Director, Model & Agentic Learning

About The Role

Johnson & Johnson Innovative Medicine is recruiting a Director, World Model & Agentic Learning to join our Data, Data Science & AI organization. This is a newly created leadership role within the Generative AI organization, reporting directly to the Head of Generative AI.

You will lead the AI science team that builds our enterprise world model and agentic‑learning capability for the R&D agentic AI platform, a reusable, expert‑curated foundation that domain teams customize, together with the mechanisms by which it improves with use. This is a durable, product‑agnostic capability. You will devise the approach, set the technical direction, and lead the team that delivers it.

• World Model: how agents represent and reason against accumulated domain understanding, instead of re‑deriving everything from raw sources on each task.
• Agentic Learning: how that understanding grows with use, i.e. getting better from operation, rather than from retraining foundational models.

• Accumulate, don’t re‑derive. Agents build on prior understanding instead of re‑reading every source, dataset, and prior result on each task.
• Know its own boundaries. The system can say what it knows, what it doesn’t, and how confident it is.
• Reason consistently. Expert judgment is applied uniformly across thousands of cases, not improvised per query.
• Improve from operation, not retraining. Every run, every expert correction, and every decision outcome makes the next result better.
• Compound across workflows. Knowledge earned in one domain or workflow surfaces automatically wherever else it is relevant.
• Keep experts authoritative. Experts own the judgment; the system does the maintenance, never the reverse.
• Stay fresh and honest. Contradictions, gaps, and staleness are surfaced, never silently buried.
• Be auditable and accountable. Every conclusion is traceable, decisions can be reconstructed and judged against their outcomes, and institutional understanding survives turnover.

• Design how agents represent accumulated domain understanding and reason against it, rather than re‑deriving knowledge from raw sources on each task.
• Build mechanisms for the system to represent its own confidence, boundaries, gaps, and contradictions explicitly.
• Ensure knowledge earned in one domain or workflow compounds and surfaces wherever else it is relevant.
• Serve the representation to the reasoning agents as queryable, grounded knowledge with provenance and confidence, and curate what they propose back by validating, deduplicating, and resolving conflicts.
• Build on the platform’s existing context, memory, and governed data layers, referencing canonical entities rather than rebuilding data pipelines.

• Design the mechanisms that turn operation into improvement. For example, active learning from expert corrections, memory‑based / in‑context learning, or outcome‑driven refinement.
• Make every run, expert correction, and decision outcome a signal that improves the next result.
• Keep institutional understanding fresh and honest as sources, evidence, and experts change over time.

• Partner with scientists and domain experts so their expertise becomes something the system can apply consistently at scale.
• Keep experts authoritative: the system maintains and applies their judgment; it never overrides it.

• Define and prove the accountability bar: demonstrate that the system produces better decisions over time.
• Make every conclusion auditable and reconstructable, and judge decisions against their real‑world outcomes.
• Partner with the J&J Technology, Generative AI evaluation, and the AI operations teams, consuming their per‑decision outcome signals as the learning signal and validating decision‑quality improvement rigorously.

• Recruit, build, and lead a team of 4–8 AI scientists.
• Attract, develop, and retain top talent in continual learning, knowledge representation, and agentic systems.
• Establish a culture of scientific rigor, ownership, and accountability within the team.

• Not a generation‑first…