Anthropic

Verified interview challenge

Design context and memory for a document copilot

Improve a document copilot that becomes slower and less reliable as conversations grow.

Unlock with Full Access

The challenge

An engineering team uses a document copilot to review design docs and RFCs over long working sessions. People ask follow-up questions, refer back to earlier decisions and move between related documents. As the conversation grows, the copilot takes longer to respond, costs more to run and sometimes contradicts what it said earlier.

You'll map the copilot's flow on the canvas and explain the changes you would make.

Improve an existing product
Improve the copilot without replacing its model or rebuilding the application.

What you'll practise

  • Context management

    Explain which information a turn depends on and how you would keep requests within the application's performance limits.

  • Memory and history

    Decide how the copilot should retain earlier decisions and use conversation history when a user returns to a topic.

  • Reliable responses

    Account for timeouts, contradictions and changes in response format, then explain how you would assess the behaviour of your design.

Your design workspacePreview
Your canvas
DocumentsQuestions & history
Copilot responsesContinuity & consistency
The interview, drawing canvas and final report are included with Full Access.