Practical takeaways for faster prototyping, customer demos, and vendor exploration
By now, the majority of product managers have vibe coded, whether we have admitted it or not
The term is even in Merriam-Webster’s slang and trending section. Vibe coding is the practice of writing code, making web pages, or creating apps by just telling an AI program what you want and letting it create the code and the product for you.
In practice, you describe the app or functionality you want in plain English (prompts). A large language model (LLM) interprets those prompts and generates code to create a working prototype. If it is not perfect, you refine with feedback until the desired behavior is achieved.
Why vibe coding matters for product managers
What amazes me as both a product manager and a consumer is how practically anyone can now build something and see an idea come to life (even my son, who does not know frontend coding, spun up a UI for a Python project on screen breaks). For product managers without formal technical training, vibe coding is empowering.
- Prototypes in minutes. You can test a flow, share an idea, or collect feedback without pulling engineers away from their work. I have thrown away many initial ideas after prototypes showed me what would not work. For example, I tested different data ingestion methods and quickly ruled out poor fits, freeing my focus for better options.
- Playful experimentation. Even lighthearted projects such as rock-paper-scissors-lizard-spock can spark creativity. Any Big Bang Theory fans? I built a version using PubNub’s services, which powered both the two-player game logic and basic matchmaking.
- Customer show and tell. At PubNub, where solutioning is often highly technical, I use our services to build demos tailored to a customer’s use case. With the PubNub MCP Server and vibe coding, I can spin up prototypes that make conversations more concrete for them and faster for me. This helps customers visualize proposals, validates feasibility earlier, and often shortens onboarding.
- Vendor exploration. Shopping for vendors has changed. In the past, evaluation required long discovery cycles with engineering involvement. Now, I can prototype with a vendor’s APIs myself, test fit early, and bring only a short list to engineering, compliance, and finance. It also works in reverse. Prospects and customers can vibe code with our MCP Server to quickly validate whether PubNub meets their needs, accelerating their decision-making.
- Unexpected inspiration. Seeing your idea reflected back in a working prototype often sparks new ideas that would not appear from writing requirements alone. For example, while mocking up a notification flow, I realized it could extend into a broader engagement system with personalized offers. Of course, the flip side is falling into rabbit holes that stray from the original goal.
Lessons I have learned so far
There are many tools for vibe coding. We have tried tools like Lovable, Replit, Bolt.new, Cursor (with Anthropic’s Claude or OpenAI’s ChatGPT), Figma Make, Visual Studio Code with GitHub Copilot, and Vercel v0. All make vibe coding accessible, and the space is evolving quickly. A few lessons stand out.
- Be focused, specific, and clear. Do one thing at a time and break ideas into small rocks. Instead of asking AI to “build an engagement app,” break it into smaller tasks such as “create a simple login flow” first then “add a chat feature where users can post messages and emojis” next. You can even fine tune the UI last. When prompting, be clear not only about what you want added but also what should stay unchanged. This avoids AI rewriting sections that already work.
- Check in your code often. Just like “save early, save often,” AI-generated code can drift from your intent. A small change might rewrite working sections, introduce bugs, or shift style. Use version control so that when you reach a working point, you capture it before moving on.
- Start a fresh chat often. Long conversations confuse AI. Open a new thread for each task or milestone. You can even ask AI to summarize and generate a clean prompt for the next chat.
- Use AI to help prompt AI. Sometimes one tool can refine prompts for another. I often ask ChatGPT to polish a prompt, then use the output as my prompt in Cursor with Claude. Always review refined prompts carefully, since AI is not perfect and can miss important details.
- Sometimes a do-over is easier. If you are spinning in circles, start over. It feels wasteful, but I have learned it is often faster than untangling broken logic.
- Document and save your prompts. This may feel old school, but it saves time when projects span weeks or when you need a do-over. For me, this has become a new form of requirements.
- Trust but verify. AI is a tool, not an end state. Review generated prompts or what its doing, and do not hesitate to search the web. More than once I found the missing piece in a forum thread that AI had skipped. When I fed it back into AI, it worked immediately.
Noticed a theme? These are the same mantras we have always followed in product management: focus, clarity, iteration, and documentation. The context has changed, but the practices remain true.
What’s next
For product managers, vibe coding is more than a curiosity. It is a new way to bring ideas to life faster, validate solutions earlier, and create stronger conversations with customers, vendors, and engineers. It does not replace product management fundamentals. It reinforces them in a new context.
We are now exploring how to move prototypes into production-ready, scalable features. This means asking questions such as:
- How does vibe coding fit into our software development lifecycle?
- How can we balance speed to market with requirements for data, security, and compliance?
- How do we incorporate AI tools and agents responsibly into workflows while maintaining high-quality user experiences and customer value?
From both a product management and UX perspective, the opportunity is exciting. The challenge is to harness vibe coding not just for rapid experimentation, but as a stepping stone to reliable, secure, and meaningful products at scale.
And from a company perspective, a few pieces of guidance are emerging:
- Empower all of your teams. Ideas can come from anywhere. Early on, many organizations focus on giving engineering teams AI tools. Extending those same tools to PMs and UX designers cuts down steps and boosts productivity. For example, a UX designer using AI to generate interactive flows can bring an idea to engineering already validated with customers.
- Be curious. My early vibe coding experiments happened on personal time, trying tools, watching videos, and simply playing around. Companies can foster that same spirit by carving out time for hackathons or internal AI challenges. These create safe spaces to test tools, stretch creativity, and uncover ideas, while also helping teams understand both the power and the limits of AI.
- Build trust and learn together. AI-driven prototypes can spark mixed reactions across teams. Some worry AI may diminish their role, while others see it as an opportunity. When prototypes are shared, there will always be skeptics and enthusiasts. The key is to frame them as conversation starters, not finished products. This builds trust and creates a learning loop across PM, UX, engineering, and beyond.
Ready to dive in and get started? Check out more resources on vibe coding as a powerful way to accelerate the path from idea to prototype, and from prototype to production. And remember, coding with AI gives teams across disciplines new ways to experiment, validate, and inspire, while keeping engineers focused on scaling the solutions that matter most.
