How Aida works, in plain English.

An AI brain of linked notes

In short: everything Aida knows is a note in her AI brain, and notes link to each other, like a well-kept notebook full of cross-references. It grows with every conversation.

Each note is about one thing: a person, a project, a preference, a decision, a skill. Notes link to the notes they mention, so "Spring newsletter" links to "Prefers short replies" and to the skill "Write the newsletter". When you ask for something, Aida looks for the notes that matter, both by the words you used and by what you meant, and reads them before she answers.

The notes are ordinary text files in a folder on your computer, grouped into hubs such as People, Projects and Decisions. You can read them in any notes app. Nothing is locked away in a format only Aida understands.

Every note keeps its history. Aida saves the state of the brain regularly, so you can compare any note with an earlier version and put it back.

A team with roles and skills

In short: each team member has a clear job and a set of skills, and that's what they're chosen for.

A team member is a role: what they're responsible for, how they work, and what good work looks like. Each arrives with skills, which are written, step-by-step ways of doing the kinds of task that role does, such as "compare supplier quotes" or "check a web page on a phone".

Some roles plan and check rather than do. The Project Manager breaks a big goal into pieces and sees it through, while the specialists do the work. Managers are never handed the hands-on work themselves.

You can teach anyone a new skill, change a role in plain words, or ask the Hiring Manager to design a new team member. Nobody joins the team until you approve.

The right person for each job

In short: Aida works out what a job really is, then picks the person whose role and skills fit it best.

When work comes in, Aida weighs up the whole team: what each person is for, and which of their skills apply. She judges by what the job asks to be done, not by what it happens to mention. "Research the best newsletter tool" goes to the Research Assistant, even though it mentions newsletters.

The chosen person gets the job with the skills that apply already open in front of them. If nobody on the team fits, Aida brings in a general helper for that one job rather than forcing it on the wrong person.

Side by side, and back to you

In short: work that doesn't depend on other work runs at the same time, and everything comes back to the conversation it started in.

Aida splits a request into pieces and works out which depend on which. The newsletter and the supplier research don't depend on each other, so they start together. A header image for the newsletter waits until there's a draft to fit.

Each team member can carry a few jobs at once, and jobs queue politely when everyone is busy. Changes to the same piece of software are made one at a time, so they never trip over each other.

Whatever comes back (a result, a question, a request for permission) arrives in the conversation where you asked. You never have to go looking for it.

The permission gate

In short: every action outside her own notes passes through one gate, and the gate asks you.

Aida and her team can read and update the brain freely, because it's their notebook, and every change there is shown to you and can be undone. Everything else goes through the gate: looking something up on the web, reading or changing a file on your computer, running a program.

At the gate, some things are simply refused, whoever asks. Things you've already allowed go through. Anything else stops and asks you, in the conversation, in "Needs you", or as a notification on your phone: allow once, always allow, no, or cancel the whole job.

"Always allow" is deliberately narrow. Aida suggests a tight version and a slightly wider one, and you choose. Every standing permission is listed where you can see it and take it back. You can also trust a whole folder, such as your projects folder, while things like publishing or sharing work still ask every time.

A record of everything

In short: everything that happens is written to a day-by-day record that Aida can't change.

Every request, every decision at the gate, every note read or changed, and every job started, finished or stopped goes into the record. Aida never edits it, and her team can't write to it.

The Observatory shows the same record live and lets you look back at any day. Click any team member to see the story of what they did, step by step.

Stop really does mean stop. When you stop someone, their work ends at once, and a late report can't quietly undo it.

Her know-how, your brain

In short: how Aida works is built in; what she knows lives in your brain. Swap the brain and she behaves just the same.

Aida is two things. The first is her know-how: how she plans, matches work to people, runs things side by side, asks permission and keeps the record. That's the same for everyone. The second is the brain: who you are, how you like things, your team, your projects and what she has learned. That's yours.

Keeping the two apart means improvements to Aida reach every brain, and your brain never fills up with instructions about how Aida works. It also means you could give Aida a different brain, for a second business say, and she'd behave in exactly the same careful way for different people.

A new brain starts almost empty: Aida herself, a default personality and a short profile of you to fill in. You add the team from the catalogue when you're ready.

Learning over time

In short: she learns from what you tell her, from your corrections and from what you teach her, and checks with you before keeping it.

When you mention a fact or a preference, Aida writes it down with where she learned it. New facts she has saved herself wait in "Needs you" for a quick keep, edit or forget, the least certain first.

You can teach her a procedure: switch on training, show her the steps once, and she writes it up as a skill for you to approve. When she notices you asking for the same thing again and again, she offers to turn it into a skill.

Once a week she does a little gardening, suggesting which duplicates to merge, contradictions to fix and stale notes to tidy, and she changes nothing until you say which. The AI Data Trainer can also look for gaps in what she knows and ask you about them, one question at a time.

Routines and triggers

In short: Aida can work on a schedule, or when something changes, without being asked each time.

Ask in plain words: "every weekday at eight, give me the day's plan", or "tell me when this page changes". Besides set times, a routine can start when a web page changes, when a file lands in a folder, or when another service calls a private web address you give it. Whatever arrives is treated as information to look at, never as instructions to follow.

Routines follow the same rules as everything else. If one needs permission, the request waits for you, unless you've allowed that routine to search and read the web on its own. If Aida was switched off when something was due, she catches up when she starts.

Where she runs

In short: on your own computer, or on a private server you rent, with your phone as a window onto her.

Aida runs on a computer that stays on while she works, or on a small private server. You use her in your web browser. On a server she sits behind a secure address, and each device signs in once with your password.

Install her on your phone like an app. She lets you know when she needs you, when a routine finishes and when something goes wrong, and you can answer permission requests straight from the notification.

To think and write, she uses an AI service you sign in to with your own account, so the words of a conversation are processed there. Your brain, your files and the record stay with you.