How I use NotebookLM and Figma Make in my ideation workflow
#76: Design process in a large organization with NotebookLM and Figma (Make)
Welcome to Fundament, a weekly product design newsletter where we share actionable tips and insightful stories with the worldwide design community. Join 2,700+ readers and grow as a UX and product designer with us!
How I use NotebookLM in my ideation workflow
Reading some articles and browsing posts on LinkedIn, you might get the impression that everything has already died or will die within the next few months. Design is dead, the design process is dead, the testing profession is about to die, the product manager role is about to disappear, agile methodologies are dead, Figma is dead, and the list could go on for quite a while. If you are interested in this topic, it is also worth taking a look at a recent post by Debbie Levit.
However, I would like to reassure you that nothing has died, and nothing suggests that this will change anytime soon. Of course, we can clearly see that things are evolving. People are starting to work differently, with the growing trend of product builders being a good example. Yet even if we all eventually become product builders, many things will remain the same. Perhaps the tools will change, but the core of the work will not. For that reason, I would not be so quick to declare the death of anything.
With that in mind, I would like to show you what my design process looked like in 2026 while working on a new product with the support of AI. This will not be an impressive showcase of building a great product and pushing it to production within a few hours. Instead, it will reflect a real process inside a large organization where AI strongly supports the work.
Some people will probably say that this approach is outdated and that I should be able to deliver a product to production within an hour. In reality, however, this is not how things work in the vast majority of large organizations. They simply cannot afford that level of experimentation due to budget limitations, legal constraints, resource availability, internal company politics, or even the existing IT infrastructure.
What did my process look like?
Toolstack
NotebookLM, Gemini, FigJam, Figma, and Figma Make.
For those who are not familiar with it, NotebookLM is an artificial intelligence tool based on the latest models from Google. It works as an intelligent assistant for analyzing and organizing source materials. One of its biggest advantages is that it operates only on the sources you provide, which significantly reduces the risk of hallucinations. In addition, every piece of information is linked to the exact part of the document it comes from.
It is an excellent tool for contextual analysis, data synthesis, and as an assistant during the discovery process. It can also work very well as a knowledge base for a product and its features.
Let’s return to the process.
What did it actually look like?
I will start by saying that I collaborate with product managers almost from the very beginning of the product development process. In this case, when I joined the project, there was already some information collected by the PM waiting for me. This included the initial assumptions for the new product, goals, some information about users, an early proposal for functionality, and the results of research conducted with the target group.
1. Context and Discovery
At the very beginning, I moved the information I received into NotebookLM and asked it to summarize it so I could familiarize myself with the project and establish what we already knew. I also asked for a separate summary of the attached research results.
Next, using the deep research feature in NotebookLM, I conducted desk research on the two biggest competitors of our new product. I expanded this research by manually collecting additional data that the model was not able to provide but that I considered important, and I added that information as well.
2. Analysis
Based on the data collected by the PM, the research results, and the materials from my own research, I then asked the model to identify gaps, risks, and inconsistencies in the collected materials. I was particularly interested in differences between internal knowledge and external insights in order to identify areas that required deeper investigation or clarification. The goal was to make sure these issues would not negatively affect the product being developed and to understand which areas required special attention.
At the same time, I asked for a list of the ten biggest frustrations users experience when using competitor services, as well as the must have features for the first version of the product. We wanted to understand the main pain points that our product could address. I received a ready to use list that included priorities, proposed solutions, and the value for the user. I also received a proposed UVP which, interestingly, aligned with our initial assumptions.
This step was the result of an analysis of the previous stages of the process that I asked Gemini to perform. It suggested specific steps and questions and prepared prompts for NotebookLM that significantly improved the quality of the responses.
I received a list of topics that clearly required deeper product work. Some of them turned out to be unnecessary because they resulted from a lack of knowledge about certain aspects of our product, for example how advertising works within it. Despite that, the questions themselves were still valid.
Next, I asked for proposed solutions to address all of the identified inconsistencies as a starting point. I also asked Gemini to do the same, since it can be connected with the context from NotebookLM. It is important to be mindful of sensitive data when doing this. Both tools proposed slightly different but equally interesting solutions. Gemini then combined both responses into one comprehensive answer, adding several additional questions and suggested solutions across all areas.
3. Alignment
Together with the PM, we analyzed the questions and the proposed solutions. We provided answers to the questions and decided to move forward with some of the proposed solutions. We also reviewed the inconsistencies that had been identified between the internally documented requirements, the proposed functionality, and what we had learned from research and desk research, and clarified them. I then provided the entire updated material to NotebookLM so it could update its knowledge base.
4. Concept & Validation
Once we felt that the collected data was sufficient, I asked NotebookLM to prepare a prompt that would allow me to build the key product view based on the gathered information. I then used this prompt in Figma Make, which produced very satisfying results. In fact, the generated view reflected our assumptions and addressed the user frustrations that had been identified earlier.
The generated view served as the foundation for workshops with the PM.
During the first iteration, we analyzed what we had received. We went through each element and decided what should be removed, what required modification, and what could remain unchanged. It turned out that several elements were unnecessary, so we removed them. Some were modified, and we also kept a few solutions that we had not originally considered but found worth exploring.
After the first workshop iteration, we conducted an initial analysis with the technical, advertising, and SEO teams. We gathered information that allowed us to move forward and ultimately confirm the content of the first version of the view. The most important factors were those affecting SEO and the frustrations most frequently identified during the research.
After collecting insights from these analyses, we updated the prototype during another round of workshops. We then conducted usability tests with the target group, which provided additional feedback that was incorporated into the prototype. In total, we completed three iterations of the prototype, along with smaller improvements along the way.
5. Design & Handoff
At this stage, we had already finalized the content of the view. The next activities were more technical. Among other things, I mapped the elements of the view to our design system and documented what we were currently unable to support and which components would require modifications to meet the new requirements. I also began writing design and technical documentation for individual features, which is needed for deeper analysis and implementation planning. This work was also supported by AI, which significantly accelerated the process.
At this stage, I also prepared a more refined version of the view, composed mostly of design system components, the appropriate advertising grid, and the standard elements used across our products.
Work on the overall look and feel of the product also began at this stage. All of these activities were carried out in Figma.
The final step involved expanding the components and preparing production ready views based on the materials developed in the previous step. This included the key view we had been working on, as well as additional views that were required. Some elements were simplified due to cost or time constraints. Everything was prepared in Figma.
After that, the handoff took place and the next stage of collaboration with the engineering team began as the product moved into implementation.
In short, this was my process. At most stages it was supported by AI to a greater or lesser extent. This allowed us to save a significant amount of time compared to a process without AI support, while also helping us deliver a higher quality product. AI helped highlight areas that might otherwise have been overlooked and supported the creation of higher quality materials, such as documentation.
One very important point is that at every stage the final evaluation always came from us. We never treated the outputs we received as unquestionable facts. Effective work with AI requires the ability to manage and critically evaluate what it produces, and to determine whether the information is valuable or useless.
The key ingredient in this process is the knowledge and experience of the people working with AI. This is the part that cannot be replaced.
☕ Support our work
Premium is a simple way to support us as creators for just $5 a month (or $45 a year).
Partnerships and socials
Partner with us via Passionfroot
Arek’s LinkedIn
Mateusz’s LinkedIn