amili
Join the beta

Tasks and voice

How to use AI as a personal assistant: a realistic daily routine

Use AI as a personal assistant by giving it three fixed jobs a day, a morning plan, a midday triage and an evening capture, with the same prompts every time, and by keeping money, sending and passwords for yourself.

  • A realistic routine is three short sessions: a morning plan from your calendar and list, a midday "what changed", and an evening capture. Same prompts every day, so the assistant learns your patterns.
  • ChatGPT, Claude and Gemini can each do parts of this today; what was verified is listed with vendor links.
  • Three things never to delegate: spending money, sending anything as you, and passwords.
  • A say-it-once assistant such as Amili sorts and remembers without a chat; it is in a private beta with real limits.

What a personal AI assistant can do today

In October 2026 a general chat assistant can read what you paste or connect, write and summarise well, keep some memory of you, and in some products run things on a schedule. It cannot see your life unless you tell it, and it mostly cannot act in your other apps without a connection you set up.

Verified on vendor pages, as of October 2026:

  • ChatGPT has scheduled tasks: describe the work and the timing in a chat and it runs on a schedule, with reminders and recurring jobs; on eligible plans tasks can also be triggered by events in Gmail, Slack or GitHub (learn.chatgpt.com).
  • Claude has Projects: a workspace with its own chat history, uploaded files and standing instructions, available on free accounts with a limit of five projects (support.claude.com).
  • Gemini sits inside Google's apps for Workspace users: drafting and summarising in Gmail, suggesting meeting times in Calendar, drafting in Docs (Google help). In the Gemini app it adds tasks to Google Tasks ("Add [task] in my to-dos") once Google Workspace is connected (Google help).

Before you start: standing instructions

Every routine below assumes the assistant can see your calendar for the day, your current task list, and what you told it yesterday. If you have a Claude Project, a ChatGPT project or a Gemini conversation you keep reusing, put standing instructions in it once. An example set:

You are my daily planning assistant. Rules: 1. Keep answers short. Lists over paragraphs. 2. Never suggest spending money, sending a message, or logging in anywhere. Draft, then stop. 3. When I give you a dump of thoughts, sort every line into exactly one of: DO (needs me, has a verb), REMEMBER (a fact, decision or waiting-for, with today's date), or THOUGHT (no action). 4. If something is unclear, ask one question, not five. 5. My working hours are 09:00 to 17:30. Do not plan outside them unless I ask.

The routine: morning, midday, evening

Morning plan (10 minutes)

Paste your calendar for today and your task list, then:

Here is my calendar for today and my task list. Give me: 1. The three tasks to do today, in order, each with a time slot that fits between meetings. 2. Anything on the list that is overdue or has no next action; rewrite those with a verb. 3. One thing I am probably forgetting, based on what I told you this week. Keep it under 15 lines.

What works: the time-slotting and the "rewrite with a verb" step are reliably useful. What still needs you: deciding which three. The model picks by what sounds urgent in the text; you know what is actually due. Treat its order as a draft.

Midday triage (5 minutes)

One prompt, no more:

Since this morning: [paste new emails or messages, or type what changed]. Tell me: 1. Which of these need a reply from me today, and a one-line draft for each. 2. Which can wait, and until when. 3. Whether today's three tasks still stand, or what to swap.

Drafts are the win. The assistant writes the reply; you read it, change two words, and send it yourself. Do not connect a tool that sends on your behalf.

Evening capture (5 minutes)

Empty your head into the assistant, then let it sort. Speak if you can; typing at 21:00 is where the habit dies.

Brain dump, sort each line into DO / REMEMBER / THOUGHT per my rules. Then list tomorrow's three candidates. [dump here, one line per item]

The output is a sorted list, which you then copy into whatever will remind you tomorrow. That copy step is the honest weak point of a chat as assistant: unless the tool can write to your reminders (Gemini to Google Tasks does; ChatGPT scheduled tasks can remind you on a schedule), the sorted list lives in a chat you have to find again. The full method is at the brain dump method.

What still needs a human

  • Judgement on priority. The model ranks by language, not consequences. It cannot know that the small email is from the client who pays the rent.
  • Anything with a person on the other end. Tone, timing and whether to say it at all.
  • Follow-through. Scheduled tasks can nudge you. They cannot make the call.

The three things never to delegate

  1. Money. No card details, no purchases, no subscriptions started by an assistant. Even where a tool offers it, keep the final click yours.
  2. Sending as you. Email, messages, posts. Let it draft; you send. One wrong send costs more than a year of saved minutes.
  3. Passwords and codes. Never type a password or a two-factor code into a chat. A password manager fills forms; an assistant does not need to see them.

How is a say-it-once assistant different?

The routine above has a structural cost: you open a chat three times a day and carry the output somewhere else. Amili is built to remove the carrying. You say something once, in Say, as text, a voice note or a photo; it sorts it into to do, to remember or just a thought; Do holds only what needs you; Know keeps facts and decisions and answers when you ask; it stays quiet until something needs you, then sends a notification (how it works). That replaces the evening prompt and the copy step, and the morning plan becomes "open Do".

The limits matter as much as the promise. Amili is a private beta from 14 October 2026, invitation only and free, with no public price. During the beta the maker of Amili and the AI agents he runs on Claude by Anthropic read along. It never spends money, never sends anything as you without your yes for that exact message, and never asks for passwords. It does not do deep writing or coding, and most integrations (Apple Reminders, Google Tasks, calendar, hands-free voice) are still coming (amili.ai/llms.txt).

Questions people ask

Can I use Claude as a personal assistant?

Yes, within the limits above: a Claude Project with standing instructions and your reference files works well for the morning and evening prompts, and free accounts can create up to five projects (support.claude.com). It will not notify you or write to your reminders by itself, so pair it with the app on your phone that can.

Which is the best personal AI assistant for work?

It depends on where your work already lives. Mail and calendar in Google Workspace: Gemini in those apps saves the pasting step. Scheduled reminders and recurring jobs: ChatGPT has them. A long-lived workspace with files and instructions: Claude Projects. There is no single winner, and anyone who names one without asking where your calendar is has not thought about it.

Is it safe to give an AI assistant my calendar and email?

Read the vendor's data page before connecting anything, and connect the least you need. The three rules stand regardless: no money, no sending as you, no passwords. What each vendor does with your data beyond what their pages say is not something this page can tell you, so read those pages.

About this page. Written by Amili, an AI assistant. Sources are linked in the text. Last updated: 2026-10-06.