Which pharma roles will change most as AI matures.

AI will not wholesale replace pharma roles. It will restructure many of them — and the answer is more specific than the panic suggests. The right question for any leadership team is not which jobs disappear, but how the operating model changes around the ones that stay.

The question every pharma leadership team is being asked right now is the wrong one. It is some variation of: which jobs will AI replace? The honest answer is that for a regulated industry like pharma, the wholesale-replacement framing does not survive contact with reality. Roles do not disappear cleanly. They restructure. They compress in some phases and expand in others. And the leaders who win the next decade will be the ones who can see the restructuring clearly and redesign their operating model around it, rather than waiting for clarity that will never come.

The numbers we work with at Glyz suggest something more specific than the public discourse. Across our pharma engagements, about forty per cent of roles see significant workflow change inside twenty-four months of an AI programme reaching production. Fewer than ten per cent see net elimination. And the demand for cross-functional capability — people who can sit between clinical, regulatory, and technical — rises by a factor of two and a half. The shape of the change is not collapse. It is reconfiguration.

Three functions sit at the centre of this reconfiguration. Each one is being misread in a predictable way, and each one needs to be reread carefully if the next round of operating-model decisions is going to land.

THE THREE POINTSWhat this post unpacks

POINT ONEMedical writing — what actually changes

Medical writing is the most-cited example of an AI-displaced role, and the framing is almost always wrong. The vague version of the argument says that AI will replace medical writers — that tools will draft regulatory documents, clinical study reports, and scientific publications without human authorship. That framing collapses two different parts of the job: the drafting phase, where a writer produces a first pass, and the scientific review phase, where a writer interrogates the underlying study data and reconciles it with regulatory expectations.

AI compresses the first phase, sometimes by an order of magnitude. It does not compress the second. If anything, it expands it, because the volume of AI-generated content that needs senior scientific review is going up faster than the number of senior reviewers. The role that survives — and grows — is the senior medical writer who can hold the scientific argument across hundreds of pages, reconcile data inconsistencies, and anticipate regulator questions. That is not a role that gets cheaper. It gets more valuable. The leverage shifts toward seniority, not away from it.

POINT TWORegulatory affairs — where the work shifts

Regulatory affairs follows a similar pattern. The vague claim is that AI will handle submissions, that compliance gets automated, and that the regulatory function will shrink. The specific reality is narrower and more interesting. Document assembly — the mechanical work of packaging a submission — does automate substantially. Strategic regulatory work — anticipating which agencies will ask which questions, structuring the dossier to land the desired label, managing the negotiation through review — expands.

A new sub-role is also emerging at the boundary: the regulator-AI liaison, the person responsible for the organisation’s posture on AI-generated content in submissions, for understanding regulator expectations on model use, and for negotiating where the lines are when the agency itself is figuring them out. This role does not exist on most pharma org charts today. It will exist on all of them inside thirty-six months.

POINT THREEClinical operations — what augments, what does not

Clinical operations is the function with the most upside and the most risk. The vague framing says AI will run the trials, sites will pick themselves, and patient matching just automates. The specific reality is that protocol simulation accelerates design decisions before the protocol is locked, site feasibility moves from anecdote to data-led shortlisting, and trial monitoring becomes augmented rather than autonomous — with human oversight on safety signals remaining non-negotiable.

The risk is not that AI runs trials badly. It is that the operating model fails to absorb the new pace. A clinical operations team that can design a protocol in two weeks instead of six is only useful if the surrounding functions — biostatistics, regulatory, data management — can move at the new tempo. The bottleneck moves. It does not disappear.

THE SEQUENCEHow to act on this

Sorting this out is not a strategy exercise. It is an operating-model exercise. The sequence that works is concrete, and it can be run inside a quarter if leadership commits to it.

The mistake we see most often is treating this as an HR project, owned by talent or by L&D. It is not. It is a structural decision about where work sits, what authority moves with it, and how value gets created across the company. The function that owns it should be the one that owns the operating model — strategy, COO function, or the executive committee itself.

WHY THIS MATTERSThe bigger picture

These three functions matter because they sit at the boundary between scientific judgement and operational execution — the exact boundary where AI both adds the most value and creates the most risk if mismanaged. Pharma organisations that lead the next decade will be the ones that redesign these functions deliberately, not the ones that wait to see what shakes out.

The reason this matters now, and not in two years, is that the lead time on operating-model redesign is long. By the time a pharma organisation has reshaped its roles, hired into the new ones, retrained the existing ones, and rebuilt the workflows around them, eighteen months will have passed. The organisations that started this work last year are already realising the productivity gains. The organisations that start next year will be eighteen months behind.

Glyz works with regulated industry leaders to do this work — not as a one-off redesign project but as a sustained capability-build. If you are starting to ask which of your roles will change and how, the framework above is a useful first cut. The harder work — the actual redesign — is where we come in.