Strategy question

OpenAI is seeing demand from law firms for workflow-specific products, while legal-tech partners want to own the application layer and model quality is still uneven across tasks. How would you decide which legal workflows OpenAI should build directly, which should be enabled via APIs/platform, and which should be left to ecosystem partners? What evidence and decision criteria would you use?

Practice this question out loud. An AI interviewer asks it, follows up like a real interviewer would, and scores your answer. Type or speak.

What this question tests

Platform strategy for deciding build versus enable versus partner across a fragmented vertical with mixed model reliability.

Segment legal workflows by two axes: how much they require deep vertical domain expertise, and how reliable current model quality is for that specific task.
Build directly only where model quality is already strong and the workflow is foundational enough to matter across most legal teams, like contract summarization or clause extraction.
Enable via API and platform where quality is promising but workflows are too varied for one product to fit all firms, letting partners customize on top of a solid foundation.
Leave to ecosystem partners where deep vertical workflow expertise, like litigation strategy specific to a practice area, matters more than the underlying model, and partners already have that domain trust.
Gather evidence through direct interviews with law firms and legal-tech partners on where model output is already trusted versus where human expertise remains essential, and revisit the split as model quality improves over time.

Proposes a concrete two-axis framework, domain specificity and current model reliability, rather than an arbitrary split between build, enable, and partner.
Gives specific examples for each category, contract summarization to build, varied firm-specific workflows to enable, and deep litigation strategy to leave to partners.
Names direct evidence gathering, interviews with firms and partners, as the way to validate the framework rather than assuming it internally.
Acknowledges the split is not permanent and should shift as model quality improves in currently weaker task categories.

Proposing to build everything directly, ignoring that legal-tech partners have real domain trust and expertise OpenAI lacks.
Treating model quality as uniform across all legal tasks instead of recognizing it varies significantly by workflow.

Which specific legal workflow would you prioritize building directly first, and why?
How would you handle a partner who feels threatened by OpenAI building adjacent to their product?