One hundred hours down to three. That’s how long it used to take Takeda’s regulatory team to produce a first draft of an IND submission, and what it takes now with Weave Bio’s AI platform in the loop. The result comes from a peer-reviewed study co-authored by scientists at both companies, and it’s the kind of number that will be hard for drug development teams to ignore.

Weave Bio, an AI-native regulatory software company based in San Francisco, announced on September 1 a formal collaboration with Takeda to deploy its platform across 14 active drug programs in multiple global markets. The partnership started with IND submission preparation and has since expanded to cover Health Authority Question (HAQ) and Response to Question (RTQ) workflows, two of the more time-intensive communication tasks in late-stage regulatory affairs.

What the study actually found

The IND phase results, published on arXiv, showed a 97% reduction in time to first draft and a 60% compression in overall submission timelines within the evaluated use case. Faithfulness to source data came in at approximately 95% under the study’s methodology. So far, the HAQ and RTQ deployment has processed more than 100 questions with no extraction errors identified in the evaluated dataset. AI answer quality averaged 68% against a target range of 50% to 70%, suggesting the system is performing at the top of what both companies had set as a reasonable benchmark.

These are meaningful numbers, but context matters. This is a single partnership, evaluated under specific study conditions. Broader generalizability will depend on how the platform performs across different therapy areas, regulatory jurisdictions, and document types. Still, a 97% time reduction in a workflow that has historically been almost entirely manual is not a small signal.

Why the human oversight piece matters as much as the speed

Takeda’s Chief Regulatory Officer Nahid Latif was direct about the intent: regulatory professionals retain responsibility for content, verification, and final decisions. That framing is not just a liability hedge. It reflects where both regulators and the industry are heading.

The FDA issued draft guidance on AI use in regulatory decision-making in January 2025. In January 2026, the FDA and EMA published joint guiding principles for good AI practice, with human accountability at the center. Weave and Takeda say they designed their collaboration around those same principles. AI handles drafting, data extraction, source traceability, and review support. Humans make the calls.

The bigger picture for regulatory affairs

Regulatory writing has long been one of the most resource-heavy parts of drug development, not because it requires rare scientific insight at every step, but because it demands exhaustive consistency, traceability, and familiarity with evolving agency expectations. Those are exactly the tasks where well-designed AI tools can take real load off expert teams.

Weave Bio has been building in this direction since its founding in 2022, with backing from USVP, Innovation Endeavors, and others. Its recent moves, including an NDA workflow built with Parexel and new global submission capabilities, suggest a company trying to cover the full regulatory lifecycle rather than a single document type.

For clinicians and health entrepreneurs watching the drug development pipeline, the practical question is whether AI-assisted regulatory workflows can compress the time from trial completion to submission without introducing new sources of error. Early evidence from this collaboration suggests the answer might be yes, provided the human accountability layer is built in from the start, not added as an afterthought.

• 97% reduction in time to first IND submission draft (approx. 100 hours to 3)
• 60% overall timeline compression across the IND preparation phase
• ~95% faithfulness to source data under study conditions
• 100+ HAQ/RTQ questions processed with no extraction errors in the evaluated dataset
• 68% average AI answer quality score, above the 50-70% target range

The full study is available at arXiv:2509.09738. More information on Weave Bio is at weave.bio.