The average Claude Code user now spends 20 hours a week inside the tool, according to Anthropic's own May 2026 usage report. That number alone says people have built it into how they work. The more interesting question is who those people are, because the honest answer is no longer "engineers." Finance teams, legal teams and marketing operations are adopting the same terminal-based tool engineering adopted first, just six to eighteen months behind.
Here's what the actual productivity numbers say, where the non-engineering adoption is really happening and which of the bigger claims deserve skepticism.
The Productivity Numbers, Not the Marketing Copy
A METR study from January 2026 measured a time-savings factor of roughly 1.5x to 13x on Claude Code-assisted technical tasks. That's a wide range, and METR's own researchers flagged it as a soft upper bound rather than a reliable multiplier, since the study measured a small group of seven technical staff. Treat the direction as real and the exact multiplier as unproven.
More grounded numbers come from broader usage data. Regular Claude Code users report 3 to 5 hours saved per week on coding tasks, with the top quartile saving 5 to 8 hours, and 45% higher reported productivity among regular users compared to non-users. Anthropic's own analysis of roughly 100,000 Claude conversations estimated an 80% task-time reduction on short tasks that otherwise take about 1.4 hours. Once task reliability gets factored in, that headline number roughly halves, from a projected 1.8 percentage points of annual labor-productivity growth economy-wide down to about 1.0 percentage point. Anthropic itself treats the 80% figure as an upper bound, not a realistic average.
Where Non-Engineering Adoption Is Actually Happening
Marketing operations is the furthest along. Anthropic's Cowork product, built on the same underlying model, is already being described by marketers as "an IDE for marketing operations": one tool for closing data silos across platforms, automating manual reporting that used to live in a spreadsheet and clearing the kind of A/B-testing bottleneck that used to require pulling in an analyst.
Finance and legal are earlier in the curve but moving. Goldman Sachs has a public partnership specifically around banking automation. Legal teams are using the same class of tool for early-stage research and document drafting rather than final work product. Neither function is anywhere close to engineering's adoption depth yet, but both are on the same trajectory engineering was on eighteen months ago.
The Claim Worth Treating Skeptically
Some vendor material puts non-engineer productivity multipliers from Claude Code as high as 25 to 30x. That number doesn't have independent verification behind it. It's worth comparing to the same pattern this site flagged in an earlier look at GTM engineers using Claude Code: wide, self-reported multiplier claims tend to reflect different baselines and definitions rather than one consistent truth. A 25 to 30x claim from a vendor guide belongs in the same skeptical bucket as the campaigns-per-week range that piece already flagged.
Why the Gap Between Functions Exists
It isn't a tooling gap. Marketing, finance and legal all have web-based systems, files and structured data Claude Code can already read and act on. The gap is closer to habit and permissioning: engineering teams already trusted a terminal-based tool with production systems before Claude Code existed, so adopting an AI-driven version of that workflow was a smaller leap than it is for a team whose entire toolkit has been point-and-click software up to now.
What This Means If You're Not an Engineer
The realistic starting point isn't rewriting a whole function's workflow. It's the same pattern that worked for early GTM-engineer adoption: pick one recurring task that's tedious but well-defined, something like a weekly report pulled from three different tools or a document draft that follows the same structure every time. Use it there first. The productivity numbers above are real enough to justify trying it. They aren't reliable enough to justify betting an entire workflow redesign on a specific multiplier before testing it on your own work.
FAQ
Do I need to learn to code to get value from Claude Code in a non-engineering role?
No. The tool works from natural-language instructions. The marketing-operations and legal use cases above are explicitly non-coding tasks: reporting, research and document drafting. Understanding what the output should look like matters more than being able to write it yourself.
Is the 20-hours-a-week usage figure typical, or is that an outlier?
It's Anthropic's own reported average across its user base as of May 2026, not a cherry-picked outlier, but it's also an average across engineering-heavy usage, where adoption is deepest. A finance or marketing team just starting out should expect meaningfully less time invested at first.
Which non-engineering function should adopt this first?
Marketing operations, based on current adoption data. It already has the most concrete use cases in production (reporting automation, cross-platform data consolidation) rather than early-stage pilots, which puts it furthest along the same curve engineering already climbed.
If you're trying to figure out where this actually fits in a non-engineering workflow, get in touch.