Overview
Working on multiple AI-enabled projects,I will share my process and insights into how I leverage AI into building digital products, which I hope to launch.
My tech stack and high level use case
- Figma to ideate and design screens
- ChatGPT 5.2 as a general technical design partner
- Claudee to code (via Co-Pilot in Visual Studio Code)
- Wordpress and Local to test and launch websites
- Supabase for databases
- Vercel for deployment
- n8n for automation
- Github to store, share, and work with others
High-level process
- Discovery – researching and defining users, product, ideas, and prototype screens
- Quick prototyping – building libraries, user flow screens on Figma
- Building with AI – documentation, defining components, building first flow
- Test and iterate – with defined components, it’s much easier to adjust designs built via AI
- Beta test
Skills acquired / learning
- Research (early-user surveys, identifying problems to solve, etc.)
- Designing branding kits and component libraries to scale builds
- Stakeholder collaboration and hand-overs (with partners and software devs)
- High-level understanding of documentation and files (README, environment variables, system prompts, file trees etc.)
- Efficient prompt engineering
Goal
To understand the problem or briefing at hand, depending on how ambiguous the projects seems
Process
I personally don’t have a particular process here. Ideas for apps and SaaS can come and be verified in a variety of ways. Generally, speaking to one industry expert can yield a lot of niche problems and then validating it with surveys or a quick MVP is the best way to validate a problem.
Learnings
- Even early on, managing expectations on what’s feasible with stakeholders is important in aligning on what a sufficient MVP looks like
- Creatively framing questions to yield particular insights in understand the problem-spaces – instead of asking general questions like “do you like this?” ask situation-type questions: “when you first used it, did you encounter any difficulties?”
ℹ️ Pro Tip: want better AI outputs for any work flow?
Ask it planning-type prompts before executing plans to refine how it works to ensure better output.
Some questions include: do you have anything to clarify? Or give me a confidence rating on the plan and why you rated it that way.
Ask it planning-type prompts before executing plans to refine how it works to ensure better output.
Some questions include: do you have anything to clarify? Or give me a confidence rating on the plan and why you rated it that way.
Goal
Initial prototyping involves building quickly and setting up the build for future scaling, particularly using clearly defined documentation and component libraries
Process
It is possible to test everything with a simple pen and paper analogue prototype, but I find the true validation of an idea comes form having a higher fidelity prototype.
Building a prototype ultimately depends on the problem, what the solution requires, and experience to anticipate what a ‘good’ prototype looks like. Specifically, in our age of AI where the use of AI itself is the product, often times I find that more time is required to refine the AI output as the solution itself (via system prompts, API setup, finding the right model) and to build the environment around it (functions like save, buttons, text edit, user flows).
Learnings
- Being clear and defining what is a ‘sufficient enough’ MVP to ship to market
- It is a good idea to define a library of core components early. This allows future scalability to be easier – just re-use and run with the AI
ℹ️ Pro Tip: want to ensure early UI consistency without burning tokens?
Ask it to generate a ASCII wireframe of the UI based on your prompt so you can make edits easily before it builds. I’d recommend learning basic CSS to produce more guided UI prompting, particularly: containers, margins, paddings, divs and flexboxes.
Goal
Testing a working solution to gain external feedback from a small group of beta users to refine the product before launch
Process
The hardest part is getting beta users. Leveraging existing friend groups or connections is the easiest and lowest-risk way to test a product. The waitlists can come once the product is more refined.
Testing was conducted by: deploying the product, observing user interaction with product in-person, and conducting surveys.
If the previous steps are done correctly, implementing feedback should not be *too* difficult. It’s a matter of prompting the AI to re-use components to expand or cull features and elements.
Learnings
- Don’t be personal about your product and always lead with a human-first compassion
ℹ️ Pro Tip: want the AI to yield a better understanding of specific UI changes?
Always lead your prompt with: these are isolated changes and follow with clearly listed problems. For each, clearly include: what’s not working, what is a working solution, and what’s the actual goal of your UI.