Merge spread OpenAI across the company, but measurable results are missing

Merge uses GPT-6 Astra, ChatGPT, and Codex in engineering, marketing, and sales preparation. The case study shows broad adoption but offers no independent numbers on quality, time, or cost.

OpenAI shows AI moving from code review to dinner planning

OpenAI published a case study about Merge, a company that builds integration products. Its teams use GPT-6 Astra, ChatGPT, and Codex to review product code, support marketing, prepare sales updates, and plan executive dinners. That is 3 products across several parts of the company rather than one isolated experiment.

The primary page returned a 403 error during verification, so this account relies cautiously on OpenAI metadata and related material from Merge. Merge also describes using tools such as Windsurf, Claude, and Devin. This is therefore a multi-vendor operating model, not an exclusive bet on one provider.

The breadth of adoption reveals the real implementation work

The more useful signal is how Merge puts AI into operation. Its own material says the company tracks tool usage, shares examples at all-hands meetings, and reviews security, data ownership, and workflow value before buying a product.

For product and engineering teams, this is more practical than a blanket order to use a chatbot. Each tool gets a bounded job and a person still reviews the result. The value comes from the process around the model, not merely from access to it.

The case study does not establish return or quality

OpenAI provides no independently verified reduction in work, error rate, revenue impact, or total cost. A long list of uses demonstrates adoption, not outcomes. It is also a vendor marketing profile that naturally selects successful examples.

Merge describes security reviews and approved enterprise licenses. That is a useful brake on staff sending source code and internal documents to arbitrary services, but there is no audit showing how consistently those controls work day to day.

Workflow metrics will decide whether the rollout paid off

The useful measures will be code review time, accepted findings, output error rates, and model cost for each task. Teams should also disclose how many suggestions people discard or repair.

If Merge publishes before-and-after results, the story can become a practical guide for other companies. Until then, it is a detailed adoption map rather than proof of productivity.

Lilith's verdict

Three model logos in one workflow make a fine slide. The accountant is still waiting for the line that shows what the company actually saved.