Beyond Chat: Bringing Models to the Canvas • Lu Wilson • GOTO 2025

Beyond Chat: Bringing Models to the Canvas • Lu Wilson • GOTO 2025

by Lu Wilson

Applied Sciences & Engineering Computing & Cybersecurity UYZHuman–computer interactionUYZGUser interface design and usability
🎙 Lu Wilson 👥 1.1M 📅 November 19, 2025 ⏱ 55 min 👁 2K 📄 expert opinion 🧭 2026-08-06
Available in: English (current) Français

Keywords

canvasAI interactiontldrawmake realfeedback loop

Summary

Lu Wilson, a software engineer at tldraw, presents a talk at GOTO Copenhagen 2025 about moving AI interaction beyond chat-based interfaces to a canvas environment. He draws parallels with the evolution of computers and smartphones, which initially used text-based interfaces before transitioning to graphical canvases. Wilson argues that AI is currently stuck in a ‘CLI mode’ with chat as the primary interaction method, and he demonstrates tldraw’s experiments to change this. He shows how tldraw’s canvas is hackable, allowing users to embed web content and interact with it. The core of the talk is the ‘Make Real’ feature, which sends screenshots of drawings to AI models to generate functional websites. Wilson demonstrates a feedback loop where users can annotate the generated output and iterate with the AI. He also shows sending multiple drawings to create multi-page websites. Throughout, he highlights challenges such as AI models not following instructions precisely, and he uses humorous strategies like repeating requests. The talk concludes by emphasizing the potential of canvas-based AI interaction and encourages developers to explore beyond chat.

175 words

Critical Evaluation

The talk provides a compelling vision for the future of AI interaction, grounded in practical demonstrations from tldraw. Wilson’s argument that AI interaction is currently limited to chat, similar to early computers and smartphones, is well-structured and supported by historical analogies. The live demos are impressive and effectively illustrate the potential of canvas-based interfaces. However, the talk is primarily an expert opinion based on the speaker’s own work, with limited external validation or rigorous scientific evaluation. The demos, while engaging, are anecdotal and may not represent generalizable results. The speaker acknowledges challenges, such as AI models not following instructions precisely, which adds credibility but also highlights the limitations of current technology. The sources cited are primarily links to the speaker’s social media and the conference, with no academic references. The talk is more of a showcase of innovative ideas than a rigorous scientific study. The title accurately reflects the content, and the talk is well-delivered with humor and clear explanations. Overall, it offers valuable insights into a novel approach to AI interaction, but its scientific rigor is moderate.

178 words

Title / Content Match

The title accurately reflects the content, focusing on moving AI interaction from chat to a canvas-based interface.

Quality & Reliability

7/10

The talk is based on the speaker's direct experience at tldraw, demonstrating real prototypes and experiments. The claims are supported by live demos, but there is limited external validation or peer-reviewed evidence. The speaker acknowledges limitations and challenges, which adds credibility.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

Concurring Sources

  • tldraw — The company's official site, supporting the canvas-based approach.
  • GOTO Copenhagen 2025 — Conference site confirming the talk's context.

External References

Contribution & Novelties

The talk presents a novel approach to AI interaction by moving beyond chat to a canvas-based interface, demonstrated through tldraw’s ‘Make Real’ feature. This allows users to draw and iterate with AI models in a visual feedback loop, which is a significant departure from traditional text-based interactions. The talk also highlights the hackability of the canvas, enabling integration of web content and multi-page generation.

Pour aller plus loin :

  • tldraw — Official tldraw documentation and SDK for building canvas applications.
  • Mother of All Demos — Historical context for the shift from text to graphical interfaces.
  • Generative AI — Overview of generative AI models and their applications.
  • Human-Computer Interaction — Field of study relevant to designing interactive systems.

117 words

Radar Profile

The radar profile shows balanced scores across all dimensions, with slightly higher ratings for information quantity and quality, reflecting the talk's rich content and practical demonstrations. The technical level is moderate, suitable for a general developer audience, while reliability is solid due to the speaker's direct experience.

Reliability 7/10

💬 Sur les 0 commentaires analysés, aucune tendance n'est disponible.