TaskAuraTaskAura
PLLog in

AI task prioritization — first pass, your call

Classifying tasks is the boring part of any system — it is where most people drop GTD. In TaskAura a language model suggests an Eisenhower quadrant, tags and context for every new inbox item. Suggests, not decides. You click or correct — the system learns your preferences within your account.

How it works

  1. 1
    Drop a task into the inbox
    Title and optionally context. Everything else the AI tries to fill in.
  2. 2
    A suggestion within seconds
    Matrix quadrant, suggested tags, optionally context (home, work). Each suggestion has a tooltip explaining why.
  3. 3
    Accept or correct
    Click = accept. Correction = the app saves your preference for similar tasks.
  4. 4
    Weekly accuracy
    Stats show accuracy — the share of suggestions you accepted without changes. If it falls, signal that life shifted or the model is losing context.

Why it works

Short-text classification is a well-studied task for language models — Brown et al. (2020) showed GPT-3 zero-shot performance on intent classification far above baseline. Eisenhower-style categorisation has the same shape: four classes, lots of context inside the task description itself.

The second effect is decision fatigue (Vohs et al., 2008). Every decision during the day burns resources — the more trivial decisions AI takes off your plate, the more is left for hard ones. Classifying a task is not a decision worth your attention.

Sources
When this method does NOT work
AI has a recency bias — freshly added tasks get classified as more urgent than they are. It is also biased toward long titles (more text = more signal). It cannot read nuance like “this is important, but only for me” — personal context stays with you. Treat suggestions as a first draft to correct, not as truth.

In TaskAura

Suggestions appear in the inbox and on quick capture. You can disable them entirely in settings or leave them only for selected fields (e.g. quadrant only, no tags). All suggestions are logged locally — they do not go back to the model as training data.
Try it free

Frequently asked questions

What if the AI keeps classifying wrong?
Every correction is saved. If despite corrections it does not improve, disable suggestions for that field — the system is optional, not required.
Do my tasks go to training?
No. The Lovable AI Gateway does not send data to train external models.
Are there rules without AI?
Yes — you can define auto-tags by keywords. Rules run locally, independently of AI.
How much does it cost?
Free for now. We are planning a premium tier for power users, but the model will stay optional.
Related