AI is starting to fit how people actually work

I sent my wife and brother an invite to Instinct. Within 30 minutes, both were using it.

That surprised me more than anything the assistant had done for me.

I’m usually the person who tries new technology early. I’ll tinker with the setup and tolerate a few rough edges because I’m curious about what’s possible.

Swati and Kshitij are more pragmatic. They’re comfortable with technology, but they expect it to fit into their lives. They’re unlikely to spend an afternoon learning something just because I think it’s exciting.

Yet here they were, putting a brand-new AI assistant to work almost immediately.

Swati started using it to organize her calendar and coordinate with clients. We’ve been using our respective Instincts to coordinate travel. Kshitij started using it to send payment reminders to customers - essentially taking on the accounts receivable work he would otherwise have had to hire someone to do.

According to him, collections have improved without him lifting a finger.

And it has stuck. They’re texting it throughout the day and finding new things to ask it to do.

I introduced them to the tool. They didn’t need me to explain every use case.

That’s the key. It just worked.

A habit they already had

There’s a growing category of personal AI assistants - Instinct, Muse, Dot, Grok Bot and others. I’ve used Muse and Instinct primarily.

What interests me is who finds these tools useful, and how quickly.

For Swati and Kshitij, the appeal wasn’t the underlying model, its memory architecture or the fact that it was an AI agent. They could care less.

What mattered to them - They could send it a message and get actual work done.

WhatsApp is already part of how they organize their lives. Asking an assistant to handle something fit naturally into that habit. They could start with something they actually needed help with, using a familiar interface.

The value was apparent enough that Kshitij even gave it access to his credit card - something he would ordinarily be reluctant to do. And which I’m still very surprised by!

I’ve felt the value too

My travel plans are usually very fluid on my US trips, even more so this time around! Dates and destinations change, and finding somewhere to stay can turn into an exercise in comparing Marriott, Airbnb, Google Flights and several other tabs. I don’t have an EA or a chief of staff. Now I do!

Instinct has already saved me a few thousand dollars on hotel bookings by finding better, cheaper alternatives. It handled the process through to booking.

The time savings have been immense. I send it a message, let it work and come back to the result. When my plans change, I can continue the conversation.

The benefit has been immediate and tangible. And while the savings are easy to quantify, having less to coordinate has been valuable in its own right.

That experience helped me understand why these assistants feel different from the chatbots I’ve used before.

Why these assistants feel different from a typical chatbot

The easiest way to understand the difference is to look at what happens after you send a message.

With a typical chatbot, you ask a question and get an answer. You might ask it to draft an email, summarize a document, suggest a hotel or help you think through a problem. But you usually remain responsible for the next step: copying the answer, opening another application, checking the details, making the booking, sending the email or updating the calendar.

That can be extremely useful. But the assistant is mostly helping you think or create. You are still coordinating the work.

These newer personal assistants are designed to take on more of that coordination.

You can tell one what you want in ordinary language: find a better hotel, rearrange my schedule, plan this trip or follow up with this person. The assistant can remember relevant context, use connected applications, browse websites, compare options and sometimes complete the task. It may ask for approval before taking an important action, but it can handle much of the work in between.

The interaction shifts from “help me figure this out” to “please take care of this.”

The difference isn’t simply that one assistant is smarter than another. These systems are built to operate over time and across tools.

A conventional chatbot generally waits for your next prompt. A personal assistant can continue working after you leave the conversation. A chatbot may give you instructions for booking a hotel. An agent can search for options, compare them with your preferences, return with a recommendation and, if you approve, complete the booking.

The experience feels different because the assistant isn’t just producing text. It is carrying a task.

What makes that possible?

A few capabilities come together.

It can remember context. You don’t have to restate every preference or detail each time. If it knows how you like to travel, what matters to you in a hotel or which commitments are already on your calendar, a new request can build on what came before.

The conversation becomes an ongoing relationship rather than a series of disconnected questions.

It can work where the work happens. These assistants can connect to applications, files, calendars, browsers and other tools. Some have a computer or browser of their own, allowing them to navigate websites and complete steps rather than simply telling you what to do.

That turns an answer into an action. Instead of saying, “Here are three hotels you could book,” the assistant can search, compare, check availability and help complete the reservation.

It can keep working when you step away. You don’t have to watch every search or approve every minor decision. You can give it a task, close the app and return later to see what it found or respond when it needs your approval.

This is especially useful for tasks involving several steps. The assistant can track where it is, what it has done and what remains.

It can ask for help at the right moment. Delegation doesn’t mean giving the assistant unlimited authority. A useful assistant should know when it can proceed and when it needs you to decide, before spending money, sending a sensitive message or making a commitment on your behalf.

The goal isn’t for the assistant to act without you. It’s for you to stay in control without managing every detail.

Together, these capabilities reduce the work required to use the AI itself. You don’t have to translate your goal into a perfect prompt, move information between applications or keep reminding the assistant what happened earlier.

You can say what you want done, and the assistant can take responsibility for more of the path between the request and the result.

That’s why a simple message can feel like something much bigger than a chat.

What this makes me question about AI at work

I spend a lot of time thinking about AI adoption in organizations. This experience at home has made me question how much effort we expect employees to put in before they see a benefit.

We give people access to tools, explain their capabilities and encourage them to find use cases. There’s a place for that. But how much of the burden of making AI useful are we leaving with the employee?

A conversation with my boss recently surfaced a related distinction: people can know where to find an AI assistant without knowing what work they can hand over to it. How do you reconcile that?

We’ve taken one step toward this at work by making our AI coworkers available where people already work: in email, Teams, Word, Excel, PowerPoint and the browser - and now on mobile too.

That matters because adoption is easier when people don’t have to leave their existing workflow to find help. The assistant is available in the context where work is already happening, with access to the documents, conversations and tools relevant to the task.

But availability isn’t the same as seamlessness and intuition. People shouldn’t have to decide which assistant to open, what context to provide or how to translate their goal into the right prompt.

The next challenge is to make delegation feel as natural as asking a colleague for help. That means giving AI coworkers the right context, access and boundaries - and making the handoff intuitive enough that people can focus on what they want done rather than how to operate the technology.

So what next?

My family’s experience doesn’t establish a universal rule about adoption. It does give me a useful hypothesis:

People may be much more willing to adopt new technology when it fits an existing habit and delivers a useful result quickly.

Swati and Kshitij are still discovering what they can delegate. I’m curious to see how far that goes. Kshitij already is starting to think that his Instinct glitches on certain tasks.

For now, I’m paying closer attention to what happens after someone’s first useful result. Do they return? Do they find another task? Does the technology earn a place in their day?

Maybe that’s the real test for AI - not whether people are impressed by what it can do, but whether they begin to rely on it without being asked.

Also, very importantly - There should never be an excuse from devs or companies that says “That’s a user error” or “your prompt wasn’t good enough “ or “you need to improve your CLAUDE.md or add more skills”.

In my family’s case, the assistant earned its place the same way any useful habit does: it made the next day a little easier.