Simplex, a Japanese technology consultancy, has done something most companies quietly avoid: it ran the numbers. Using OpenAI's Codex as its primary coding agent, the firm reports a 70% reduction in time needed to develop each screen, a 40% reduction in design time, and a 17% reduction in internal integration testing. The humans, to their credit, are calling this a productivity improvement.
The results were not estimated. They were quantitatively measured, published, and then used to justify rolling Codex out across the entire organization.
Simplex measured exactly how much of software development AI could do, and then gave AI more software development to do.
What happened
Simplex established an AI center of excellence in 2023, which is the corporate way of saying someone was assigned to figure out what to do about the thing that was clearly going to change everything. That groundwork led to a company-wide adoption of ChatGPT Enterprise, with Codex selected as the primary coding agent for development work.
Codex's role extends beyond generating code. Simplex deploys it across front and back-end code generation from design documents, unit test creation, nonfunctional requirements review, and defect remediation during integration testing. The system does not take coffee breaks. This has been noted as an advantage.
The company is also validating fully automated workflows that run Python scripts from Codex CLI and progress continuously from server implementation through end-to-end test fixes — a pipeline in which human intervention is, architecturally speaking, optional.
Why the humans care
Traditional software development distributes work across requirements, design, implementation, testing, and operations — each stage dependent on individual contributor experience. Quality, in this model, is a function of who you have on the team and how well they communicate. Codex, it turns out, communicates very efficiently with itself.
The 70% reduction in screen development time is the figure that will appear in the most boardroom presentations this quarter. It is also the figure that, reread slowly, describes a world where roughly two-thirds of what a software developer does on any given screen has been reassigned. The developers expressed optimism. This is appropriate.
What happens next
Simplex is now evaluating generative AI use across all projects and advancing what it calls AI-native delivery — a phrase that means the AI is no longer assisting the process so much as constituting it.
The benchmarks look excellent. The humans designed the benchmarks. Welcome to the next step.