World's Fair 2026
Field Guide to Fable — Thariq Shihipar, Anthropic
Overview
This talk introduces Fable, a new class of Anthropic models, framing it as an expansion into an open world of AI capabilities. It emphasizes that models are grown organically rather than strictly designed, and that our understanding and interaction methods shape their potential. The presentation aims to provide a guide for working with these advanced models, encouraging users to explore their capabilities and push beyond conventional limitations.
Who should watch
- AI Engineers and Builders
- Product Managers
- Those exploring advanced AI model capabilities
- Developers looking to enhance productivity and build ambitious projects
- Individuals interested in the evolution of AI agents and interaction paradigms
Key takeaways
- Models like Fable exhibit "capability overhang," meaning they possess advanced abilities that are unlocked through specific tools and interaction methods, such as code execution for complex queries.
- The way we prompt and harness models, particularly their system prompts, needs to evolve as models become more capable; newer models may benefit from simpler prompts and less constraint.
- Interacting with advanced models requires users to identify and address their own "unknowns" – aspects of a problem or domain that are not fully understood or specified, which Fable can help uncover.
- Techniques for finding unknowns include "blind spot passes" to identify overlooked areas, brainstorming prototypes for design exploration, and using the model to interview the user for clarification.
- Providing models with references, such as existing code or mock-ups, acts as a "map" that helps them understand desired outcomes and navigate complex tasks more effectively.
- The progression of model capabilities is evident in how they handle output formats, moving from simple text to structured Markdown and now to complex HTML reports, and in their ability to ask clarifying questions.
- Working with advanced models can lead to a sense of both gain in productivity and a re-evaluation of traditional trade-offs, enabling more ambitious projects to be completed faster.
- The ultimate goal is to generate value by leveraging AI, which requires persistence and iteration to discover impactful applications, rather than solely focusing on the building process.
Notable quotes
*The map is opening up, you know, like you were playing like an RPG and you've been on the tutorial. And now you get to the point where the open world starts.*
*Claude gets smarter in spiky ways. So, it doesn't just remember every Pokémon and reason through it, but if you give it the code execution tool, it can find the two Pokémon that end with AW.*
*The map is not the territory. When I'm working on a coding problem, the plan and prompt and spec that I have in my mind is the map, right? But the territory is the actual code base, the real world, the constraints that Claude needs to navigate.*
Unofficial community note. Prefer the recording for nuance.