Product managers and leaders, make your GenAI product vision real!

Documents and static prototypes just don’t work to fully embrace the GenAI revolution. Proof, magic, understanding, and validation for AI features and products only comes with real, working applications

Viral Shah
Published in
8 min readJan 18, 2024


It’s March 2023 and ChatGPT just made GPT3.5-turbo available to the masses. We at Databutton had been readying ourselves for months, waiting for this tremendous disruptor to all software products. What would this mean for us, at the time a low code platform for building data apps? It took on many flavors:

  • Our users wanted to build apps leveraging the best AI tools and services (GPT3.5-turbo being the top one)
  • People started copy/pasting code generated from ChatGPT into Databutton
  • We built our own coding assistant, Databutler into the product

Going AI Native

Taking all these learnings, we discovered we could rebuild the foundational product experience to be AI-native with huge customer benefits like creating higher quality apps while closing an even bigger knowledge gap. In our case, AI-native means interacting directly with AI agents to build the entire application. But making that vision real? Wow, that was hard. We had to consider things like:

  • Conversation design and tone of voice
  • Stretching the limits of contextual awareness for agents
  • New modes of human-computer interaction and quality of outputs
  • Information and UX design when outputs are probabilistic
An attempt at designing user-agent conversations in Notion

How do you contribute to such an open ended problem space as a product leader or product manager? You need to get into the details. I tried so many things, but it quickly became clear that “defining” things up front was a hope and pray strategy.

Our brilliant Head of Design putting together contextual generative UX elements in Figma, but would they work?

Truly participating as a product leader

In the end, we had to get our hands dirty. What did that mean? Stretching ChatGPT to its absolute limits as a prototyping device. Here’s an example trying to generate markdown that can be converted to an app wireframe, and building real apps on Databutton v1 that we could test internally and with users.

All product leaders and managers need to go that deep and build real apps to communicate, validate, and sell their ideas.

Let’s zoom out and look at the landscape 👇

GenAI needs to be foundational to your strategy

Businesses need to understand and act on the changes GenAI is bringing in order to stay relevant and win. GenAI is a skill leveler, increases productivity, and is bringing down the cost of goods and services. Those are macro trends that are going to have a huge impact.

What does that mean for product companies? Product leaders need to establish a clear and actionable AI direction that can transform their product experience to make their users and customers more productive and satisfied solving the core job-to-be-done. Product managers need to translate that direction into tangible features that wield AI to its fullest extent, used in the context of their product’s workflows and information.

Many great product companies are doing just that. A great example of that is Intercom’s new AI direction, which they’ve gone all-in on.

Intercom is embracing GenAI and changing direction

Paul Adams, CPO of Intercom talked about their new AI direction on Lenny’s Podcast.

What day did ChatGPT launch? November 29th, I think, last year. Ever since that day, I literally wake up every day thinking about AI pretty much. And, I read as much as possible and still feel like I’m way behind in it…Everything has changed. So, we literally ripped up our strategy almost entirely, and started again, from first principles and said, “Okay, why do people use Intercom?” Intercom is a customer support product. And then, very soon after that, Sam Altman, who’s the founder and head of OpenAI, said, “Hey, one of the first industries that’s going to be disrupted is customer service.”

But how do you make a new AI vision, or even roadmap item, tangible? Product managers and leaders need to act as the driving force. An ever changing set of capabilities and opportunities requires you to stay on top of what is possible. There are very few pre-written rules when it comes to working with and building experiences leveraging GenAI — the only way to know what works is by building. This means going beyond specing and wireframing to building apps and actually using the AI tooling available for communication, validation and direction setting.

Thankfully, the opportunity is ripe for PMs with code generation tools hitting the market. You can use every tool in your belt, like describing detailed logic in text and wireframing, to actually build apps. PMs are situated best to understand your business requirements, your customer needs, and the actual technical dependencies in order to build real apps leveraging GenAI.

