Benchling offers cloud-based research and development software that helps biotech companies develop products quickly and efficiently while remaining compliant. According to Alan Pierce, a software engineer, the company operates in a “high-stakes domain” in which precision and strict data management are critical, which is why it gives its customers full ownership of their use of AI.
At the same time, the opportunity for AI to speed up science is massive. Benchling’s analysis shows scientists spent up to 25 percent of their time just on capturing and aggregating data, tasks that can be automated with the right AI solution.
Pierce’s team builds the company’s AI platform, providing guidance to other teams on when AI is most appropriate to use — and when it’s unnecessary. Like every Benchling employee, he and his peers are encouraged to view AI as an innovative tool for solving tough customer problems, not as a replacement for traditional, reliable approaches.
“We work hard to make sure we’re applying AI in situations where it’s truly useful, not just building AI for the sake of it being AI,” Pierce said.
Benchling’s embrace of AI is grounded in three key ingredients: “excitement, curiosity, and a healthy level of skepticism.” Guided by these principles, Pierce shared, he and his teammates continuously craft new ways to make it easier for biotech companies to deliver breakthrough products — all while cultivating a shared understanding of AI’s strengths and limitations.
“Customer trust is at the core of our approach to AI,” Pierce explained.
About Benchling
Benchling is a unified, cloud-based platform built to accelerate how biotech companies discover, develop, and scale life-changing products. The platform brings together scientific workflows, structured data, and collaboration tools in one place — helping teams move faster and with greater confidence.
Core products include Benchling Bioresearch, which enables scientists to plan and manage experiments, biomolecules, and results in a shared environment, and PipeBio, a specialized suite of sequence analysis tools that supports biologics discovery teams through lead selection and optimization.
Together, these tools — and others across the Benchling R&D Cloud—provide a connected system of record that supports the full lifecycle of modern biotech R&D.
The customers know best how and when AI is appropriate, so at Benchling, they treat all AI features as opt-in.
“We let the customer decide which providers to use, and customer data is never used for training,” Pierce added.
An Exciting New Development
Benchling is a unified home for a wide variety of interlinked scientific data, and efficiently capturing and structuring that data is critical to powering scientific insights. One of its newest tools, the Data Entry Assistant, does exactly that. It allows customers to ingest data from sources like PDFs, Word documents and unstructured Excel sheets, all without needing to reformat the data for machine ingestion.
When Pierce pitched the idea for this solution last year, no one was certain it would even work, given that the generation of LLMs at the time likely wasn’t capable of powering this feature in a way that meets real-world customer needs. Yet as his team developed more techniques and newer models were released, the feature started to work, and after ample foundational work and a few failed experiments, he and his peers found a breakthrough when they reworked the mechanism for splitting up the task into smaller pieces.
“That one tweak brought the scale, reliability and latency to a point that we can stand behind,” Pierce said.
The new solution has already proven to be a game-changer for the company’s customers, enabling them to spend less time on data entry and unlock data analysis and decision-making options that weren’t possible previously.
Pierce said he’s especially proud of the cross-model verification system that his team built for the data entry assistant. He explained that, rather than simply producing an AI-generated response, the tool runs the same task through multiple AI systems and cross-references the different model results to ensure accuracy. His team runs different model families, such as Claude and GPT, for different operations to ensure that they’re independent.
“This doesn’t just detect problems; variations on the strategy can auto-correct errors and significantly improve latency for simpler cases,” Pierce noted.
He believes that this strategy captures how working with LLMs requires a different mindset from traditional software — one that might be closer to how a structural engineer approaches bridge-building. Unlike code, he explained, a bridge isn’t “right” or “wrong;” rather, it’s made up of materials that are never perfect yet come together seamlessly to create a reliable solution.
“By characterizing the tools available and combining them thoughtfully, we can build up a combined system with very high reliability, even if its individual components may never be fully perfect,” Pierce said.
Keeping Pace With Science and AI
Technologists at Benchling don’t just get the chance to sharpen their technical skills — they get to immerse themselves in science, too.
Pierce said that he often learns new scientific facts and capabilities while collaborating with the company’s customers. For instance, he recently learned that viral DNA can contain overlapping reading frames, so that the same portion of DNA can code for multiple proteins based on where the translation starts, creating what seems like a clever puzzle in the DNA.
According to Pierce, having in-house scientific experts at Benchling makes it easier to engage in high-level conversations with customers, as it enables team members to broaden their knowledge while offering optimal support. This is especially important in LLM work, where deep domain knowledge is often critical.
“If you want a scientific agent to make the best decisions, you need to be able to step into the mind of a scientist,” Pierce said.
“If you want a scientific agent to make the best decisions, you need to be able to step into the mind of a scientist.”
As he and his peers stay informed scientifically, they also stay up to date on the latest advancements in AI, which he admits can be difficult at times, given the technology’s rapid evolution. To keep their AI skills sharp, they engage in networking events and participate in the company’s internal hackathon, where they get to test out new ideas. During the company’s most recent hackathon, Pierce explored use cases and technical details behind LLM-driven code execution.
To track up-and-coming AI models, his team has connected their evaluation test suite to a variety of different model providers, including ones that can’t be used in production yet, for new model and feature releases.
“This helps us easily keep tabs on the wider developments in the space to know when a model is promising enough that we should work to bring it to production,” Pierce explained.
Stay Informed About AI
As AI continues to evolve, Pierce and his teammates do their best to keep up with its transformation by keeping others informed about its applications. For instance, he shared, when his team rolled out ChatGPT and Gemini internally, they set up a #building-with-ai Slack channel to share tips and troubleshoot. His team felt that it was important to give others knowledge around how to best use the technology, considering many employees lean on AI for a variety of purposes, such as preparing example data for customer demos, ramping up on scientific topics, troubleshooting technical problems and summarizing notes.
The Chance to Join a Noble Mission
For Pierce and the rest of his team at Benchling, AI has unlocked a myriad of opportunities, giving the company’s customers more opportunities to work more effectively — and make a major impact.
Pierce said that this mission to accelerate significant scientific work is one of the main reasons why he decided to join Benchling. The company’s partnership with some of the world’s largest pharmaceutical companies gives him and others the chance to see the real-world challenges organizations face in developing new drugs and therapies and find the best solutions to make this work easier.
Pierce has had the chance to partner with organizations driving a wide range of critical causes, including the development of childhood cancer treatments. So every day that he and his peers show up to work, they’re not just writing code and performing tests — they’re helping others change the world.
“The more we can speed up their science, the sooner these drugs and therapies make it to patients and save lives,” Pierce said.
