What happened
GitHub has introduced canvases, interactive visual surfaces within the Copilot app where developers and agents collaborate in real time. Users create a canvas by typing /create-canvas and describing what they need; the agent updates it as it works, and users can interact via clicks, edits, and other actions that feed back to the agent or process locally.
Why it matters
Many developer tasks—like triaging issues, visualizing codebases, or finding knowledge holders across tools—are harder in a pure conversation format. Canvases let you see and manipulate information directly, spotting patterns and taking action faster than cycling through prompts and responses. This shifts Copilot from a chat tool into a collaborative workspace.
What to watch
The article provides five concrete canvas examples already in use: an issue triage helper (swipe cards to ship or reject), an interactive codebase diagram (hover, drag, filter to explore architecture), a worktree session view (see and clean up active/stale work with one click), a prompt coach (review past interactions and get improvement tips), and a knowledge finder (search Slack, Teams, email, docs to locate people with context on a file or topic). Developers can try canvases now in the GitHub Copilot app.
Summaries like this, in your inbox every morning.
The introduction of canvases reflects a maturing understanding of how AI agents and developers work together. While conversational interfaces—where users iterate through prompts and responses—work well for many tasks, they become inefficient when the problem requires visual reasoning, simultaneous interaction with multiple elements, or real-time feedback. By embedding interactive, visual surfaces directly into the Copilot app, GitHub is extending the agent paradigm beyond chat into what is essentially a shared workspace. The agent remains a participant (updating the canvas, responding to interactions), but the interaction model shifts from sequential text exchange to direct manipulation—swipes, clicks, drags, filters—alongside agent updates. This design choice acknowledges that certain workflows (triage, architecture exploration, cleanup, decision-making) are fundamentally visual; presenting information as cards, diagrams, or lists, rather than prose, lets users spot patterns and connections faster and take action with fewer keystrokes.
For example, today's edition would include:
AI-summarized, only the topics you pick — one digest a day via Email, LINE, or Slack.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. Q&As are published on this page for other readers too.
Anthropic released Claude Opus 5.5 today and cut its price 20%, with input at $4 per million tokens and output…
Anthropic and OpenAI, which spent early September warning that model capabilities are outrunning the safeguard…
OpenAI released GPT-6 Sol and GPT-6 Luna on September 22, trained the same way as its top-tier GPT-6 Astra, an…
In September 2026, Amazon Web Services and Salesforce announced that Salesforce CRM data will be usable from A…
Anthropic announced Claude Opus 5.5 on September 22, the first model in its new Claude 5.5 series, and began o…
John Platt and his Google team built Empirical Research Assistance (ERA)
