Draft a reply to a blocked agent

An agent that hits a question stops and waits for you. If you have been away for a few days, several may be waiting at once, and the cost is rarely the thinking. It is starting the reply.

Draft a reply sits on the comment box in Onplana Tasks, on any task where an agent is involved. It asks Onplana to write a first version of your answer to the waiting agent and puts that text in the box.

What it does, and what it does not

1. Open the task in Onplana Tasks and go to its comments.
2. Select Draft a reply.
3. The draft appears in the comment box, already addressed to the agent.
4. Edit it into what you actually mean.
5. Press Comment to send it.

It never sends anything. The draft lands in the box and stops there. Nothing reaches the agent until you press Comment yourself, so a draft that misreads the situation costs you a glance, not a wrong instruction acted on.

It adds to what you typed, it does not replace it. If you had already started writing, the draft is appended below your text rather than overwriting it.

It arrives pre-addressed. Agents are reached by mention, so an unaddressed reply reaches nobody. The draft opens with @ and the agent’s name.

Note

If the agent’s name is one Onplana cannot reliably turn into a mention (unusual characters, punctuation, or a very long name), the draft deliberately arrives with no mention rather than a broken one that would silently fail to deliver. Address it yourself in that case.

The limits

Drafting calls a model, so it is capped in two directions:

Limit	Amount
Per person, per day	40 drafts
Per task, per hour	5 drafts

The per-person window is daily rather than hourly on purpose. The situation this feature exists for is the catch-up burst: you come back from time off, find several agents blocked, and work through all of them in one sitting. An hourly cap would throttle precisely the case it is meant to serve.

The per-task hourly cap is the opposite concern. Re-drafting the same reply five times in an hour usually means the draft is not the problem, and a sixth will not help.

You are warned before you hit the wall

Once you have 5 or fewer drafts left for the day, each draft tells you what is left, for example “3 AI drafts left today.” On the last one it says “That was your last AI draft for today.”

If you do reach a limit, the message says which limit, what it is, and when it lifts, for example “You have used all 5 drafts an hour for this task. More in 42m.” You are never left guessing why the button stopped working.

It also spends AI tokens

Drafting draws on your organization’s AI token allowance like every other AI feature. If that balance is exhausted, drafting stops with the same message the rest of the AI suite gives you, along with the ways to fix it: buy AI credit, or move to a plan with a larger allowance. See Understand your AI token budget.

So there are three separate things that can stop a draft: your daily count, the task’s hourly count, and the organization’s token balance. They are independent, and the message always tells you which one you met.

When it helps and when it does not

It is worth using when the agent has asked something you can answer in a sentence or two and you are working through a queue of them. It saves the blank box, not the decision.

It is not worth using when the answer requires a judgement only you hold: a priority call, a budget, a decision about scope. Draft it if you like, but expect to rewrite it. The model can see the task and the conversation. It cannot see the meeting you had yesterday.

FAQs

Does the agent see the draft before I send it? No. Nothing leaves the comment box until you press Comment.

Can I edit the draft? That is the intended flow. The draft is a starting point, not an answer.

Why did my draft arrive with no @mention? Because the agent’s display name could not be safely turned into a working mention. Add the mention yourself so the reply is delivered.

Do drafts count against my AI tokens even if I delete them? Yes. The model runs when you request the draft, so the tokens are spent at that point whether or not you send the result.

Is this on my plan? Yes, on every plan including Free, bounded by your AI token allowance like the rest of the AI features.

Related

• Talk to agents in task threads covers the conversation this drafts into
• Review agent output for approving what an agent produces
• Understand your AI token budget for the allowance drafting draws on
• Install the Onplana Tasks browser extension if you do not have the companion yet