Skip to content
ClaudeClaude

Claude Fable 5: Working At The Frontier

We asked 5 teams what’s now possible with Claude Fable 5. In this film, teams at Thomson Reuters, Hebbia, Cognition, Cursor, and Base44 describe the before and after in their own words. Working at the Frontier is an ongoing series where the teams building with Claude tell their own story. Learn more about Claude Fable 5: https://www.anthropic.com/claude/fable Office Hours: https://claude.com/office-hours

Jul 6, 20262mWatch on YouTube ↗

CHAPTERS

  1. 0:00 – 0:14

    Fable makes previously unthinkable problem classes approachable

    The episode opens by framing Fable as a step forward that expands the set of problems people can realistically attempt with AI. It sets the theme: new capabilities unlock new workflows rather than just faster versions of old ones.

    • Fable changes what kinds of tasks feel feasible to attempt
    • Unlocks problem areas people weren’t actively targeting before
    • Positions the rest of the video as proof via real-world customer examples
  2. 0:14 – 0:45

    Hebbia: compressing the finance deal lifecycle with higher accuracy

    Hebbia describes how Fable impacts finance work by speeding up and tightening the deal lifecycle, which directly affects competitiveness. On a rigorous internal dataset, they report roughly a 20% increase, suggesting strong performance under difficult evaluation conditions.

    • Finance value prop: faster deal cycles = competitive advantage
    • ~20% increase on a rigorous, difficult dataset
    • Performance holds as users push more complex questions
    • Correct answers encourage expanding use cases
  3. 0:45 – 0:47

    Scaling complexity: users push boundaries as reliability improves

    Building on Hebbia’s results, the narrative highlights a feedback loop: as the model returns correct answers on harder tasks, teams become willing to attempt even more complex problems. This frames adoption as confidence-driven escalation.

    • Complexity increases over time as users gain confidence
    • Reliability enables experimentation on harder tasks
    • Boundary-pushing is driven by consistent correctness
  4. 0:47 – 1:09

    Cursor: a coding agent that stays on track with minimal nudging

    Cursor contrasts prior workflows—constant prompting and steering—with Fable’s ability to maintain direction during deep problem-solving. The key claim is reduced overhead: developers can delegate more and intervene less.

    • Cursor uses Fable for professional software development
    • Less need to repeatedly nudge or course-correct the model
    • Stronger “stay on track” behavior during long tasks
    • Enables deeper problem-solving with minimal instructions
  5. 1:09 – 1:39

    Base44: automating system-prompt rebuilds and expanding product scope

    Base44 explains how Fable dramatically reduced the effort to rebuild a system prompt—from multiple top engineers over days to an automated run completed in hours. The outcome is near-complete coverage (90–95%) and the confidence to pursue new product directions.

    • Base44 positions itself as a full-stack “vibe coding” platform
    • System prompt rebuild estimated: 3 engineers for a couple days
    • With Fable: ~4 hours to get ~90–95% of what was needed
    • Unlocks new product areas the team previously avoided
  6. 1:39 – 1:45

    From fear to exploration: new possibilities enabled by higher leverage

    This segment emphasizes the psychological and strategic shift that comes from improved tooling: teams move from avoiding certain initiatives to actively exploring them. The focus is on how capability changes decision-making, not just speed.

    • Higher leverage changes which projects feel safe to attempt
    • AI capability affects roadmap and ambition
    • “New areas and possibilities” become viable
  7. 1:45 – 2:16

    Thomson Reuters: using Claude for high-stakes legal drafting

    Thomson Reuters describes the legal domain’s demand for precision and correctness, using motions as an example of complex documents that typically require days or weeks. The claim: with Claude, tasks once requiring too much context and accuracy become plausible for AI assistance.

    • Company context: expert information that must be right
    • Legal motions are among the hardest documents to draft
    • Historically too much context/precision for AI to handle
    • Claude shows real potential to support this workflow
  8. 2:16 – 2:24

    Compounding quality in legal solutions as models improve

    Continuing the legal theme, the episode notes a compounding effect: as foundation models improve, the quality of downstream legal products rises disproportionately. This suggests iterative model upgrades translate into accelerating product capability.

    • Model improvements translate into better end solutions
    • Compounding effect on legal product quality
    • Long-term trajectory matters as much as current performance
  9. 2:24 – 2:55

    Cognition (Devin): surviving massive enterprise toolchains

    Cognition discusses deploying AI in enormous engineering organizations with decades of internal tools—an environment where many models fail. They describe Fable as a step change that increases trust and enables the model to take on larger projects in real-world enterprise settings.

    • Cognition built Devin, positioned as an AI software engineer
    • Customers can have ~30,000 engineers and extensive internal tooling
    • Typical models struggle with legacy/internal environments
    • Fable is described as a step change and increases trust for bigger projects

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.