Why AI is so hard to actually ship inside big pharma

Week 5 of my ML/AI journey

A note before I start: everything here is written at a deliberately high level. Nothing in this post targets any specific company I work at or have worked at. Any sensitive details have been blurred for confidentiality. All opinions are my own.

A lighter week on paper

I had a CIT 595 midterm last week, so most of my hours outside of work went into studying for it. Not much building, but a lot of thinking and connecting.

I spent time catching up with UIUC folks working across different sectors, hearing how they are moving their careers forward, and reconnecting with former colleagues from my past pharma and biotech roles. A lot of them are working with LLMs inside large enterprises now, and it was fascinating to compare notes on how each of us is trying to build. On my own side, I have been playing with life-science AI features in my personal projects and contributing to an open source project.

The real topic: shipping AI inside a big enterprise is hard

This week I want to write about something I keep running into: how genuinely difficult it is to get AI and ML solutions adopted inside a large enterprise, especially in a regulated industry like pharma.

Internally, I have been trying to scale up productivity solutions. Not just within my own team, but ideally to the department and even global level. And I keep hitting the same walls.

Here is what the pushback actually looks like.

There is no workflow for building internal AI solutions. No documentation on how to use a tool once it exists. No clear guidance on who to reach out to, what the approval steps are, or how anything moves from idea to production.

There is no QA process. And it is worse than just missing. There is no formal workflow to even define how we would QA an AI tool in the first place. Same story for long-term maintenance and ownership.

The result is that I can build something genuinely useful for a task, something that works, and still not be allowed to use it, because it has to pass through multiple layers of constraints that were never designed with this kind of tool in mind.

The vendor side has the same problem

I have also been trying to bring in AI features from the software vendors we already use. Working with their consultants, I realized their platforms already have a range of AI and predictive analytics features sitting right there.

But bringing them in hit the same kind of wall. Without a perfect, fully defined use case, you cannot get a quote from the supplier. And without a quote, you cannot move forward. Meanwhile, multiple departments are often chasing the same kinds of solutions in parallel, with no unified strategy tying it together.

And underneath all of it is the deepest issue: AI readiness depends on digitization, and in a lot of large pharma environments, full digitization is not complete yet. If the underlying data and processes are not fully digitized, what exactly are you building the AI layer on top of?

A smarter approach I learned about this week

I was feeling pretty boxed in by all of this until a conversation with another coworker reframed it for me.

His point: some startups are sending forward deployed engineers directly into large pharma enterprises. They find any space they can plug into, sit alongside the internal teams, and work from the inside to figure out how to scale their product into the environment. It is a mix of consultant and embedded engineer.

I think that is a genuinely smart approach. You build the relationship, you gain the inside awareness of where the real friction is, and you solve the problem from within instead of pitching at it from outside. For any startup trying to break into this space, that is a model worth studying.

It also happens to be the kind of role that would let people like me, who have the domain context and the passion to bring this work in, actually make an impact. A lot of the systematic constraints are not things one person can change. But you can still try, and you can still find the seams.

Where I am landing

There are big structural things I cannot change. But I am learning a ton, building a lot, and gaining new perspective on how this industry actually works.

Most importantly, I feel so much more confident.

My skill set is niche. To some people it might even look a little insane, this particular blend of chemical engineering, pharma, and applied AI. But my growth has never been linear, and it does not come from the standard path. It comes from a different angle entirely. And lately I have noticed I am not just building and learning anymore. I am giving advice and contributing to others too. I am genuinely happy with where I am right now.

I am proud of my chemical engineering background. It was rigorous. It taught me not just domain knowledge but how to think, how to work, and how to work with others. Layered with my computer science education now, it is becoming the foundation for exactly the direction I want to grow my career.

New roles and new kinds of jobs are forming in this space as it develops. If you are passionate about this intersection, get ready.

Let’s keep it up.