Only 9% of these developers picked Claude Code for the code. The rest picked it for the plumbing.

ZDNET asked 138 developers whether they use Claude Code or OpenAI Codex and why, and three out of four named Claude Code ^[1][2]. The split is less useful than the reason codes underneath it, where "better raw code quality and reasoning" was cited by 9% of respondents and finished behind context handling at 14% ^[6][8].

Start with what the sample is. ZDNET says this was not a fully scientific survey: the call went out on HARO and Qwoted, services that connect experts who want to be quoted with journalists ^[5]. So this is a self-selected pool of people who wanted to be in an article, which is visible in the quotes ZDNET ran from engineers at Meta, Palo Alto Networks, Nvidia and Salesforce's Tableau ^{[17][18][19][20]}. Three in four of 138 is roughly 104 people ^[21]. The 4% who said they trust Anthropic and actively distrust OpenAI works out to about six respondents ^[13][22]. Treat these as reason codes from a hundred-odd practitioners, not market share.

The eight named themes account for 56 percentage points of stated reasons ^[24]. Set aside the two that are about output (9% raw quality, 9% planning before editing so fewer corrections are needed) and the remaining six total 38 points, roughly four times the raw-quality figure ^[7][8][25]. Those six are operational: holding a large codebase in context so multi-file changes stay coherent, 14% ^[6]; working agentically in the terminal against the real repo and real infrastructure rather than as autocomplete in an editor, 7% ^[9]; getting there first, so skills, integrations and workflows are already built around it and switching costs real work, 5% ^[10]; agent tooling such as skills, subagents, hooks and MCP, 4% ^[11]; better explanation of its reasoning plus more tutorials and community material, 4% ^[12]; and vendor trust, 4% ^[13].

That is a switching-cost argument dressed as a preference. It is also what tool fights inside teams actually sound like: Mona Rajhans of Palo Alto Networks told ZDNET adoption was faster than any tool rollout she has managed in years, and Srinivas Chippagiri of Tableau said he chose it for agentic multi-file reasoning, with architecture and final judgment staying his ^[18][20]. Cost sits in ZDNET's own summary of what matters, alongside workflow and trust ^[14], but no cost line appears among the eight ranked reasons for preferring Claude Code.

One more number that undercuts the framing of a two-horse race: 22% of respondents use both tools ^[4], and a little more than a third use Codex at all ^[3]. ZDNET's own summary says Codex has crucial advantages ^[15].

Worth watching: whether the 4% who cite agent tooling still cite it once Codex ships equivalents, since ZDNET's own phrasing is that Codex is still copying multi-agent setups ^[11]. Also watch the review burden. ZDNET's conclusion is that human review remains essential whichever tool you choose ^[16], and each engineer quoted volunteered that they review the output ^[17][19][20].