In the weeks since I first wrote about the possibility that we’re living through an AI bubble, the debate has continued to rage unabated. During his company’s third-quarter earnings call, Nvidia CEO Jensen Huang flatly dismissed the premise. Sanofi CEO Paul Hudson proposed a new metric: ROAI (return on investment in AI). Some experts argued that what we’re seeing is a “rational bubble”; others claimed the bubble is much bigger (and thus much less rationalizable) than previously envisioned.

All of this is to acknowledge the obvious: That nearly every organization in and around healthcare has plenty at stake in the AI bubble debate. Might there be bubble-like froth in valuations and vendor promises? Absolutely. Might there also be plenty of areas and functions in which AI is solving thorny content- and compliance-related problems without a modicum of hype? Also yes.

There’s a lot more to unpack, starting with the reality that healthcare providers are already using AI. The catch is that they’re doing so differently than how consumer tech companies imagine.

A DHC Group study of clinicians found that the dominant use cases are literature search (78%) and diagnosis assistance (46%). Eighty-six percent say AI already influences treatment decisions, while 83% insist they prioritize their own judgment when the machine disagrees.

That could have profound implications for omnichannel teams. If HCPs look first to AI for answers to their questions, then a brand’s best marketing may be making sure that AI can find, parse and present machine‑readable truth about a therapy at the exact moment a physician needs it. You do not need a mascot or other tchotchkes; you need structured data, consistent labels and evidence that survives being summarized.

“The way through is sober execution: AI‑ready data products that break silos, interoperable multi‑agent workflows with audit trails and UIs that force better questions – so we convert hype into validated gains and avoid mistaking a headline spike for scientific progress,” notes Sajid Sayed, director, digital acceleration, medical communication and information, oncology business unit, at AstraZeneca.

The softer kind of bubble, the one we feel in meetings, is a story bubble. The plot goes like this: every creative brief becomes a prompt, every rep becomes a conductor and every HCP journey becomes a mesh of personalized micro‑moments. The only thing left between us and the promised land is a budget and whatever guardrails are in place.

Parts of that plot are already playing out. IPG Health is using AI to pre‑check campaigns against brand guidelines and sniff out MLR issues before a designer falls in love. Ostro is incorporating AI to enable two-way communication for email and SMS campaigns. Incyte has used generative imagery to make the invisible feel seen as part of an effort to raise awareness of the symptoms of myeloproliferative neoplasms (MPNs). These are not paper prototypes; these are pilots that passed risk review and launched. 

But the bigger challenges are the ones in which an LLM’s charming confidence collides with compliance. You can produce 50 variants of a headline in one minute, but only three of those will survive legal scrutiny (and none should be sent to a physician without human review). That’s where the “governance mode” for which many of us have been advocating enters the picture.

In brainstorm mode, you ask the model to surprise you. In governance mode, you pin the model version, lock the decoding, log provenance and route high‑risk copy to retrieval‑and‑assembly rather than free generation. It is not as thrilling as an agent that writes your brand plan, but it is how you publish without worrying that “likely” could become “possibly” somewhere between lunch and the 3 p.m. build. (If this sounds like lived experience, well, it is for anyone who has watched a phrase wobble under load.)

So back to the bubble. Yes, it exists in some corners and for some valuations and for some poor souls raising a Series A that belongs in 2031. Also no, there’s no such concern in the places where AI is already a better spell checker for our pipeline, a better librarian for our medical references and a better triage nurse for our content backlogs.

ZS’s 2025 outlook found that 93% of life sciences tech leaders are planning to increase data, digital and AI spend this year, even as they confess that only a fraction of employees with access actually use the tools weekly. It’s a very pharma kind of contradiction: Budgets go up while adoption drags, because there’s a general sense that the value is real even if the friction is high. 

Here’s something we don’t say out loud often enough: In pharma, we’ve been using AI for years without calling it that. We’ve created predictive models for targeting, next‑best actions that are basically policy engines and patient‑finder heuristics that might have impressed us more if someone slapped a logo on them.

The difference now is that language models have collapsed the distance between intent and interface. Suddenly the creative director, the brand lead and the MLR reviewer are all “using AI,” whether that means asking for 10 patient‑friendly analogies for a MOA or summarizing six PDFs in a single paragraph that finally gets the coverage explanation right.

If you want a simple yes-bubble/no-bubble heuristic, follow the controls. When the risk people move from bans to playbooks, you are exiting the theater of hype and entering the shop floor.

