# How human help works, and why you approve every task

> Some steps need a person in the real world. Here is how we plan for a human at Deputant to help your agent, and why nothing happens until you say yes.

Source: https://deputant.com/blog-how-human-help-works.html

Deputant team · Oct 8, 2026 · 3 min read

AI agents are good at many tasks on a screen. They can compare prices, fill in a form, and book a table. But some steps happen in the real world. Someone has to call the store. Someone has to drop off the return. Someone has to wait on hold.

An agent can't do those steps well yet. So we are building human help.

## What human help is

When your agent gets stuck, it can hand the task to a real human at Deputant. The human does the step your agent can't do. Then the human hands the task back to your agent, or back to you.

Here are some tasks we expect human help to cover:

- Calling a store to ask about an order
- Dropping off a return
- Calling a restaurant that doesn't take bookings online
- Talking to a support line that only answers by phone

These are small tasks. They are also the tasks that stop an agent cold. One phone call can be the only thing between you and a finished task.

## You approve every task

A human never starts a task on their own. Your agent asks first, and you decide.

When your agent wants human help, you will get a request. It will tell you three things:

- What the task is
- What the human will see from your vault
- What the human will do

You say yes or no. If you say no, nothing happens. The task comes back to you, and you can finish it yourself or try another way.

Why ask every time? Because a person acting for you is a big step. It should never be a surprise. You should always know when a human is working on your behalf, and what they can see.

## You set the rules

You also choose which kinds of tasks humans may do for you at all.

Maybe phone calls are fine, but nothing that uses your card. Maybe returns are fine, but only for one store. Maybe you don't want human help at all. That is a fine choice too, and it will be easy to turn off.

Your rules come first. If a task is outside your rules, it won't reach a human, even if your agent asks. If a task is inside your rules, you still approve it before it starts.

## Humans see only what a task needs

A human at Deputant won't get your whole vault. They will see only what the task needs, and only for that task.

Say the task is a call about a late order. The human may see the store name, the order number, and your name. They won't see your passwords. They won't see your other cards. They won't see your memories, unless one matters for the task and you approved it.

When the task is done, that access ends.

We think this is the right default for any helper, human or agent. It is the same idea behind the vault: each one gets only what it needs.

## We want agents to need less human help

Human help is a bridge. It fills a gap that agents have today.

Each time a human steps in, it shows us a step that agents can't do yet. We want to build tools so agents can do more of those steps without a person. Over time, we want your agent to need less human help, not more.

Some steps may always need a person. An agent can't carry a box to the post office. For those steps, we want human help to be simple, safe, and clear.

## What to expect in early access

We're still building human help, and it will be limited while we roll out access.

When early access opens, human help will be limited. Real people do this work, and we will start small. At first, human help may cover only some kinds of tasks, or only some places. We will tell you what it can and can't do before you count on it.

If that sounds useful, join the waitlist. We will email you when early access opens.

## Supercharge your digital life

We're rolling out access in waves. Join the waitlist, and we'll email you when your spot opens.

[Join the waitlist](https://deputant.com/#join)

[All posts](https://deputant.com/blog.md)