Skip to main content
All posts

The Agent Proposes — You Confirm: Safer Task Creation

The AI agent in AI Planner can create tasks from chat. Since 0.11, when parsing topics and drafts it does not write to the database silently: it shows a proposal first — you confirm or edit — then the card lands on the matrix.

When confirmation is required

  • The agent parsed a long text or list and proposes several tasks
  • The wording was ambiguous — the model drafted a best guess
  • Capture from a screenshot or thread with many entities in one reply

An explicit command like “create a task: …” can still apply immediately — you already stated intent clearly.

What it looks like in the UI

  1. You ask the agent to parse chaos or a topic.
  2. Chat shows proposed fields: title, quadrant, due date.
  3. Confirm to create as shown or dismiss.
  4. After confirm, the task appears on the matrix and Today.

This is not checkbox safety theater: fewer junk cards and accidental due dates when the model guesses context.

Not an autonomous agent

AI Planner does not run background bots editing your projects. The agent is a planner chat tool. Approve-before-create reinforces that: even active parsing stays under your control. Agent overview: companion article.

Same in the Telegram Mini App

The same agent runs in the Telegram Mini App. Confirmation works the same — handy on mobile before creating a card.