OpenAI has acquired Orchestra Bio, a service whose AI agents track drug development and identify which timelines will be affected when one part of a program goes wrong

OpenAI has acquired Orchestra Bio, a service whose AI agents track drug development and identify which timelines will be affected when one part of a program goes wrong

OpenAI announced the acquisition on October 20 without disclosing the price. Founded in 2022, Orchestra Bio lets users assign AI agents (programs that carry out multistep tasks autonomously) to track drug development progress and timelines. OpenAI will integrate the team and technology with Rosalind, its models and tools for the life sciences.

OpenAI estimates that taking a drug from target discovery (identifying the molecule the drug is intended to act on) to approval in the US takes an average of 10–15 years. Drug development involves interdependent activities, including preclinical studies, drug batch manufacturing, and clinical trials, carried out by different departments and contractors, often in disconnected systems. In OpenAI’s example, a failed drug batch postpones the start of a clinical trial. According to PharmTech, departments and contractors see only their own portions of the schedule, so the consequences are often recognized late. Orchestra’s agents identify the activities and timelines affected by a problem and compare possible responses; scientists and program leaders decide what to do before the delay occurs.

In a 2025 blog post, Orchestra’s founders wrote that they had spent years moving scientific data from paper to the cloud and had seen fragmented planning and missed signals derail scientific work. They believed that managing drug development needed software tools as capable as those used to manage scientific data. The service represents a program plan as a graph, a diagram connecting each activity to those it depends on, and links it to a financial model showing how much money will be needed and when. Signals about progress are buried in emails (“the manufacturer mentioned quality control problems”), meeting transcripts, and PDF reports; language models read these documents and assess whether the program is progressing toward its planned scientific milestones. What-if scenarios work like branches in software development: users copy a development program or budget and explore alternatives in the copy.

Investor 8VC, which led the 2025 funding round, writes that the team began by asking how to represent a development program, and built its agents on that foundation. In August 2025, Orchestra emerged from stealth with about $20 million in funding.

Rosalind has two components: the GPT-Rosalind model helps scientists work through questions in biology and chemistry, while Rosalind Workbench connects it to scientific tools and data so researchers can analyze results and plan experiments. As Decrypt puts it, Rosalind helps researchers decide what to test, while Orchestra is intended to manage what happens after that decision.

OpenAI’s competitor Anthropic is also moving into science: it acquired Coefficient Bio, a biology AI startup, in April and confirmed that it had its own biology laboratory in September. According to Axios, both companies want to expand beyond programming and cybersecurity.

Megha Sinha founded Kolter AI, a company that uses AI to plan changes in drug manufacturing and regulatory registration. In July, she told PharmTech that funding had gone into AI for drug discovery and planning pilots, while implementing changes across departments and markets remained manual: “Strategy moves at boardroom speed; execution still moves at spreadsheet speed.” In its article on the acquisition, PharmTech places Orchestra in that same gap between planning and execution.