Problem

Right now, workspaces hosting tasks are named task-${RANDOM_WORKSPACE_NAME}.

This does not make it easy to identify the task later at a glance.

Proposed solutions

1. Use coder/agentapi to dynamically rename the workspace based on the initial prompt. This will depend on the AI agent inside the workspace being available.

1a) As above, but allow using a remote lanaguage model to name the task. This is a common practice in existing AI tooling.

2. Use an off-the-shelf text summarization algorithm (e.g. https://github.com/DavidBelicza/TextRank) to condense the task prompt.