Traditional tools like docs, wireframes, high fidelity prototypes don’t work

GenAI features, roadmap items, or vision simply cannot be expressed in written docs, wireframes, or even Figma prototypes. The proof and tangibility is simply non-existent. Let’s quickly break down what these tools are good for.

Written docs

A great tool to communicate ideas and direction with clarity is the popular Amazon working backwards method of writing press releases. It’s great in its customer centricity and focusing on outcomes. However, how can one write a realistic vision for that press release without understanding what is possible?

This is a fantastic framework when the rules of the game are set, but how do you evaluate opportunities without knowing what’s possible — especially in your context?

Without a nuanced understanding of GenAI capabilities, low fidelity ideas aiming for 10* experiences dominate — can’t AI just solve everything for the customer, what about a magic button? This will inevitably lead to wasted efforts and no tangible value for customers.

Low fidelity wireframes

Wireframes are a great tool for information and workflow design. But the reality is that the entire space of human-computer interaction is being disrupted by AI. You cannot stick to the standard methods of users providing intent to your product and getting the intended output anymore.

Wireframes don’t work for GenAI features. Image credit: Eddie Lobanovskiy

High fidelity prototypes

This is mostly Figma prototypes nowadays. The goal is to make the product feel as real as possible to communicate and learn. But there’s a huge weakness here for GenAI tools — how do you express data and probability-heavy experiences like conversation design, tone of voice, variability of output, and more in static images? Sure, you can have static text that pops up or images that are “generated” when clicking a button, but it doesn’t convey the most important parts of your product experience.

The proof, magic, understanding, and validation comes only with real, working applications.

You need to go deep and build with real data and LLM APIs to communicate, validate and sell

In reality, this has always been a true for data and intelligence-driven products like:

  • Ranking algorithms for any search functionality
  • Recommendations for music and video platforms
  • Dashboarding and BI tooling inside SaaS products

When I worked on AI and search features for Office 365, we always had to communicate and validate features with live-data prototypes. Static tools just didn’t cut it.

Live-data prototypes

From Marty Cagan in Silicon Valley Product Group:

“The main purpose of a live-data prototype is to actually prove something — normally it’s to prove whether an idea (a feature, a design approach, a workflow) really works. In order to know this, we typically need to do two things. First, we need the prototype to access our real data sources — like actually search our live inventory and show products that are really available right now. Second, we need to be able to send live traffic, in quantity, to the prototype.

You do have to keep in mind two big limitations. First, this is code so it requires your developers to create the live-data prototype and not your designers. Second, this is not a product, and you can’t run a business with it, so if the tests go well, you’ll still need to allow your engineers to take the time to productize the code.”

The key phrase to me here is to prove whether an idea really works. Using the classic IDEO framework — product managers and leaders should be able to answer what is feasible, desirable, and viable. GenAI is a space where feasibility needs to be tested out in tandem with desirability and viability. You cannot project or predict what is correct yet.

The value of rapid iteration is immense when working with probabilistic AI models

Through our experience building Databutton we’ve seen in particular that rapid iteration is invaluable. Small changes in prompt engineering, providing context, and more yield incredibly different results. This makes it all the more important to be in the drivers seat and actually build real applications.

Databutton is the copilot for product managers to generate apps

Want to meet your new favorite colleague?

At Databutton, we believe the future belongs to narrators, and we’ve built the perfect copilot to leverage the communication, problem solving skills, and domain knowledge that product managers possess in order to generate applications.

You no longer have to prioritize development time to communicate or validate an idea. You can test many ideas at once, or many directions for solving the same problem.

Databutton apps are full stack React/Python apps, so you can work with any data source or API, build out complex logic, and expose it through excellent interfaces. You use your skills to generate real code using the best frameworks, making collaborating and handover to devs a breeze.

So, product managers, are you ready to meet your new teammate? Sign up at and let us know what you think on Discord.



Viral Shah

Co-founder and CPO at Databutton: Build AI apps at the speed of thought