Simplex rethinks software development with Codex

Simplex is a technology partner that works across consulting, systems development, and operations. To improve productivity in systems development, the company has quantitatively measured the impact of generative AI and applied those learnings across multiple projects. Building on that experience, Simplex is now evaluating generative AI use across all projects and advancing AI-native delivery in applicable projects, with the goal of improving productivity across the organization.

After ChatGPT launched in 2022, Simplex established a center of excellence in 2023 to create the foundations for employees to use AI and to validate AI-native development processes. Building on that work, the company adopted ChatGPT Enterprise across the organization and selected Codex as its primary coding agent, accelerating an effort to rethink how software development gets done.

In traditional software development, people typically divide work across requirements definition, design, implementation, testing, and operations. Tasks such as interpreting design documents, deciding how to implement a feature, defining review criteria, and isolating or fixing defects often depend on the experience of individual contributors. As a result, quality and development speed can be shaped by individual skill and by how knowledge is shared across the team.

As generative AI started to spread through software development, it was often used as an assistive tool for human developers. More recently, agentic systems have made it possible to delegate multi-step tasks to AI. In development environments, AI is starting to move beyond support and take on work that advances projects directly.

To scale that shift, Simplex adopted ChatGPT Enterprise as the foundation for company-wide deployment and uses Codex as its main coding agent.

Codex's role at Simplex goes beyond code generation. The company uses it across design and testing, including front and back-end code generation from design documents and reference implementations, creation of test code including unit tests, review and remediation for nonfunctional requirements, and fixes for issues found during internal integration testing. Simplex is also validating automated workflows that run Python scripts from Codex CLI and move continuously from server implementation through fixes for issues found in end-to-end tests.

Simplex is developing and testing new approaches to AI-driven software delivery with Codex and ChatGPT, focusing on CRUD-based web applications as an initial use case.

Through that work, the company has measured meaningful time savings across multiple stages of development:

Note: AI-generated results may vary depending on the system settings and input data.

Ujihiro says the impact goes beyond reducing 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. It is also helping us build a model where senior expertise can be applied more broadly across development. As a result, roles are becoming clearer on the ground: people focus on final decisions and accountability for quality, while AI handles implementation, review, and fixes.”

Simplex's experience with ChatGPT Enterprise and Codex highlights several lessons for organizations moving from AI experimentation to operational adoption:

Simplex is not trying to replace each step of the traditional development process with AI one for one. Instead, the company is working to redesign the development process itself around AI. Rather than following a linear sequence of requirements definition, design, implementation, testing, and operations, Simplex is exploring an approach that defines rules and constraints up front, then improves quality through repeated integration and automated evaluation.

Ujihiro sees a future where, as databases, API catalogs, and standardized design rules mature, Codex could take on much of the implementation and validation work. “For relatively simple systems, there is potential to generate products automatically from an RFP,” he says. He also expects more areas where, depending on the function, it may be more effective for AI agents to execute business tasks directly rather than building them as source code.

The next challenge is not just making code generation more efficient. It is rethinking how systems should be built, how they should be maintained, and where people should retain responsibility in an AI-first operating model.