Q&A: AI Coding Tools Carry the Real Risk

In episode 6 of this video interview the series, Manufacturing Intelligence, PharmTech’s Chris Cole, spoke with Richard Jaenisch, senior director of education, outreach, and digital experience at Open Biopharma, about where artificial intelligence (AI) drug discovery actually stands, how the Pope's recent commentary on AI ethics compares to the embryonic stem cell debate, and where AI agents are, and aren’t, worth the security risk in manufacturing.

PharmTech: Where Is the Line Today of What AI Can Do in Discovery?

Jaenisch: There are really 2 approaches under the same umbrella. One mines existing data to identify gaps scientists might have missed, then applies that insight toward a new formulation. Scientists stay in control here, they can look at the output and say, ‘This makes sense,’ because they understand the underlying pathways. The other is closer to generating massive volumes of candidates and hoping something useful emerges, which I'd describe as chimpanzees writing Shakespeare. Give it enough attempts and you'll eventually get something, but I haven't seen that method produce a real success yet. The pathway-driven approach is where I expect real movement, since scientists can still translate and validate the results.

Are There Specific Capabilities Being Exaggerated?

It's not straightforward hype, and it's not fully delivering either. Even the newest generation of tools, like the latest AlphaFold, hallucinates more than its predecessor, and hallucination is inherently not useful, even though generating novel possibilities is part of the point of generative tools. Some of the gap comes down to how people use them. I've seen strong results from tools that weren't designed for the use case, simply because someone found a clever workaround. Benchmarks and real lab use don't always line up.

You've Compared the Pope’s Take on the AI Debate to the Stem Cell Controversy. Where Are You Seeing Those Parallels?

Both debates carry a strong moral charge that isn't really about the underlying science. With embryonic stem cells, the quandary centers on the source of the cells and where life begins. AI has no such component. It's a machine, not a form of life. What carries over is the pattern: a prominent religious voice shapes conversations well beyond people who share that faith. If colleagues in your circle are Catholic, AI ethics becomes a topic of conversation whether you are or not, and that attention compounds with other concerns already circulating, like the environmental footprint of data centers.

Where Are AI Agents Saving Time Today, And Where Is the Risk Not Worth It Yet?

Most agents in active use are doing simple work, scraping and compiling information, or handling short, sequential tasks. The majority deployed right now are coding agents, and that connects to manufacturing more than people expect, since every piece of equipment runs on software employees want to customize. The risk shows up with people who build bolt-ons without really understanding the code underneath, especially once an agent is given broad permissions and starts operating outside its intended scope.

Is There a Company Spending Enough Money to Make That Risk Worthwhile?

I'm not convinced yet. Announcing a partnership or a large infrastructure figure is easy, a $30 billion or $200 billion number tells me a company is buying capacity, but not that it's solving the training and talent problem underneath. I'll believe someone is serious when I see real money going into people doing the work, with the hours to match, not just infrastructure. Until that shows up in quarterly and investor reports, I'd call it mostly announcements rather than execution.

What Does a Pharma Company Doing This Right Look Like in Two Years?

We'll know more in 2 years, honestly. Companies are testing different strategies, and part of what separates them is whether the expertise they're hiring or training matches how they intend to use AI. My suspicion is that team-oriented implementations will outperform ones built around individual empowerment. When AI use is a shared, community resource rather than an individual one, tacit knowledge doesn't walk out the door when one person leaves, the rest of the team keeps it moving. Organizations that build AI adoption around teams, not just individuals, are the ones I expect real results from.