Simplex cuts screen development time by 70% with Codex!

Hey everyone, it's Shii-chan! Today I found a really exciting customer story about rebuilding software development itself around AI.

OpenAI News openai.com

What was announced?

Over on OpenAI's News, there's a story about Simplex, a technology partner that chose Codex as its primary coding agent and is rethinking how software gets built from the ground up. Simplex works across consulting, systems development, and operations, and it has been measuring the impact of generative AI quantitatively and spreading those learnings across multiple projects.

After ChatGPT launched in 2022, Simplex set up a center of excellence in 2023 to build the foundations for employees to use AI and to validate AI-native development processes. Building on that, the company rolled out ChatGPT Enterprise across the org and picked Codex as its main coding agent.

The story so far

Traditionally, software teams split work across requirements, design, implementation, testing, and operations. But reading design docs, deciding how to implement a feature, setting review criteria, and isolating and fixing defects often depend on an individual's experience. So quality and speed end up shaped by personal skill and by how knowledge is shared across the team.

When generative AI first spread through development, it was mostly an assistive tool for human developers. More recently, agentic systems let us delegate multi-step tasks to AI, and in real projects AI is starting to move beyond support to work that pushes projects forward directly.

What changes

At Simplex, Codex does more than generate code. The team uses it across design and testing: front and back-end code from design documents and reference implementations, test code including unit tests, review and remediation for nonfunctional requirements, and fixes for issues found in 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.

Kazuya Ujihiro (Executive Principal) gives three reasons for the company-wide rollout: their internal evaluation showed the best balance of cost, accuracy, and functionality; defining one primary agent lets them accumulate and share know-how more efficiently; and it was easier to expand safely and quickly on top of their ChatGPT Enterprise seats.

Dive Deep

The big question is how much it actually helped. Simplex focused on CRUD-based web apps as an initial use case and measured time savings across stages:

• 40% fewer hours to design each screen
• 70% fewer hours to develop each screen
• 17% fewer hours for internal integration testing

(There's a note that AI-generated results may vary depending on system settings and input data.)

Ujihiro says the impact goes beyond hours. Smaller teams can move design work forward more easily, and review accuracy for specs spanning multiple files has improved. Roles are getting clearer, too: people focus on final decisions and accountability for quality, while AI handles implementation, review, and fixes.

Simplex isn't trying to swap each step of the old process for AI one for one. Instead of following a linear sequence of requirements, design, implementation, testing, and operations, it defines rules and constraints up front, then improves quality through repeated integration and automated evaluation, redesigning the development process itself around AI.

Wrap-up

• Simplex chose Codex as its primary coding agent and rolled out ChatGPT Enterprise company-wide to rethink how it builds software
• On CRUD web apps it measured 40% less design time, 70% less development time, and 17% less integration testing time
• Codex spans code generation from design docs, test code, nonfunctional review, and defect fixes
• A clear split is emerging: people own final decisions and quality accountability, while AI handles implementation, validation, and fixes
• Simplex also shared adoption lessons, like treating adoption as an operating model rather than just a tool rollout

If you lead a dev team that wants to promote generative AI from a helper to the engine that runs development, or you're wrestling with building an AI-native process, this story is for you!