Your app gets 3 days. A habit takes 66.

The average phone has 80 apps installed. The average person opens 9 a day.

Look at which 9. Social, messaging, entertainment. About half of all phone time goes to those categories. Even ChatGPT and Claude fit the pattern: they are chat apps.

Now look at what happens to a new app. 77% of users abandon it within 3 days. By day 30, around 4 or 5 in 100 are left. Meanwhile the research on habit formation says a new daily habit takes 66 days on average to become automatic. The range in that study was 18 to 254 days.

Put those two numbers next to each other. A habit takes 66 days. Your app gets 3.

That is the wall every new app runs into. It is not a design problem or a feature problem. It is a habit problem. People will not change their ways unless the new thing is 10x better. Your app needs ChatGPT-level differentiation to earn a new habit. A normal app will not do.

Stop fighting the 9 apps. Join them.

The fix is not a better onboarding flow. The fix is to stop asking for a new habit at all.

Combine with the apps your users already open. This is a general trend in the app space right now, powered by AI, and it splits cleanly by market. B2B products connect to Claude and ChatGPT through MCPs. B2C products connect to the messaging apps, where WhatsApp alone has 3 billion people.

Most people hear "AI in a chat" and picture a support bot answering FAQs. That picture is about three years out of date. Here is what an agent living in a chat interface can actually do today.

What can an AI agent in a chat actually do?

An AI agent in a chat interface can answer from one business's exact knowledge, start conversations on its own, run scheduled routines, time offers to user behavior, take payments and bookings, hand off to a human, and through an MCP, operate the business itself. Here is each one in practice.

It can hold your entire knowledge, and answer like you. Not generic answers. An agent can be loaded with one business's exact catalog, prices, schedule, and method, and answer only from that. I recently built one for a hostel group with three properties. Each property's agent knows its own rooms and its own prices, and never quotes another property's. A coach's agent answers questions the way that coach would, trained on their method, in their voice.

It can start conversations, not just answer them. This is the shift most people miss. A chatbot waits. An agent acts. It sends the morning welcome to today's check-ins. It messages each member of a program with a daily check-in tied to their own goal. It follows up with every lead that went quiet. The conversation arrives in an app the user was going to open anyway.

It can run routines on a schedule. Morning messages, checkout-day review requests, event announcements, weekly summaries. Things a person used to copy between a spreadsheet and a chat window every day, running on their own, every day, without drift.

It can watch behavior and time its moves. An agent sees who is engaging and who is drifting. When a member goes quiet, it reaches out before the silence becomes churn. When a subscription lapses, it makes a win-back offer at the moment it matters. When engagement shows someone is ready for more, it offers the next step in the ladder. Timing is the whole game in retention, and timing is exactly what software does better than memory.

It can transact. Send the payment link, confirm the booking, deliver the digital product, upgrade the plan. The chat is not a brochure for the business. It is a working surface of it.

It can hand off to a human, and know when to. A well-built agent knows its edges. Price negotiation, a complaint, a sensitive question: it hands the thread to the owner instead of improvising. Automation with human control is not a compromise. It is the design.

And on the B2B side, through an MCP, it can operate the business itself. This is the part I would not have believed two years ago. My own company runs through a chat interface. The agent I work with can build a branded app, write and schedule a week of content, run outreach, read analytics, manage bookings, and send campaigns, because every one of those capabilities is exposed to it as a tool. My users get the same thing: they connect their business to Claude, and their app becomes something they talk to.

None of this requires the user to learn anything. That is the entire point. Every capability above arrives inside an interface the user already opens 9 times a day.

What we just shipped at Tribed

We build branded apps for coaches and experts, so the 3-day wall is our wall too. Last week we shipped our answer for the B2C side.

Every Tribed app now comes with an engagement agent on WhatsApp. Members connect it during onboarding. From then on it runs the daily check-ins tied to each member's goals, the win-back offers when someone drifts, and the upsells when someone is ready. Every message deep-links back into the app, so each touch educates the member into the app one small step at a time. The member sees the value over weeks, not in a 3-day trial window. The app stops betting everything on the fragile first days and starts compounding.

And because this is a coaching product, consent is built in: members opt in during onboarding and can stop the messages with one word.

The bet

If you are building in B2B and your product has no MCP, I think the slowness of your adaptation will kill you inside a year.

If you are building in B2C and you are not inside your users' messaging apps, you are asking users for 66 days of habit-building they will never give you.

Stack on the habits your users already have. Do not compete with them.

FAQ

How long does it take users to build a habit around a new app?

Research from University College London found a new daily habit takes 66 days on average to become automatic, with a range of 18 to 254 days. Most apps never get that long: 77% of users abandon a new app within 3 days of installing it.

What is an MCP and why does it matter for B2B software?

MCP (Model Context Protocol) is an open standard that lets AI assistants like Claude and ChatGPT connect to your product and use it as a set of tools. Instead of learning your interface, your users work with your product from inside the AI chat they already use daily.

Can an AI agent message app users on WhatsApp?

Yes. With the user's opt-in, an agent can send daily check-ins, win-back offers, and upsell messages over WhatsApp, each deep-linking back into the app. Consent matters: members should opt in explicitly and be able to stop all messages with one word.

Tribed builds branded apps for coaches and experts: your program, your community, and an AI coach trained on your method, in an app your clients keep. Book a 15-minute call.