central question is whether non-engineering departments can safely build production-ready applications with language models without ballooning engineering overhead.
Standardizing the Developer Lifecycle
Before adopting a unified system, developers at cult.fit relied on a mix of coding agents and assistants. This fragmentation created conflicting conventions for code repositories, agent instructions, skills, and development workflows, which prevented teams from sharing knowledge and reusing capabilities. Cult.fit addressed the bottleneck by standardizing its entire software development lifecycle around Codex.
"Once we standardized on Codex, we could build our software development lifecycle around a single platform."
Santhosh Kumar, CTO at cult.fit, explained that this architectural standardization allowed the business to introduce specialized engineering workflows. These systems aggregate information from observability platforms and logs, investigate incidents, and create pull requests with proposed fixes for an engineer to review, while security agents help identify and address vulnerabilities, and other on-call agents investigate recurring production issues. According to Kumar, moving to a more fully spec-driven development model reduced the development timeline of a new social feed for the cult.fit app from an estimated more than a year to two months.
Citizen Development and Direct SaaS Replacement
The architectural shift extended past the core software team into operational departments. Instead of routing routine queries through centralized data analysts, business units began utilizing Codex directly against company data. Non-engineering teams have built more than 20 internal applications in recent months. The human resources department, for instance, constructed its own workflow within Codex to handle performance reviews, scheduling, feedback collection, and the work of bringing that information together.
"Software has been democratized. With OpenAI’s solutions, our people are able to build their own applications around whatever their teams need."
This democratization translated directly into lower licensing expenditure. When commercial subscription rates for its internal dashboard software increased sharply, an engineer rebuilt the functionality using Codex over a weekend. Running the capability on its own application eliminated more than $100,000 in annual software costs. Alongside these internal tools, cult.fit integrated the OpenAI API to power consumer features, including an AI Trainer built into the member experience where over 45% of cult.fit members are first-time gym goers.
The counter-argument to self-built internal tooling centers on long-term maintainability: non-engineers lack formal software training, creating potential risks around application security and technical debt. Cult.fit managed this boundary by establishing twice-weekly engineering office hours and embedding engineers in selected functions to provide technical and domain support.
Can non-engineering employees across 77 cities continue to maintain more than 20 self-generated internal applications without creating governance overhead that outweighs the $100,000 in annual software savings?
What changed
Software licensing costs are falling sharply, but the engineering burden on internal teams is merely shifting rather than disappearing. While in late March businesses experimented with compact models without high costs for hiring machine learning experts, non-technical departments are now generating production code directly and leaning on full-time programmers for architectural safety.
Our stories: Codex AI Models on Hugging Face: Ready-to-Use Tools for Business (March 28); OpenAI Codex: The New Operating System for Business (June 13).