The Software Layer Nobody's Building (Yet)

Here's what keeps me up at night: we've cracked the code on AI for molecular design, we're shipping gene therapies like they're consumer products, and bispecific antibodies are becoming the new commodity. But we're still solving yesterday's bottlenecks with yesterday's tools.

The Infrastructure Illusion

Everyone's talking about AI becoming biopharma infrastructure, and sure, that's real. Target discovery, molecular design, simulation workflows that used to demand months of manual coordination now run on neural networks. Impressive. But here's the thing that nags at me: we've automated the thinking part while the deciding part remains a complete mess.

Protocol design, patient stratification, site selection, imaging reads, safety monitoring. These are where AI actually moves the needle on development outcomes, not just optics. Yet the software ecosystem here is fragmented, proprietary, and frankly archaic. We're watching companies deploy best-in-class AI for molecular biology while their clinical teams are still swimming through spreadsheets and email chains trying to figure out which patient cohort actually matters. The gap between discovery acceleration and development clarity is becoming absurd.

Where the Real Margin Gets Squeezed

Cell and gene therapies are the darling of 2026, right? They work. Bayer's advancing dopaminergic cell therapy into phase three, gene therapies are getting FDA approvals, the science is legit. But here's what doesn't work: scaling them. The gap between what works clinically and what works operationally has become a chasm.

Manufacturing complexity is exploding. Supply chain orchestration is barely keeping pace with demand. We've solved how to make a therapy; we haven't solved how to make it reliably, at scale, while maintaining economics that don't require a patient to mortgage their house. This is where software could actually be transformative. Not AI that designs better molecules. Software that designs better operations. Real-time visibility into manufacturing variance, predictive inventory management that talks directly to clinical demand signals, automated compliance frameworks that don't require a team of people reading regulatory documents like ancient scripture.

Nobody's really attacking this seriously at the software level. That's a gap worth noting.

The Modality Moment

We're watching a fundamental shift in what gets funded and built. Large molecules, cell and gene therapies, and antibody drug conjugates are projected to power revenue growth over the next couple years. But buried in that trend is something more interesting: we're moving away from the "one modality to rule them all" thinking toward rational combinations.

GLP-1 amylin combinations for obesity and diabetes. PD-1 by VEGF bispecifics for cancer. These aren't random mashups. They're therapies designed with explicit awareness that single mechanisms have limits. The software question becomes: how do you design, manufacture, and manage clinical evidence for increasingly complex combination therapies? How do you track which component is doing what in a patient? This gets geometrically harder very quickly, and the tools to manage it are still stuck in the linear drug paradigm.

The China Factor That Nobody's Ready For

Chinese biotechs have moved way past monoclonal antibodies and ADCs. They're deep in RNA therapeutics, T-cell receptor therapies, bispecifics, protein degraders. The innovation bandwidth is genuinely impressive, and it's arriving at a moment when global R&D is fixated on regulatory approval cycles and manufacturing constraints.

What this means for software: the competitive moat isn't going to be proprietary molecules anymore. It's going to be operational speed and transparency. The company that can go from concept to phase two clinical data faster, with fewer surprises, is the company that wins. That requires software infrastructure that doesn't exist yet at the level it needs to.

The Launch Gamble

Because of pricing pressure and regulatory headwinds, companies are front-loading commercial investment earlier in product lifecycles. They're betting hard on strong initial market capture. This is where AI-driven sales strategies and direct-to-patient models come into play, but here's what strikes me: you can't AI your way out of a mediocre product. What you can do is use software to make sure your launch doesn't fail because your supply chain imploded or your patient identification was one month off.

The real innovation in 2026 isn't going to be the molecules. It's going to be whoever builds the software layer that connects discovery, development, manufacturing, and commercial execution into something that actually feels coordinated.