Simplex Says Codex Cut Screen Dev Time by 70%

Japan-based technology firm Simplex has released measured results from its Codex software development rollout, reporting that the AI coding agent reduced per-screen development time by 70%, cut design time per screen by 40%, and trimmed internal integration testing hours by 17%. The company says these gains came from applying the tools across live projects, and it is now scaling that approach across the entire organization.

How Simplex Structured Its AI Development Stack

Simplex began laying the groundwork in 2023, a year after ChatGPT's public launch, by establishing an internal center of excellence to evaluate and validate AI-native workflows. From there, the company deployed ChatGPT Enterprise across all staff and designated Codex as its primary coding agent — a deliberate decision built around concentrating institutional knowledge on a single tool rather than fragmenting teams across multiple platforms.

According to Kazuya Ujihiro, Executive Principal at Simplex, three factors guided that choice. "First, our internal evaluation showed it offered the best balance of cost, accuracy, and functionality. Second, we wanted to define a primary agent so we could accumulate and share usage know-how more efficiently. Third, it was easier to expand safely and quickly on the basis of our ChatGPT Enterprise seats."

Codex's role inside the company spans more than code generation. Simplex uses it to produce front-end and back-end code from design documents, generate unit tests, evaluate code against nonfunctional requirements, and address defects caught during integration testing. The company is also running automated pipelines through the Codex CLI, using Python scripts that move continuously from server builds through end-to-end test remediation.

Breaking Down the Results

Simplex validated its numbers against web applications built around standard CRUD operations, treating that category as an initial benchmark before broader rollout. The 70% reduction in screen development time is the headline figure, but the data across all three phases tells a more complete picture of where the gains actually land.

Design work dropped by 40% and testing by 17%. The spread across those stages indicates the productivity improvements are not limited to code output alone — they reach into earlier planning work and downstream quality processes as well.

Ujihiro noted the impact extends beyond engineering hours. "Codex has made it easier for smaller teams to move design work forward, and it has improved the accuracy of reviews for specifications across multiple files," he said. "Roles are becoming clearer on the ground: people focus on final decisions and accountability for quality, while AI handles implementation, review, and fixes."

Rebuilding the Development Process Around AI

Simplex is not grafting AI onto its existing workflow one step at a time. The company's stated goal is to redesign the development process itself around AI capabilities, shifting away from a linear sequence toward a model where rules and constraints are defined upfront and quality is refined through continuous automated evaluation.

Ujihiro believes the trajectory will steepen as supporting infrastructure matures. "For relatively simple systems, there is potential to generate products automatically from an RFP," he said, pointing to a future where AI agents handle implementation and validation directly, with engineers retaining accountability for decisions that require human judgment.

Simplex is now evaluating AI-driven development across all projects, building on what its pilots produced. Whether the efficiency gains measured on simpler web applications hold at greater complexity will be the defining question as the company moves from validated experimentation into standard practice. Full deployment details are available in the OpenAI case study on Simplex.