How to Build a Personal AI Assistant in an Ordinary Chat With No Code

The word "agent" is often imagined as a separate program that starts itself on a schedule and reports the result on its own. In practice, for one permanent role a much simpler thing is enough: an ordinary chat in ChatGPT, Claude, or Gemini that has been told once who it is, who it works for, and what specific task it solves each time.

What actually happens inside

Such an "assistant" doesn't start itself on a schedule and doesn't work in the background. It's a chat with a pinned instruction: a name, a role, a response format, and context about you. Every time you open this chat and send new input, the model applies the pinned role to the new task instead of inventing an answer from scratch.

The difference from an ordinary one-off prompt is that you don't have to write the instruction again. You describe once what the helper should do, and then you just send fresh data: a to-do list for the day, a client's question, an idea for a post.

Where this is set up with no code

In ChatGPT this feature is called Custom Instructions (profile settings), or Projects if you need to keep several context files next to the instruction. In Claude, Projects play the same role: a pinned instruction plus knowledge base files that the chat keeps at hand for every request. In Gemini a similar feature works through Gems, although Google has already announced it will rename it to Skills from November 2026.

In all three cases the logic is the same: the instruction lives in the chat or project settings once, instead of being pasted by hand at the start of every conversation.

A template for a working instruction

To build such a helper, the instruction needs to cover four points:

1. 1. Name and role. Who the helper works as and for what specific task, not "help me with everything".
2. 2. Input. What you will send each time: a to-do list, the text of an email, a brief for a post.
3. 3. Response format. What the output should look like: three points, a table, a finished text, a question at the end.
4. 4. Context about you. The tone you usually use with clients, or the goal you're moving toward this week.

An example of a minimal instruction for sorting a to-do list in the morning: "You help me sort my task list for the day. Every morning I send a to-do list. Pick three tasks that really move me toward my goal this week, and one that can be dropped today without consequences. At the end, ask one clarifying question so tomorrow's review is more accurate."

Which tasks such a helper really covers

The helper sorts the to-do list by priority every morning. It lays out one idea into several posts and formats for a week ahead. It writes a draft reply to a client in a recognizable tone, which you only have to check and send.

In my experience, setting up such a helper takes 10 to 15 minutes, and for simple roles the free version of the chat is enough. The most common mistake at the start is pinning an instruction that's too general, like "be a helpful assistant". The more precisely one specific task is described, the more stable the result from run to run.

Where the limits of this approach are

This is not an autonomous agent in the full sense. The helper won't open the chat itself and won't check email on a schedule in the background while you're busy with something else. To get that kind of automation you need the next layer: connecting the model through an API to a scenario in n8n, Zapier, or a similar tool, where a request goes to the model on a timer or an event with no human involved.

For personal routine and repeating creative tasks, a pinned instruction in an ordinary chat covers most of the need without a single line of code. And when the role grows into several linked steps and you need it to run without your involvement, it makes sense to move to the next level: real automation on top of the model.

I write about how I build such setups and move from a manual chat to full automation in the channel «Растём с ИИ» ("Growing with AI"). If you need to build a system for your specific task, you can start with a free consultation.