How to Use LLMs as a Product Manager: Real Examples

In this article, I want to share my experience using large language models (LLMs) to solve product management tasks.

Product management isn't just about the business side of things. It's also about getting your hands dirty and understanding how solutions are built from a technical perspective. Sometimes, the best way to grasp an idea is to create something yourself - whether it's a quick prototype, a visualization, or a proof of concept. This approach doesn't just help me personally. It brings clarity to the whole team. Especially when we're dealing with something new that doesn't exist on the market yet, building even a basic version helps us see what we're trying to achieve more clearly.

LLM is not just a "question-answer" tool. It is a powerful mechanism that allows you to create something new and creative. In this article, I will present three examples that show how I use LLM for creation and experimentation:

• Creating a 3D Snake Game to Assess LLM Capabilities
• Developing small applications for Proof of Concept (PoC)
• Rapid Prototyping for a B2B Matching App

These examples will help you better understand the potential of LLMs for speeding up workflows and optimizing tasks.

Let me say right away: LLMs are not a panacea. They will not replace human experience, but their use in terms of speed, efficiency, and the ability to visualize ideas can be a game changer.

Case 1: Creating a 3D Snake Game to Assess LLM Capabilities

To understand how powerful modern LLMs are, I decided to create a 3D Snake game. This was an experiment aimed at testing how quickly and easily something could be developed using LLM capabilities. With Claude 3.5, I got a basic prototype of the game in just a few prompts.

This case showed me that, by using LLMs, you can create something interactive in just a couple of hours and start testing a hypothesis right away. Of course, the game wasn't a masterpiece, but creating it gave me a complete understanding of the tool's capabilities and how it could be used for more complex projects.

If you're interested, you can check out the result at this link.

Case 2: PoC for Similarity Search

Now let’s move on to developing small applications for Proof of Concept. In this case, I was working on an idea for a search and match mechanism through a chat interface, using similarity search and embeddings. Instead of developing a full-scale application with a backend and all the associated costs, I decided to use the OpenAI API and Google Sheets.

Why this approach? The answer is simple: minimizing time and costs. Of course, you can set up a full-fledged application, and I've had similar experiences using Cursor App, but it's a long and complex route. Here, I just needed to validate the idea, and Google Sheets was perfect for this.

I opened ChatGPT and asked it to create a script for Google Sheets that would help me implement the idea. This resulted in a lightweight and flexible PoC. Here’s what the technical part of the solution looks like:

• Data Structuring and Processing: Company profile data (e.g., industry, location, offerings, and needs) is collected and structured in Google Sheets. This allows for quickly testing core concepts.
• Search and Match Algorithm:

Embedding Generation: The OpenAI model is used to create embeddings for company descriptions and their needs. This helps capture the semantic meaning of the texts.

Cosine Similarity Search: Similarity scores are calculated between companies to find the best matches based on needs and offerings.

• Two-Step Matching Process:

Attribute-Based Matching: Initial matching checks for exact matches based on key attributes like industry and region.

Similarity-Based Matching: Similarity scores are calculated for more detailed and nuanced.

• Scenario-Based Logic: The application takes into account different scenarios, such as searching for investments or cross-regional matches for global partnerships.

The entire backend for the PoC is implemented in Google Apps Script, which makes it a simple, accessible, and cost-effective solution for initial idea validation.

Case 3: Rapid Prototyping for a B2B Matching App

Sometimes, as a product manager, you need to showcase your idea to stakeholders, designers, or the team. Instead of drawing your ideas on a napkin, using an LLM like Claude 3.5 can help bring your concept to life in a matter of minutes. In this case, I wanted to conceptualize a B2B matching app with Tinder-like functionality for companies. Imagine an app where companies can browse potential partners, swipe through matches, and connect based on mutual interest. Here's how I used LLMs to rapidly develop a React component that outlines the core interface and functionality. Link to the prototype

Rapid prototyping with LLMs helps you create an interactive representation of your concept, allowing you to communicate your ideas more effectively to everyone involved. This approach is especially useful when there is no existing reference on the market, and you need to show what your vision looks like in practice.

Conclusion: You don't always need to build something large and complex to test a hypothesis. Using lightweight tools like Google Sheets and LLMs helps save time, avoid unnecessary costs, and quickly understand whether an idea is worth pursuing further.

Use these approaches to make your work more flexible and efficient. Don’t be afraid to experiment with LLMs - it's a powerful tool that, when applied correctly, can become your greatest assistant.

In the next article, I'll describe the approaches I use for building apps, managing my projects, and creating texts. Stay tuned for more practical insights that can help you leverage modern tools in your own workflow.