Benchling CEO on bringing AI to biotech cloud workflows after a brutal industry downturn

Summary

Benchling co-founder and CEO Saji Wickramasekara spoke from the JP Morgan Healthcare Conference, where the mood in biotech is notably better than it has been in years. Benchling spent its first decade moving scientific workflows off spreadsheets and on-premise systems into the cloud. The current focus is AI: putting models and simulation directly inside scientist workflows and building agents to automate the procedural grind of drug discovery.

The analogy Wickramasekara reaches for is GPT before ChatGPT. The underlying models for bio already exist, but making a medicine is complex enough that assembling those capabilities into an interface scientists can actually use in the lab is the hard problem — and the one Benchling is trying to solve.

The most immediate opportunity isn't agentic reasoning or autocomplete. It's retrieval. One customer ingested 10,000 previously untouched experiments into Benchling, and a scientist preparing to run studies that would have taken eight months discovered the work had already been done years earlier. That kind of institutional memory recovery is the low-hanging fruit, and Wickramasekara argues life sciences is full of examples like it.

The overhype question

On where AI excitement outpaces reality, Wickramasekara is careful but pointed. There's genuine enthusiasm around hypothesis-generation models, but drug discovery is roughly 9,999 steps after the hypothesis. The regulated, animal-study, and manufacturing-adjacent stages are where the real opportunity to compress timelines sits — and where AI is underinvested relative to the noise around early-stage discovery tools.

On token speed and silicon partnerships, Benchling has deals with both Nvidia and Anthropic, but Wickramasekara argues speed is not yet the constraint. Most life science organizations outside San Francisco are still just getting basic AI copilots switched on, slowed by legal, regulatory, and security reviews. Scientists already juggle complex data across multiple tools, and that last-mile integration problem comes before inference latency.

The biotech downturn

The industry backdrop matters. The sector went through what Wickramasekara describes as a dot-com-bust equivalent over the past few years. The mechanism was straightforward: COVID and mRNA created a funding surge, but drug development takes seven to ten years, and investors got impatient. Capital rotated out, interest rates rose, and geopolitical uncertainty compounded the pain. As a reference point, the industry spends close to $200 billion a year on R&D and produces roughly 50 new approved medicines per year — a ratio that frames both the inefficiency AI is targeting and the long cycle times that made the downturn so brutal.

The mood at JPM is recovering. The XBI has been trading better, M&A has returned capital to investors, and Eli Lilly becoming the first trillion-dollar pharma company has reset ambitions across the sector. Wickramasekara notes that about 90% of attendees at JPM are not actually at the formal conference — it functions more as an industry gathering than a banking event, with most activity in satellite events. Benchling is hosting an AI and bio panel at its San Francisco HQ the same evening, with Eli Lilly, Isomorphic Labs, ProFluent, and Anthropic on the bill.