How to Compare Fable 5.1 vs Astra for App Builds

TL;DR

Astra delivered a usable Mac clipboard app faster and with fewer tokens, while Fable produced a distinct design that required more iterations to reach the same usability. The video explains evaluating speed, feedback, and how second-round prompts shape final outcomes, helping you choose where to apply each model.

Transcript

I gave one prompt to two models and I got two completely different apps, both trying to solve the exact same problem. Copy paste. Hell. Fable built ledge, a narrow list that slides in from the right side of my screen. Astra decided on shelf, a wide bar across the bottom with nice big cards and a preview pane. They both got the exact same brief. The... Read More

Key Insights

• Astra can deliver a usable app quickly, with lower token use and faster iteration, which accelerates overall development. This speed enables discovering additional requirements and refining the product in fewer cycles.
• Astra’s design emphasizes bottom-up interaction early, using a bottom shelf metaphor that users can adopt quickly, including clear copy success feedback and drag repositioning for layout flexibility.
• Fable provides meaningful, distinct design choices that feel different from Astra, prompting explicit follow-up questions and different edge-handling decisions that shift the design space.
• Second-round prompting matters more than the first prompt, as it allows the model to adapt to evolving requirements and user feedback.
• Speed of iteration directly impacts the quality of the final product because early builds expose issues the prompt may not anticipate.
• The ability to express precise adjustments to layout, shortcuts, and feedback reduces cognitive load and makes the tool feel more reliable in practice.
• Neither model wins every job; the best approach is to route different aspects of the task to the model best suited for them, creating a multimodel workflow.
• User testing with a busy person validated practical utility beyond mere functionality, confirming product viability and adoption potential.
• More videos from this channel:
  • What Is Real AI Misalignment Beyond Skynet?
  • How AI Doom Fades Into Reality and What Matters
  • What Is the Continuum of AI Capabilities and Risks?
• More from this channel: AI News & Strategy Daily | Nate B Jones

Questions & Answers

Q: What prompt did the author give to both models?

The author asked both Claude Fable 5.1 and GPT-6 Astra to build a native Mac clipboard manager that remembers copied text, images, and links, with a keyboardshortcut to pull back items. The goal was to create a seamless copy paste experience without requiring memory of the best tool, while allowing further iterations if needed.

Q: Why is the second round of prompting important?

The second round of prompting matters because it reveals preferences that could not be captured in a five-line prompt. It lets you refine edge behavior, confirm successful actions, and adjust interfaces based on actual usage feedback, which improves usability and reduces cognitive load during real work.

Q: What were the two apps or interfaces named in the video, and how did they differ?

The two interfaces were Shelf and Ledge. Shelf is a wider bottom shelf with large cards and a preview pane, designed for quick visual access, while Ledge is a narrow list that slides in from the right, offering a more compact, code-like presentation. They represent different form factors for the same task.

Q: How did speed affect the final outcome?

Speed affected the final outcome by allowing more rapid exploration and error correction. Faster initial builds let the creator try changes sooner, notice issues, and guide subsequent iterations with fresh insights, ultimately shaping a more usable product in less time.

Q: What small but important UX detail did Astra handle well in version 1.1 and 1.2?

Astra added a visible copy success message after a copy action and improved keyboard behavior to wait for key release before returning focus, preventing unintended interactions with underlying apps and reducing user confusion during repeated actions.

Q: How did the author feel about the feedback loop between the models and the user?

The author felt that a productive feedback loop is crucial, as it allows the user to steer the development with concrete preferences and real-world use, turning model outputs into practical, adopted tools rather than theoretical demonstrations.

Q: What was the test user’s reaction to the final products?

The test user, described as very busy, expressed that the app was something she would love to use, installing it immediately and starting to rely on it for daily tasks, which validated the solution’s real-world utility beyond the demonstration.

Q: Why does the author advocate a multimodel approach?

The author argues for a multimodel approach because different models excel at different aspects of a task. By assigning the right model to the right job, you capture a broader range of intelligent outputs, leading to a better overall solution than relying on a single model for everything.

Summary & Key Takeaways

• In the video, two models are given the same task to build a native Mac clipboard manager, resulting in two different app concepts named Shelf and Ledge. The creator emphasizes that speed and iteration count influence final usability more than initial intelligence. The takeaway is to use each model where its strengths fit the task, and to iterate quickly to refine requirements.

• The author highlights practical design decisions such as hotkeys, visual layout, and user feedback like copy confirmations. These small details determine how useful the tool feels in real work, not just how it is built. The comparison shows how differing form factors and prompts lead to divergent but valuable outcomes.

• A key message is that choosing between models is job-specific. The second round of prompting often reveals preferences and unmet needs, enabling better alignment with real user workflows. The video argues for multimodel strategies and treating models as tools to optimize a solution space rather than winners of a single benchmark.