To that point, in April 2024 most big pharma companies still had broad ChatGPT bans on the books. By late 2025, they were talking about adopting agentic workflows in the compliance‑adjacent zones that used to be off-limits. Nowadays, a majority of leaders say governance is both the main barrier and the main use case. That’s not what you see in Beanie Baby markets; that’s what you see when a technology starts behaving like plumbing.

If you’re a life sciences marketer or omnichannel lead, a few ground rules can help you maintain your footing. First, you should measure usefulness where HCPs actually use AI. They are looking up literature, checking interactions and pressure‑testing their own thinking. If your brand’s evidence isn’t where those prompts land (and in structured, machine‑readable form), then your investment in personalization is lip balm on a broken leg.

Second, you should build for throughput, not fireworks. Your team’s pain is rarely “we have no ideas.” It’s “we can’t get the good ideas through PRC at the speed the marketplace expects.” The quiet, durable ROI is in tools that turn messy drafts into reviewable copy, that map claims to citations and that refuse to hallucinate under pressure.

Third, you should accept the intern metaphor. You can pair AI with human editors, route risky tasks to deterministic systems and keep a tamper‑evident log of everything the machine touched. The latter will come in handy when you’re asked to answer the inevitable “how did this sentence get here?” question.

And yes, it’s worth keeping an eye on the froth, because markets can stumble even when the tech is working. The Bank of England warns about correction risk; fund managers are nervous; CEOs say “trillions” out loud. If macro sentiment sours, some vendors will vanish.

But the pipes you install this quarter (making your content machine‑legible, your segmentation dynamic, your PRC packet self‑auditing) will still be good plumbing next quarter. To borrow a line from Amazon that Zvi Mowshowitz relayed in a different manner: Industrial bubbles are the kind you can live through, because they tend to leave useful infrastructure behind even if some shareholders get burned. Your job is to make sure your customers can still drink the water. 

The marketers I trust most sound neither euphoric nor grim. They’re putting their dollars where they can see the before and after.

“Yes, we have been sold a vision of the future that takes on almost sci-fi-like proportions. Yet beyond some entertaining content in our feeds, we are struggling to make the big leaps forward in our everyday lives,” explains Sasha Giacoppo, PhD, head of CX design at Pfizer. “The gap between expected and perceived is growing, and that looks and feels like a bubble.”

Giacoppo adds a caveat, however. “If you start with where we were a year or two ago, there have already been an incredible number of smaller yet impactful changes occurring, little by little, one after another,” he continues. “Those are the wheels of transformation in motion. That’s real, and will eventually bring us closer to a critical mass.”

“AI is now the first place people go to for answers,” one agency lead recently told Fierce Pharma. Great, so let’s make sure the first place is honest, current and ours to lose. “AI is currently like a really promising intern,” said another. Okay, then for now let’s keep that intern away from anything that could end with a warning letter.

There’s a lot of real estate between those two sentences. In fact, between those two sentences is where the bubble talk fades and the work begins. 

If you look at how life sciences leaders are allocating money and attention, the picture sharpens. Companies are devoting a larger portion of their budgets to incorporating AI in the creation and targeting of their commercial messaging. Leaders now expect agents to stabilize operations under regulatory pressure. Data readiness and governance may be the bottlenecks, but they don’t signify a lack of imagination. And clinicians, the people who actually make decisions, say they’ll use AI as a second brain but not as a spine. 

None of that suggests anything along the lines of “this will all evaporate.” Rather, it suggests both positive momentum toward widespread use and a heightened, responsible degree of caution.

In the end, it depends whether you’re basing your bubble/no bubble opinion on valuations that assume every startup will become the next OpenAI, or on AI’s ability to help solve real problems that pharma marketers actually have. 

As Nivi Mogali, director, enterprise omnichannel analytics capabilities at GSK, puts it: “We may not be in an AI bubble so much as in a period of rapid experimentation within healthcare and life sciences.” The real test, she adds, “will be how quickly today’s AI initiatives mature into sustainable and scalable impact, from discovery to patient engagement. What ultimately matters is, whether AI shifts from proof of concept to proof of value in transforming health outcomes.”

As an industry, we’ve moved past “should we use AI?” and toward “how do we use it responsibly and profitably?” That, ultimately, isn’t a bubble question so much as it is an operational one. And pharma has always been much better at operations than at predicting the future.

If we’re currently living through an AI bubble – and the bubble bursts – what will be the impact on your business, and on the life sciences business writ large? Drop us a note at hello@kinara.co, join the conversation on X (@KinaraBio) and subscribe on the website to receive Kinara content.