The discussion highlights a successful playbook for vertical AI: start with a narrow, high-value 'context complete' workflow (e.g., transactional legal work) to prove value, then expand into a broader platform vision (e.g., an 'IDE for lawyers' or a full 'agent development life cycle'). This focused approach allows for deep integration and builds initial customer trust.
Building on rapidly evolving foundation models requires a dynamic strategy that anticipates the platforms' roadmaps. The speakers stress the risk of building features that will be commoditized, advising a focus on defensible areas like deep workflow integration, proprietary data access, and superior, domain-specific user experience.
The rapid pace of AI innovation has rendered traditional long-term roadmaps obsolete. Companies are shifting to highly agile, continuous planning cycles, with Sierra re-evaluating every 4-6 weeks and Harvey using a three-month cycle. This model requires a cultural acceptance of ambiguity and rapid change.
In enterprise AI, deep, continuous collaboration with customers is non-negotiable. Effective strategies include embedding teams with clients (Harvey's 'forward deployed' model), sharing roadmaps for feedback, and identifying 'frontier customers' whose advanced needs pull the product into the future.
Superior user experience (UX) is a critical and often underrated differentiator in AI applications. It can be used to guide the model, compensate for its weaknesses, build user trust, and drive adoption through thoughtful localization (e.g., British English) and personalization.
Keep pulling the thread on Aatish Nayak, Sachi Shah.