Most of my work with AI is for businesses: custom agents, team training, public sector projects. Building Nino was different. Nino is a personal AI assistant on Telegram that you can have live in about two minutes. No company procurement, no IT department, no project manager. Just one person deciding, in a few seconds, whether this is worth their time.
That shift from business to consumer changes almost everything. Here are the principles that guided us, and the lessons I would share with anyone building a consumer AI product today.
Why a personal assistant, and why Telegram
Nino is built on OpenClaw, an open-source platform for personal AI agents that live in messaging apps, with memory, tools and scheduled tasks. In our business projects through Claws, we kept seeing the same thing: people adopt agents much faster when the agent lives where they already spend their day.
For most people, that place is a messaging app. Not a new app to install, not a dashboard, not a browser tab they will forget. A chat. Telegram made sense as a starting point because it is open to bots, works on every device, and people already know how to use it.
The bet behind Nino is simple: a personal agent should feel like texting someone helpful, not like learning new software.
Five lessons from building Nino
Lesson 1: Onboarding is the product
In consumer products, the first two minutes decide everything. If setup feels like work, people leave before they ever see what the assistant can do.
That is why "live in two minutes" was not a marketing line for us. It was a design constraint. Every step that could be removed, we tried to remove. Every question we asked the user up front had to justify itself.
The lesson I would pass on: measure your onboarding in seconds and steps, not in features. A brilliant assistant that takes fifteen minutes to configure will lose to a decent one that works immediately.
Lesson 2: Memory creates trust, and responsibility
What makes a personal assistant feel personal is memory. When it remembers your preferences, your recurring tasks and the context of earlier conversations, it stops being a generic chatbot and starts being useful.
But memory is also where trust is won or lost. People need to understand, in plain words, what the assistant remembers and why. They need to feel in control of it.
My view is that consumer AI products should treat memory as a privilege granted by the user, not a data asset collected by the company. Be clear, be modest, and give people control. Anything else eventually backfires.
Lesson 3: Proactive beats reactive
Most people do not wake up thinking "what should I ask my AI today?". A chatbot that only answers questions gets used for a few days, then forgotten.
What changes the relationship is when the assistant does something on its own, at the right moment: a reminder at the time you asked for it, a morning summary, a follow-up on something you mentioned. OpenClaw's scheduled tasks make this possible, and in our experience this is where a personal agent starts to earn its place on someone's phone.
The lesson: design for the moments when the assistant reaches out, not just the moments when the user does. And be careful with it. Too many messages and you become noise.
Lesson 4: Say no to most features
Large language models can do an enormous range of things. That is a trap for product builders. It is tempting to list everything the assistant could do, and end up with a product that is good at nothing in particular.
The discipline I believe in is to pick a small number of everyday jobs and make them feel effortless. Reminders, quick answers, drafting a message, organising a few things. Then add carefully, based on what people actually use.
This is the same principle I apply to business agents, which I describe in AI agents in business: what actually works in 2026. Narrow scope is not a lack of ambition. It is how you earn the right to expand.
Lesson 5: Language and culture are not details
A personal assistant talks to people in their own language, about their own lives. If it sounds foreign, stiff or slightly off, people feel it immediately.
This matters a lot to me, because ThalerTech Group works across Europe, Africa and the Middle East. It is one reason we built Fares AI, an assistant that speaks Arabic dialects, rather than assuming Modern Standard Arabic or English would be enough. Consumer AI has to meet people where they are, linguistically and culturally.
What building Nino taught me as a founder
Beyond the product lessons, Nino reminded me of a few things about building in general.
Speed of learning matters more than speed of shipping. Launching fast is only useful if you are actually listening to what happens next.
Consumers are honest. A business client may tolerate friction because a contract was signed. A consumer simply stops using the product. That honesty is uncomfortable and extremely valuable.
Small teams can ship real products. Open-source foundations like OpenClaw mean a focused team no longer needs to build everything from scratch. The leverage is in choosing what to build on top.
I have written more about why I keep starting new things, and what I would do differently, in Why I keep building companies.
Key takeaways
- In consumer AI, onboarding is the product: measure it in seconds and steps.
- Put the assistant where people already are, such as a messaging app.
- Treat memory as a privilege granted by the user, with clear control.
- Proactive help builds habits, but too much becomes noise.
- Keep the scope narrow and respect language and culture from day one.