AI-enabled pharma is set to generate big rewards for the sector. Because AI can optimize resource allocation by helping guide experimentation decisions and reduce manual workloads for researchers, it’s expected to create more than $350 billion in annual value worldwide for the pharmaceutical sector, according to a CB Insights report. That same report found that top pharmaceutical and life sciences companies already have AI platforms to accelerate and streamline their research and development processes.

But the true future of AI-enabled pharma R&D requires more than grafting AI onto existing processes. It needs a closed loop in which each experiment provides data for the next to accelerate discovery.

One life sciences enterprise recently used the Lab-in-the-Loop model to develop a fully automated research chemistry workstation linked to an AI-powered experiment database. This approach is forecasted to significantly increase R&D throughput and make small-molecule R&D more efficient.

What does Lab-in-the-Loop mean?

Lab‑in‑the‑Loop (LitL) is an AI‑powered, closed‑loop system that comprises lab instruments, experimental data, and machine-learning (ML) models. As lab instruments generate experimental data, the ML model ingests and analyzes it. The result is a continuous cycle of learning from every experiment run, including failed runs.

The model employs the results to make predictions that the LitL planning module can use to plan the next round of experiments, and that guidance gets more accurate and precise with each iteration. What sets LitL apart from past pharmaceutical and life-sciences lab automation efforts isn’t just the use of AI for continuous feedback, but its ability to massively compress discovery timelines by running hundreds of complex, self-correcting experiments in parallel. Other industries should be watching this space, too, because the Lab-in-the-Loop model can be adapted for other sectors.

Full integration across the pipeline

Each Lab-in-the-Loop can be connected to others along the organization’s R&D pipeline, from initial discovery through clinical candidate selection. The result is a set of embedded continuous-learning processes that can refine decisions and accelerate progress at each stage of the development cycle. For example, autonomous agents can monitor experiments around the clock, propose next runs, and coordinate with other agents in R&D, quality control (QC), and chemistry, manufacturing, and controls (CMC).

Multimodal datasets, processes, and orchestration

AI agents can be trained to specialize in different types of data. One enterprise might have LitL agents designed to work with small molecules, large molecules, nucleic acids, and other biologic components, as well as agents that handle assay, sensor, and outcome data. For example, one global pharmaceutical company has developed a specialized LLM to optimize mRNA sequencing and predict mRNA properties for vaccine development. The LLM was trained on codons from “more than 10 million mRNA sequences from a diverse set of organisms” and has outperformed other mRNA prediction solutions, according to researchers.

LitL agents can be orchestrated to share data from different sets when appropriate, to allow for synthesis, collaboration, and expanded learning along the pipeline. A life-sciences enterprise is using multimodal datasets to help identify lung and breast cancer patients most at risk for their cancers spreading to the brain. The company’s predictive platform analyzes clinical, genomic, imaging, and other patient data from hundreds to thousands of individuals to generate personalized reports, helping doctors identify which patients need closer monitoring for earlier detection and intervention.

Active learning creates agent-expert partnerships

Agents can automate the documentation that creates much of the workload during research and experiments. That reduces the amount of repetitive analysis and documentation the research team has to do, and leaves the human experts in charge of guardrails, experimental constraints, and major decisions.

When agents are capable of identifying the most valuable next experiments and continuously optimizing for maximum learning, they also help the human experts in the lab focus on the highest-value tasks. As these experts work, the agents can learn from their decisions and actions to improve agentic performance over time.

Business impacts of Lab-in-the-Loop

LitL can help organizations in multiple ways when it’s implemented comprehensively. The ability to ingest and analyze huge datasets quickly compresses timelines, so discovery, experimentation, candidate selection, and process optimization can all happen sooner. At the same time, organizations can better control costs and improve productivity when AI models reduce error rates, pinpoint next-best steps, and support more efficient use of lab instruments and human expertise.

LitL also allows organizations to extend the reach of their potential for innovation. There are some areas where no amount of manual labor could fully investigate a class of candidates or formulations. For example, there are millions of different macrocycle peptide compounds. AI is the only practical way to explore large numbers of these compounds for drug-development potential. That’s one reason companies in the space have seen large investments from leading venture capital firms.

A strategic path to Lab-in-the-Loop

Developing an effective LitL program requires strategy and a willingness to start small. The first step is to audit the wet lab’s instrumentation and determine how to connect that equipment to the system. The next step is to choose a single workflow to bring into the loop, such as automating protein purification. Choose metrics that will show the pilot’s effectiveness, such as error reduction or cycle-time reduction. Proactively address potential failure points. For example, the loop can’t close if model outputs are not integrated into real-time decision making, and AI models trained on biased historical assay data will fail to generalize. Once the pilot is performing well, it can be extended gradually across the R&D pipeline. Similar LitL models can be adapted for CMC and QC.

Lab-in-the-Loop enables pharmaceutical and life sciences organizations to finally close the gap between AI potential and practical application. In a rapidly evolving market, the risk of underinvesting in these capabilities far outweighs the upfront budgetary hurdles. The companies building LitL infrastructure now will own the cost-of-discovery advantage, while their competition will spend the decade catching up.