Anthropic’s AI workplace: 4 signals for pharma media

With a pair of recent reports about how AI is changing its own workplace and what its employees and the US public expect next, Anthropic offers pharma media a useful glimpse of where work could be heading.

We would not expect every agency, publisher or pharma company to match Anthropic’s intensity of AI use. What makes the research interesting to pharma media, though, is the direction of travel: more delegation, different responsibilities and new ways to put human expertise to work.

The pace will vary. Anthropic is also a rapidly growing AI company studying its own business. Still, its findings deserve curiosity and scrutiny. Here are four findings worth a closer look, and what they could mean for pharma media.

1. More work becomes possible. Choosing the right work matters more.

What Anthropic found: Employee-directed interactive agents recorded six runtime hours per assumed human work hour by August 2026. Output also rose across functions, including sales and legal. Parallel machine runtime cannot be counted as human hours saved, and non-random adoption limits causal conclusions.

What it could mean for pharma media: Teams could explore more audiences, test more ideas and make smaller-market campaigns viable. That is an attractive prospect—provided somebody asks whether the additional work is useful.

A counterweight comes from Veeva, which estimates that roughly 80% of approved content goes unused by field teams. This concerns field engagement, rather than the whole media market, but it makes the point: A bigger library needs more than a faster printing press.

The opportunity could therefore be to spend more time understanding what deserves to be made.

2. Judgment gains value, but experience still needs an entry point.

What Anthropic found: Junior employees’ share of technical hires fell from 27% to 9%, while the company’s workforce tripled. That establishes a changing hiring mix, rather than absolute recruitment losses. Hiring assessments increasingly emphasised autonomy and judgment.

What it could mean for pharma media: Planners, analysts and editors may spend more time challenging recommendations, interpreting evidence and defending decisions.

But “everyone becomes more strategic” leaves something important unexplained: How do people acquire that judgment?

Some routine work is also training. As it becomes automated, agencies could create better apprenticeships through supervised decisions, campaign investigations and exposure to clinical audiences. Tomorrow’s experts will still need somewhere to be beginners.

3. Managing the work becomes part of the work.

What Anthropic found: The share of interactive agent sessions devoted to building instructions, reviews, dashboards and other mechanisms for managing work rose from around 10% to 22%.

What it could mean for pharma media: AI could change the job description as much as the production schedule. Someone will need to set objectives, check evidence, investigate exceptions and decide when a system’s recommendation should be challenged.

That creates opportunities in oversight, measurement and workflow design. It also gives efficiency calculations another cost to include: supervising the machinery.

The same discipline applies to expectations about clinical audiences. In Doximity’s survey, 75% of physicians wanted reclaimed administrative time for personal time. Respondents could select several answers, and the sample was not population-based. Even so, it is a useful reminder that saved time does not automatically become available attention; the clinician’s newly liberated diary may already have plans.

4. Automating tasks leaves the future of jobs open.

What Anthropic’s survey found: Staff expected AI to perform, on average, 75% of their current work by 2030, compared with 28% among U.S. workers. Yet 95% of staff expected Anthropic’s headcount to grow over the following year. These are personal expectations, rather than company plans or established outcomes.

What it could mean for pharma media: Greater automation could accompany growth, broader responsibilities or smaller teams. The outcome will depend on demand and how organisations choose to deploy their capacity.

It also leaves room for the longer-term healthcare question: If other careers change, might more people choose clinical work? Perhaps,but Anthropic’s research does not establish that transition.

WHO’s projected 11 million health worker shortfall by 2030 shows the need. Its observation that unemployed health workers can coexist with unmet demand shows the constraint: Training and funded employment must connect. A larger pool of interested candidates would only begin the story.

What follows for the 2030 budget?

These findings suggest several possible destinations for the productivity dividend. Finance could retain savings. Brands could reinvest in audiences and experiments. Agencies could charge differently for expertise, while oversight and technology absorb more spending.

WFA’s research already describes a broader shift away from agency payment based purely on labour, offering corroboration for that commercial tension.

Sequoia takes the argument further with “service-as-a-software”: AI businesses selling completed work and charging for an agreed result. Its subsequent services thesis identifies work already outsourced as a natural entry point.

For pharma media, that could mean buying campaign reporting, asset adaptation or parts of campaign operations as defined services delivered by software-led providers. Technology suppliers could compete for budgets currently allocated to agencies, while agencies could package their own expertise into similar offerings. Which services could become repeatable purchases—and where would clinical understanding, strategic judgment and accountability command a premium?

Anthropic’s studies cannot tell us whether pharma media budgets will rise or fall. But they do give us a better question to take into the meeting: If producing more becomes easier, what will make the work worth more?