What Makes an AI Assistant Personal?
A useful answer starts with your question. A more helpful answer also considers your life.
“What should I make for dinner?”
It’s a simple question. But the answer changes depending on who’s asking.
Maybe you love cooking and have an hour to spare. Maybe you just got home, there are three things in the fridge, and washing another pan feels like too much.
Both people asked the same question. They need different answers.
That difference is at the heart of CAM, an everyday AI companion we’re building around the person asking.
The answer should fit your situation
An AI assistant can offer plenty of dinner ideas. The more useful task is helping you choose something that works tonight.
Say you tell an assistant:
“I have rice, eggs, and peas. I’m tired, and I want something easy.”
An answer that fits might be:
“Egg fried rice could work. You’ve got the main ingredients, and we can keep it to one pan.”
The value comes from using the details you shared in that interaction to narrow the options and make the next step easier.
Personal context can be just as useful outside the kitchen. A weekend suggestion can change when you share that you prefer quiet places. An explanation can change when you say you’re unfamiliar with the topic. A project plan can change when you mention you only have twenty minutes.
Good personalization gives those details a purpose.
You shouldn’t have to explain everything again
Many everyday conversations depend on something said earlier.
“What was that restaurant Ashley recommended?”
“Which gift idea did we settle on?”
“What was the first thing I wanted to tackle in the garage?”
We remember fragments: a person, a reason, a rough date. We don’t always remember the title of the conversation where we discussed them.
Yet finding an old answer can mean scrolling through a long list of chats, opening several threads, and trying different search terms.
The experience we want to build with CAM is more natural: ask about something you previously shared, using the details you remember.
We’re designing CAM’s personalized memory to support that across conversations. An assistant needs to distinguish a passing comment from a lasting preference, recognize when information may be outdated, and make it easy for you to correct or remove what it remembers.
If it cannot find the answer, it should say so.
Personal should still mean under your control
Remembering more isn’t automatically helpful.
You might want an assistant to remember that you prefer simple explanations. You might not want it to retain a sensitive conversation. And a preference you shared six months ago may no longer apply.
Our approach to CAM is guided by a few design goals:
- Use relevant context when it improves the answer.
- Make remembered information understandable and manageable.
- Let people correct or remove information.
- Ask when an important detail is missing.
- Be clear about uncertainty.
An assistant should also understand where information came from. Remembering a restaurant recommendation you shared is different from having access to someone else’s messages.
Being personal should never require pretending to know more than it does.
Everyday usefulness is the goal
There’s no shortage of impressive things AI can generate. But an ordinary day is full of smaller moments where a little help matters.
Choosing dinner. Understanding an unfamiliar term. Finding a recommendation. Turning an overwhelming task into a manageable first step.
Our ambition for CAM is to make those moments easier through conversation and relevant context.
CAM is currently in development. The examples here illustrate the experience we’re working toward, rather than features available to use today.
We’ll share what we’re building and learning along the way.
Life has questions. Just ask Cam.