Rippling moved most of its AI coding work from Anthropic to OpenAI

The Information reports that Rippling moved most of its AI coding work from Anthropic to OpenAI in five months. The shift tracks a vendor's model choice becoming a line a CEO defends on a bill.

Rippling is an HR software firm. In May, Anthropic models handled 57 percent of the coding work of hundreds of engineers at the company. OpenAI models handled 16 percent. Five months later, OpenAI handles 52 percent and Anthropic 18 percent. CEO Parker Conrad told The Information that the company sought to cut costs. He said OpenAI proved cheaper.

What the numbers say

The three facts above all come from The Information's reporting on October 24. Rippling is big enough for its coding-tool split to be a signal of its own. The company has hundreds of engineers writing production code with AI assistants.

The reversal is nearly symmetric. The share Anthropic lost is close to the share OpenAI gained. Together the two providers covered roughly three-quarters of Rippling's coding work in May. They still cover about seven in ten of it now. Rippling has not stopped using Anthropic. It has moved most of the volume to the cheaper vendor.

Why one customer matters

A single company does not set a market. One like Rippling is more than a single data point. It is public about its engineering culture. It writes a lot of code with AI tools. It has the kind of budget that moves on cost.

Anthropic's product lead inside coding workflows has been visible for a year. Many teams defaulted to Claude in Cursor, Claude Code and Codex over OpenAI's own models. The Information says the pricing gap is now wide enough to overturn that default. It says the shift shows up at a scale a reader can see in the usage mix. The outlet does not publish the per-token prices Rippling actually paid. The CEO's quote is the only on-the-record source for the move.

What a team can do with this

For a team paying for AI coding, there are three plain actions.

Measure the mix. Most teams know which assistant their engineers prefer. Fewer know which model is doing most of the token work. The split can shift without anyone noticing.

Run a cost test. Pick a representative task. Run it on the current provider. Run it on the next-cheapest one. Compare the token bill and the output quality on the same prompts. Rippling's CEO gave the outcome of that test, not the method.

Keep both wired in. Rippling did not stop paying Anthropic. It kept the option. That is what let it move volume without a project to port everything.

The caveat is that The Information is the only source for the numbers. Rippling has not published them in its own channels. The per-token prices and the task mix behind the split are not reported. Treat the figures as what Rippling's CEO told one reporter. That is still a specific claim from a named company.

Source

Anthropic's Margin is OpenAI's Opportunity, The Information. Paywalled.