# Anthropic's Thariq Shihipar on Fable: A Field Guide _Anthropic's Thariq Shihipar presents a 'Field Guide to Fable,' detailing its 'grown' nature, the evolution of AI agents, and how to leverage it to discover 'unknown unknowns.'_ **Published:** 2026-07-06 **Source:** https://www.startuphub.ai/ai-news/ai-research/2026/anthropic-s-thariq-shihipar-on-fable-a-field-guide --- Thariq Shihipar, a member of technical staff at Anthropic, recently presented a "Field Guide to Fable" at the AI Engineer World's Fair. In his talk, Shihipar provided insights into Anthropic's Fable model, its development, and how to effectively work with it. He highlighted the evolving nature of AI agents and the importance of understanding their capabilities and limitations. Anthropic's Fable ModelCore a 'grown, not designed' AI modelFrom the article 4 mentionsIn his talk, Shihipar provided insights into Anthropic's Fable model, its development, and how to effectively work with it.developed viaIterative DevelopmentContextcontinuous learning and refinement processFrom the article 2 mentionsShihipar described Fable as a model that is "grown, not designed." This metaphor emphasizes the iterative and adaptive nature of its development, contrasting it with traditional software engineering.enablesEvolving AI AgentsContextprogression through distinct functional stagesFrom the article 5 mentionsHe highlighted the evolving nature of AI agents and the importance of understanding their capabilities and limitations.allowsDiscover Unknown UnknownsEffectleveraging Fable's capabilities to find unexpected insightsFrom the article 3 mentionsA significant portion of Shihipar's talk focused on the concept of "unknown unknowns" and how to leverage Fable to uncover them.leads toRethink TradeoffsContextmoving beyond 'good, fast, cheap' paradigmsencouragesEmbrace FutureOutcomeimplementing Fable and its emergent capabilities ## Understanding Fable: A Growing Model Shihipar described Fable as a model that is "grown, not designed." This metaphor emphasizes the iterative and adaptive nature of its development, contrasting it with traditional software engineering. He explained that working with Fable involves a continuous process of learning and refinement, akin to nurturing a biological organism. This approach allows for emergent capabilities and a more intuitive interaction model. ## The Evolution of AI Agents The presentation traced the progression of AI agents through distinct stages. Initially, agents primarily functioned through chat interfaces, requiring explicit context for every task. This evolved to agents with "given arms to execute," capable of performing actions through code. The latest stage, exemplified by Fable, is an agent that "acts proactively." Shihipar noted that this advancement was driven by the need to better manage context, as early models struggled with the sheer volume of information required for complex tasks. ## Navigating the Unknowns with Fable A significant portion of Shihipar's talk focused on the concept of "unknown unknowns" and how to leverage Fable to uncover them. He proposed using a framework, akin to an "Unknown Matrix," to categorize different types of knowledge gaps: known-knowns, known-unknowns, unknown-knowns, and unknown-unknowns. Shihipar suggested that Fable can be used as a tool to identify these unknowns by prompting it with specific questions or scenarios. For instance, asking Fable to identify coding challenges or to generate multiple design directions for a dashboard can reveal aspects that the user might not have considered. ## Rethinking Tradeoffs: Beyond "Good, Fast, Cheap" Shihipar presented a compelling argument that in the age of advanced AI, the traditional "Good, Fast, Cheap: Pick Three" mantra is becoming obsolete. He stated, "We believe that tradeoffs are not real." This provocative claim suggests that AI models like Fable can help achieve multiple desirable qualities simultaneously, such as being both "good" and "fast," or "good" and "cheap." He illustrated this with the example of using Fable to generate code and documentation in hours that would previously take weeks, thus challenging conventional project management constraints. ## Implementation Notes and Future Directions Shihipar shared practical advice for developers working with Fable, emphasizing the importance of "implementation notes." He recommended keeping a log of deviations from the plan and choosing conservative options when encountering edge cases. This practice, he explained, helps in understanding the model's behavior and refining prompts effectively. He also touched upon the concept of "blindspot passes," where users can ask Fable to identify potential blind spots in their understanding or work, thereby improving prompt engineering and uncovering hidden complexities. ## Conclusion: Embrace the Future Shihipar concluded by reiterating that "the only way to prove that agents work is to do the best work of our lives, faster than ever before." He encouraged the audience to embrace the capabilities of models like Fable, to "go explore" and "make it real." His message conveyed a sense of optimism about the future of AI development, highlighting the potential for these advanced tools to unlock unprecedented levels of productivity and creativity. --- Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory.