Canvas is a new way to write and code with ChatGPT

Original Article ↗ Hacker News Discussion ↗

What Canvas Is and How It Works

  • New ChatGPT mode with a split view: chat on one side, an editable “canvas” for code or prose on the other.
  • Supports inline edits, selection-based prompting, and targeted rewrites without regenerating entire artifacts.
  • Early users report it can run and display some Python outputs, acting like a lightweight IDE in the browser.
  • Current beta appears mostly single-file; no full-project / directory context yet.

Comparison to Claude, Cursor, Copilot & Others

  • Widely seen as a response to Claude’s “Artifacts” and Cursor-style AI editors.
  • Differences vs Claude: Artifacts regenerate whole outputs and are not inline-editable; Canvas emphasizes persistent, directly editable documents.
  • Compared to Cursor/Copilot: Canvas is less integrated with local codebases and IDE workflows, but has a slicker in-browser UX.
  • Some think Canvas is “chasing” competitors; others note VS Code + Copilot + extensions already cover much of this.

Use Cases, Strengths, and Limits for Coding

  • Praised for: single-file scripts, small web apps, refactoring snippets, explaining code, generating tests, one-off data/scripts.
  • Criticisms: lack of multi-file/repo awareness, difficulty handling headers/interfaces in other files, and unsuitability for large, complex systems.
  • Some report strong help in porting or restructuring code; others say models still hallucinate APIs, versions, and non-existent functions.

Writing, Content, and “Slop”

  • Example blog-post demos are derided as banal, generic “LLM slop” that bloats the web with low-value content.
  • Non‑native speakers and people weak at writing find such tools invaluable for clarity, tone, and professionalism.
  • Tension between authenticity/voice vs efficiency and polish is a recurring theme.

OpenAI Strategy, Business, and Ecosystem Impact

  • Debate over why OpenAI is investing in consumer UX vs staying an API/infrastructure provider.
  • Some argue ChatGPT/Canvas builds a sticky consumer moat and better training data; others worry OpenAI is undercutting third‑party tools built on its API.
  • Comparisons drawn to browser wars and Amazon cloning successful products on its own platform.
  • Concern that foundation models are plateauing, and these features are incremental “demo-ware” to sustain hype.

Developer Workflow, Skills, and Future of Work

  • Many experienced devs use LLMs as “junior developers” they review, reporting real productivity gains.
  • Others see a risk: fewer meaningful tasks for juniors, harder skill development, and fragile AI-generated codebases.
  • Broader anxiety about long-term displacement of routine programming, with maintenance and deep reasoning still seen as human strengths—for now.