Onboarding your AI assistant: Start with one task

Perhaps there is a task you put off every week. Nothing huge. But it stays on your mind until you finally do it. That is a useful place to start when you are onboarding a personal AI assistant.

Hand over a manageable task, look at the result together and explain what should be different next time. This helps you find a way of working that suits you. You do not need perfect instructions or connections to your entire digital life to begin.

Even good assistance needs a beginning

An experienced human assistant arrives with skills. They still need to get to know your priorities, your tone and the small things you care about. At assistant provider Athena, working preferences and shared expectations are part of onboarding. Athena describes that process here.

There is a practical idea we can take from this for AI assistance: start with real work and use it to discuss what matters to you. You provide the personal context. Your assistant should help turn it into a clear task.

This takes some attention at first. It does not mean accepting poor results indefinitely. If you keep correcting the same mistakes, you need a better way of working, or the task may not yet suit the available tool.

Choose a task whose result you can judge

A useful first task comes up regularly and produces something you can assess. You might organise notes for a current project, prepare research or revise a text. There are more possibilities in our guide to everyday AI assistance.

Take this fictional example: you are planning a small workshop. Ideas, unanswered questions and decisions already made are mixed together in your notes. Next time you return to the project, you want to see where to pick things up.

A first request could look like this:

I want to keep track of my workshop project. Sort these notes into: already decided, still open and the next useful step. Keep it short. Flag anything unclear. Do not change any dates or commitments.

This establishes the material, the result you want and the limits of the task. If an important detail is missing, the assistant can ask a specific question. This first task does not require access to your whole inbox or calendar.

Use the result to show what you mean

Perhaps the first overview is too long. “Make it better” leaves the problem unclear. You could be more specific: “Show me the next decision first. I only need the background when I ask for it.”

In her work on leadership and collaboration, Ann Hiatt explains why general feedback offers little direction without a clear problem and an observable goal. Her article on goal setting also offers a useful perspective on these smaller corrections.

You do not need to rewrite the whole result yourself. Say what already works and name the one change that would make it more useful. In the workshop example, this could become an agreed project preference: a short overview, the next decision first and open questions kept separate. You should be able to recognise that preference in the next result.

Agree on what should carry forward

“Keep it especially short today” differs from “For this project, I always want a short overview”. A good working relationship keeps track of that distinction. In her professional writing, Bonnie Low-Kramen stresses the importance of explicitly discussing expectations between assistants and the people they support. Great Expectations

With an AI assistant, a simple question can help: “What have you remembered about this project?” Ask for incorrect or outdated notes to be corrected. This lets you check what the next task will build on.

When we talk about Bhalithan learning, we mean using your conversation context, files and agreed notes. It is not a promise that he remembers everything accurately or that every conversation retrains the underlying AI model.

Delegate more as the work proves useful

Once the project overview works for you, you might ask for a draft of the workshop website. The assistant already has some context for the project. You still agree on what this new task should produce and which decisions remain yours.

Delegating a task also involves deciding which authority comes with it. The concept paper Intelligent AI Delegation addresses this explicitly. In everyday terms: knowing your working style does not replace approval to send or publish something.

With Bhalithan, for example, you review the specific email draft before it is sent. A repeated task only becomes an automatic routine when it is explicitly agreed and technically supported. Asking him to organise your notes once does not start a daily report.

Personal assistance can also mean fewer interruptions

Good assistance should take your attention into account. Reflecting on her work with Simon Sinek, Monique Helstrom describes how much the timing of questions affects a working relationship. Her personal account of communication gives a useful example.

Discuss whether you want a short overview, which decisions call for a question and when you need quiet. Be equally deliberate with personal information: start by sharing what the task needs. A selected passage may be enough. Connecting an account later should serve a specific purpose; passwords do not belong in the Telegram chat.

How to tell whether it helps

After a few similar tasks, look at the total effort. How much did you have to explain, check and correct? Is the result usable? Does the assistant follow your agreements? Is less work left for you?

A fixed number of days tells you little about this. A small writing task and an ongoing project require different amounts of context. Use the results to decide what else you want to delegate.

How getting started with Bhalithan is meant to work

Bhalithan is a personal AI assistant in Telegram with a workspace of his own. Start with something you need help with, in a text or voice message. Your files and agreed project notes provide a basis for your continuing work together.

Customer onboarding is still being tested. The pilot supports tasks through text and voice, along with connected mailboxes that require approval before sending. Before you start, we confirm which connections and services are ready for your task. Where a suitable connection is available, Bhalithan prepares the setup in his own browser; you handle any required sign-in through your private account area. He can then continue the setup as far as the service allows.

If a task has come to mind that you would like to hand over, talk to us about getting started.