Big Pharma AI Plans Have a Talent Debt Problem

Even the most Tech-Forward Organizations Are Reeling from the Compounding Cost of an AI-Unready Engineering Bench

Big Pharma runs on parallel tracks of investment, largely without anyone comparing notes. 

AstraZeneca has a knowledge graph mapping 10.9 million biological relationships. GSK built a research data platform from scratch to feed its AI and machine learning teams. Bayer runs a generative AI assistant used by more than 40,000 of its own employees. Pfizer is laying the data lake its wider AI transformation program will eventually run on. 

These four offer some of the most visible examples of a pattern that holds across the industry's biggest names, and a growing number of the mid-sized ones too: data is closely-guarded and AI’s value is exclusive to the well-resourced company that deploys it.

All four pharma giants are short of a key resource: people. 

The story is more nuanced (and complex) than a board update

A knowledge graph or a data platform can be easily justified to a board. Sure, it’s difficult to staff. But not for the reason most hiring plans assume. 

In fact, the engineer who can make one of these systems work requires someone who can do more than deliver a basic understanding of large language models or vector databases. They also need to know why a change to a commercial dataset warrants a different level of scrutiny than a change to a marketing dashboard; why a model trained on clinical data carries a different class of risk than one trained on support tickets; and how work actually moves through validation in a regulated organization before it touches a real decision.

That combination, AI fluency plus pharma-specific context, is rare. It’s a needle in a haystack situation, difficult enough to find that research across the sector puts the ratio at roughly sixteen open roles for every one person who genuinely has both. 

In other words, the wider AI labor market is already tight. Pharma is running hotter than that.

What the existing bench is actually missing

It’s clear everyone in pharma with AI ambitions is paying a premium for talent. Look inside the hiring data and a more precise picture appears, and it keeps repeating wherever you look across the sector. 

Take AstraZeneca. The company has already redirected a disproportionate share of its engineering function specifically towards AI roles, well beyond what a typical cross-industry comparison would predict. GSK, meanwhile, has gone further still. In fact, they’re among the most AI-concentrated hiring patterns we have seen anywhere in the sector. 

This is not a handful of companies racing ahead of a slower industry. It is closer to what an entire industry looks like once it has leaned hard into AI hiring and still cannot close the gap. The details hint at what’s going on. The clear majority of AstraZeneca's and Pfizer's open AI engineering roles are for Integrators, engineers who connect AI into systems that already exist, rather than Builders who create something from nothing, and the same skew shows up at Bristol-Myers Squibb and Johnson & Johnson, some of them even more heavily. Genuine AI research talent, the people inventing new model architectures, is not actually the industry's biggest gap. 

Bayer is worth holding onto as the exception: its hiring is noticeably less AI-concentrated than its immediate peers, a reminder that even inside this pattern, companies sit at different points on the same curve. Bayer tilts towards Builders instead of Integrators; bucking industry trends and highlighting that while scarcity is shared, approaches to solving it vary widely. .

The scarce role sits in the middle: someone who can take a model, a pipeline or an agent and wire it safely into a regulated, validated, often decades-old technology estate. That is a useful thing to know, because it is also a learnable skillset. Which raises an obvious question: if the shortage is concentrated in a teachable skill, why is almost every pharma organization trying to solve it through hiring alone?

Short answer: it’s complicated. More on this below. 

Buying has its limits

Hiring externally into a sixteen-to-one market carries a premium which quickly compounds. Every offer made to close a vacancy is a signal to every other pharma company all competing for the same shortlist. Talent demand far outpacing supply pushes the next offer higher again. 

But spend alone doesn’t solve pharma’s talent shortages. It only complicates unlocking the thing pharma needs most: working knowledge of how a specific organization's validation, compliance and clinical governance process actually functions. A new hire, however capable, still has to learn that from a standing start. The people who already have it are already on the payroll.

Losing one of those people to a competitor's higher offer is the more expensive version of the same problem. The organization pays market rate to replace them, and loses, permanently, the specific knowledge that took years to build. Information that was never priced in but is priceless and expensive to lose. 

The case for bulking up your current workforce

The next best alternative starts with a question most engineering organizations have not asked precisely enough: which current employees sit closest to the profile the AI roadmap actually needs? 

Answering that requires the same kind of structured assessment used to define the gap in the first place: mapping existing engineers against a target skill profile, across the technical, delivery and compliance dimensions that actually matter. It requires more than enrolling everyone in a generic AI course and hoping the right people opt in.

Some companies are ahead of the curve in this area. But most are aware of a hard truth: People who can genuinely operate at the intersection of AI and drug development are rare because the two halves of that skillset take very different amounts of time to build. Regulatory judgement, clinical context, and an instinct for how a change actually moves through validation are earned over years inside a specific kind of organization. AI tooling fluency is comparatively teachable, in months rather than years, to someone who already has the harder half. That asymmetry is the whole case for looking inward before looking outward.

Enterprise engineering organizations that have run this kind of structured reskilling, mapped against a target profile rather than a generic curriculum, see it change what a team can actually ship; vaulting the practice from theoretical into the real world.

There’s no mystery anymore: when hiring is not an option, reskilling is the next best thing. Specialist placement covers the work that genuinely cannot wait: a live platform build, a regulatory deadline, a project already behind schedule. 

For example, AstraZeneca closed a fifty-engineer gap this way in eight weeks, a timeline that would be difficult to reach through open-market hiring alone in a market this tight. In their case and in a growing number of others, reskilling covers the rest: the medium-term shape of the team, built from people who already understand the organization and will still be there in three years.

The actual bottleneck

Nearly every major pharma company, including, but far from exclusive to the ones named in this piece, has an AI roadmap that would satisfy any board. 

The problem: Very few have a resourcing plan that fully matches it. This is an emerging and persistent issue because building one means admitting something slightly uncomfortable: the people needed aren't available at scale; as in, they don't exist. This isn't a hiring gap. It's a workforce planning and strategy gap, and it requires immediate reframing.

The organizations that get ahead over the next two years will be the ones who retrain and reskill existing employees as they recruit AI-native engineers as they join the workforce. 

The question becomes: Who is actually going to build this, and how much of that answer was already inside the building?

Where Andela fits

Andela works both sides of this problem for pharma clients: placing pre-vetted specialist engineers where the timeline cannot wait, and building structured reskilling programs that run in parallel rather than in sequence. 

The AstraZeneca placement referenced earlier, fifty engineers in eight weeks, is one of ours. On the reskilling front, we worked with GitHub to re-train 200 internal facilitators—people who then went on to train 3,000 developers on GitHub Copilot. The results speak for themselves: 92% facilitator pass rate and an 80% increase in tool adoption.

Big Pharma companies have budgets. They have advanced tech and sophisticated data. What’s needed: an honest map of where the existing bench actually sits against the roadmap it has been asked to deliver.

‍