Building Modern Parent

What I learned from building an AI bot in three days this week.

When my friends had their first child, they spent more time researching nurseries than planning their wedding. Dozens of tabs open, conflicting reviews, hidden fees—and mounting frustration. Watching them navigate this chaos made me wonder: Could there be a simpler way?

I spent most of this week exploring this question by building an AI bot—codename Modern Parent—to help young parents navigate some of the earliest challenges of parenthood, especially choosing the right nursery.

Why an AI bot?

After conversations with new parents, a clear pattern emerged: they were overwhelmed by fragmented, impersonal information scattered across countless websites. Every parent's situation was unique, yet the available resources felt generic and disconnected.

Nursery selection is often most frustrating. It seems straightforward until you actually do it: location, cost, educational philosophies, facilities, discounts, government support schemes, Ofsted ratings, waiting lists—each nursery has dozens of variables to consider. Even worse, nursery websites often lack details, making comparison hard.

AI could organise this fragmented information and personalise it. Imagine your mother offers to care for the kids on Fridays, or perhaps you're enrolling twins and want the exact price with a discount applied. This personalised clarity is what Modern Parent aims to provide. Down the line, the AI bot could call nurseries to gather missing information or arrange visits, evolving from a bot into a full-fledged AI agent working on behalf of a new parent.

A mockup of Modern Parent.

Choosing the Tech

My goal was a minimal viable product (MVP) built within three days, prioritising speed over perfection, allowing me to quickly return to customer development.

Here's what I used:

• WhatsApp as the user interface, since every parent is already there.

• Wassenger to manage WhatsApp interactions effortlessly.

• OpenAI API for handling AI conversations (with the new Responses API).

• OpenAI API also doubled as my vector store, simplifying things.

• ScrapingBee for extracting nursery data.

• Render for easy deployment.

• Replit Agent to rapidly spin up a landing page.

• Cursor as my AI-powered IDE.

• ChatGPT as my thinking partner to clarify ideas along the way.

The process of software creation is evolving incredibly fast. Today, it’s less about writing code and more about collaborating with an AI model to design and build the software together. This process is way more fun than simply writing code the old-fashioned way.

Thinking through the architecture with ChatGPT instead of mulling it over in my head is both faster and more interesting. It’s just so much faster and more fun to work in a team of two (with AI) than just on my own.

Instead of staring at a blank page or debugging alone, I described my problem to ChatGPT—‘How should I store personalised nursery preferences securely?’ It replied with clear ideas, immediately removing mental blocks.

Defensibility: Personalisation and Nuance

One of the biggest questions I hear entrepreneurs discuss these days is defensibility. If software becomes easier and cheaper to build, what makes it defensible? Why invest in building a product if someone can just ask AI “build me a copy, but even better” and it will do it?

When I imagine AI getting exponentially better in the coming years, I see two potential options. One is that one or a few general-purpose AI agents will dominate most use cases. That is, one AI agent will be able to help a new parent looking for a nursery and a truck driver with a broken transmission in the middle of the night.

Another is that we’ll see an absolute explosion of AI agents that specialise in relatively narrow domains (e.g. helping new parents) and are defensible through network effects, proprietary data and other traditional business moats. This would be similar to how we have lots of smartphone apps for different tasks.

In this specific case, I think that defensibility will come from the data collected during use and from deeply understanding human needs. Generic nursery comparison tools already exist, but real parents’ lives aren't generic. What’s valuable—and hard to replicate—is understanding the quirks of real life: Grandma covering Fridays, a nursery with bilingual staff, or special discounts.

Off-the-shelf vs Custom Development

Why did I choose to build Modern Parent from scratch instead of using one of the countless off-the-shelf platforms?

The main reason was to give me the flexibility I needed. For example, this project requires scraping nurseries in specific geographies, which I wouldn’t expect an off-the-shelf platform to help me with. On the contrary, if I had spent a day or two building on someone else’s platform only to discover that it has an important limitation, it would be expensive to switch to a different platform.

Another reason was that from a technical perspective, I don’t think I would have saved much time by using off-the-shelf tools since AI-assisted development is so fast. There is a learning curve everywhere, but I’d rather face the learning curve of the new Responses API from OpenAI because this knowledge will be easily reusable.

In my experience, AI agent platforms like agent.ai or Lindy AI or Relevance AI, are great for specific, focused workflows with clear inputs and outputs, especially for non-technical users. However, for this project, it was easy to choose to write code from scratch.

Three Things That Surprised Me

Maybe the biggest surprise was how hard it is to adjust the tone and character of the default gpt4o model. It has a certain personality and it’s not very easy to override it with simple instructions, so I expect far more work on the system prompt in the coming days.

The second biggest was how cheap LLMs are. Sure, we’ve all seen the headlines about dropping costs. Still, it was a surprise to see that three days of development and non-stop use of OpenAI API cost me less than $5.

The last surprise was how easy and fun was to do this project. I felt like my experience from years ago (I used to be a professional software developer and studied CS) was very relevant but I wasn’t blocked if I didn’t have experience with some of the newer tools. A quick conversation with ChatGPT would help me find the way forward.

The Future is Personalised

After three days immersed in this project, I could feel the promise of highly personalised AI solutions. This bot will remember if you tell it that you live in Twickenham, London and that you had your first child 2 months ago, and will use this information to help you find the right nurseries. Building it helped me realise that we take it for granted that we need to google generic information and then personalise it to our needs. But it doesn’t have to be this way.

That’s exactly the kind of intuitive, personalised experience Modern Parent—and truly valuable AI—should strive to deliver

If you’d like to try Modern Parent, just send it a whatsapp message. Currently, it only has nursery data for Twickenham, London, but I’ll get it started on scraping the entire London this week.

I'd love your thoughts—what other parenting challenges do you think could benefit from thoughtful, personalised AI solutions?