Chemistry provides a demanding environment for AI development: chemical data span many instruments, simulations, scales, notebooks, repositories, publications, and proprietary archives. AI methods have already accelerated chemical discovery, but the next step is to design AI for chemistry from chemistry outward rather than adapting tools built for unrelated domains.

This NSF workshop will bring chemistry, AI, data infrastructure, automation, software, publishing, and industry stakeholders together to define a shared vision and actionable pathways for coordinated data handling, methods development, and infrastructure. The long-term objective is to make experimental and computational chemical data, models, and tools more accessible, searchable, interoperable, and shareable.

The meeting is designed as a roadmap workshop rather than a research symposium. Participants will work toward concrete pilot concepts, writing responsibilities, and a community-facing report that can guide near-term action and longer-term coordination.

1. Start with the big science questions: what should AI help chemists do soon, and what should it make possible over the next decade?
2. Figure out what data we actually need, including experimental data, simulation data, metadata, provenance, uncertainty estimates, and negative results.
3. Ask what the models need to be good at, from better representations and generative models to benchmarks, uncertainty tools, active learning, physics-aware methods, and interpretable AI.
4. Decide what should be shared across chemistry and what should stay specialized for particular subfields, simulations, journals, repositories, or industry settings.
5. Leave with a few pilot ideas that people can actually act on within 1-2 years, while still pointing toward the longer 5-10 year vision.