Current AI pricing, whether per-seat or token-based, fails to capture the true value delivered. The industry needs to move towards models that are more closely aligned with the work performed and outcomes generated, even if pure outcomes-based pricing is challenging for generalist tools like ChatGPT.
While foundation model companies have advantages like early access to new models, startups can carve out defensible niches. By focusing 100% on specific verticals and leveraging proprietary data not available on the open web, startups can build valuable businesses in markets too small for incumbents to pursue.
The paradigm of user interaction with software is shifting. The future lies in proactive AI that can initiate conversations and actions, moving beyond the current reactive, prompt-based model. This will likely lead to a decline in the importance of traditional graphical user interfaces for many applications.
Building AI products requires a different approach than traditional software. The 'eval' (evaluation) has become the new product specification, defining success and guiding development. This data-centric approach is a core, yet underhyped, aspect of creating effective AI systems.
Keep pulling the thread on Peter Dang.