Drug Development Cut by 2-3 Years: Pharma AX Hinges on Connectivity

According to Pyo Jun-hee, Director of the AI New Drug Research Institute at the Korea Pharmaceutical and Bio-Pharma Manufacturers Association, global pharmaceutical companies estimate that AI can shorten drug development timelines by 2-3 years. Some companies are advancing their technology with the goal of eventually reducing development periods by 4-5 years. Director Pyo explained that in a drug development landscape where multiple companies compete over the same targets, shortening development time can be a critical factor in securing market leadership and business competitiveness.

This announcement was made at an event themed 'The Pharma-Bio Industry Led by AI Convergence: What Must We Prepare For?' and was reported on November 3, 2026. At the same event, Lee Ho-jun, Executive Director of Amgen's Center for Observational Research, presented on drug development and approval decision-making using real-world data (RWD) and real-world evidence (RWE).

How Far Has AI Advanced in Drug Development?

The scope of AI applications introduced by Director Pyo has expanded from small-molecule compound discovery and candidate screening to protein, antibody, and peptide design, as well as genomic analysis. The cases and application areas presented include the following.

• Insilico MedicineDeveloped a pulmonary fibrosis treatment candidate by using AI to identify disease targets and optimize compound structures, advancing it to Phase 3 clinical trials.
• GenentechUsed AI to design antibodies and iteratively learn from experimental results, improving properties necessary for development such as binding affinity and expression levels.
• Clinical TrialsAI technology is being applied to optimal patient group selection, support for patient enrollment, and management of participant adherence to trial protocols and the likelihood of achieving endpoints.

Director Pyo explained that even when AI generates numerous candidate compounds, actual drug development requires comprehensive evaluation of efficacy, toxicity, stability, and manufacturability. As a result, a method is being used where AI-generated candidates are selected for experimentation, and the resulting data is fed back into the model for further learning.

What Distinguishes AX from Adopting Individual AI Models?

The way pharmaceutical companies apply AI within their research environments is also changing. While the past focus was on introducing individual AI models at specific research stages, the trend is now shifting toward building R&D systems that connect data generation and analysis, decision-making, and experiment execution. Director Pyo described this as the AI transformation (AX) process for pharma-bio companies. His explanation, presented in sequence, is as follows.

1. The stage of utilizing AI technology
2. The stage of embedding AI into R&D workflows
3. The stage of redesigning the entire operating system around AI

Autonomous laboratories combining AI and robotics were also presented as a key example. Director Pyo explained, "When AI designs experimental conditions, automated equipment prepares and synthesizes samples, and the analysis results are then fed back to the AI model," adding that "AI uses this to set the next experimental conditions." In an actual automated lab that has been built, equipment carries out synthesis according to AI-designed conditions, and subsequent experiments are designed based on high-performance liquid chromatography (HPLC) analysis results.

Director Pyo explained, "Beyond adopting individual technologies, it is important to build a research environment that connects data, decision-making, and execution."

What Must Be Determined Before Using Data?

Executive Director Lee Ho-jun emphasized that before utilizing AI technology, it is essential to clarify what problem is being solved through data generation and analysis. He stated, "Even when using AI, it is difficult if the purpose of why and for what it is being used is unclear," explaining that research approaches differ depending on whether the goal is to support clinical trial design, secure evidence needed for drug approval, or address safety concerns.

The example he cited was a study analyzing treatment patterns for small-cell lung cancer patients across different countries. It was found that the usage rate of a particular treatment reached about 80% in the United States, but only 20-30% in Korea and Spain. Executive Director Lee explained that designing a multinational clinical trial based on the U.S. treatment environment without accounting for such differences could lead to difficulties in forming control groups and recruiting patients. He noted that an inappropriate design could extend the patient recruitment period and increase costs associated with follow-up observation.

A case was also introduced in which existing patient data was used as an external control group in the development of a treatment for a rare disease. This involved analyzing historical patient data related to a treatment for acute lymphoblastic leukemia (ALL) and using statistical techniques such as propensity score matching to construct a comparable control group. Executive Director Lee explained that while such research does not fully replace clinical trials, it can provide additional evidence for judging treatment efficacy in situations where securing a control group is difficult.

He added, "To utilize AI and real-world evidence in the drug development process, specialized personnel are needed who understand the research objectives and can assess the limitations of the data and the analysis results."

Implications for AX at Korean Companies

The point of overlap between the two presentations is not the models themselves, but the connections before and after them. Director Pyo emphasized the linkage of data, decision-making, and execution, while Executive Director Lee emphasized the purpose of use and the specialized personnel needed to assess it. Even companies outside the pharma-bio sector considering AI adoption should first map out what decisions and actions the results will lead to, before selecting a model.

The case of country-specific treatment patterns illustrates the pitfalls of directly importing overseas standards. As shown by the treatment usage rate diverging between about 80% in the U.S. and 20-30% in Korea and Spain, the same indicator can vary significantly by country—so before applying standards or models designed abroad to domestic data, it is essential to first verify the actual distribution in the local context.

There is one thing worth checking within your own organization this week. For each AI initiative already adopted or under review, check whether you can write down in a single line who makes what decision based on its results. If that line remains blank, the priority is to redefine the purpose of using AI before focusing on the model itself.

Source: "Drug Development Cut by 2-3 Years"...AI in the Pharmaceutical Industry Expanding from Design to Clinical Trials