Building with AI as a Product 'Builder' in 2026

Personal Planner and Tracker

Platform

Web, Mobile (Browser)

My role

Solo Builder

Tools

Claude, Lovable

Try Persimmon

Short on time? Listen to quick audio version

LEARN BY DOING

Goals

Goals

  1. Build a tool that fits my needs perfectly
    I used to use a planner tool called Friday, which got sunset a while back. I loved how simple it was. It let you visually block out your day on a calendar, so at any point you knew exactly what you were doing and what came next. That alone made me noticeably less anxious and more organized. The second need came from a different habit. I had also been using Toggl, to understand where my hours actually went. Whether a task took the ten minutes I assumed it would, or somehow ate my whole afternoon. I have been looking for something like it ever since, but everything I found was either too basic to be useful or so overbuilt I didn't want to learn it just to plan my day. I wanted both of these things, a simple time-blocked planner and simple time tracking living in one lightweight tool.

  1. Try building something end-to-end and upskill
    AI tools have made shipping possible for solo builders and honestly, I wanted to try it - build something from start to finish, on my own, instead of just wishing it would exist. And as a product designer, I wanted to push my boundaries and upskill. I started with a "prompt-to-code" tool to see how far it could actually go, building something fully functional, at least for my own use to start. I picked "prompt-to-code" deliberately, not "AI-coding." It felt like the right first step, before jumping into writing code alongside AI. I also wanted to test the combination itself - Lovable to build and Claude to think and write the prompts.

3 tools and it did not include Figma

Tools

Tools

Notion - vision, document


Notion is the starting point. Raw thoughts and ideas from my head reside here. This gives me a canvas before AI and building even enter the picture. A beat to think alone before the process picks up speed.

Later, I come back here to make a few notes about what I build and ideas for later versions. So, it really becomes a 'home' for the project.

Notion - vision and document

Notion is the starting point. Raw thoughts and ideas from my head reside here. This gives me a canvas before AI and building even enter the picture. A beat to think alone before the process picks up speed.

Later, I come back here to make a few notes about what I build and ideas for later versions. So, it really becomes a 'home' for the project.

Claude - ideate, iterate


In the truest sense, this replaced Figma for me and became my canvas (also a deliberate tool-switch I wanted to try out). What comes out of my Notion, I refine it here: sketching a few different directions before settling on one, analysing screens against UX principles and writing prompts, sometimes from a rough voice note and sometimes from a screenshot of something that didn't come out right.

Claude - ideate, iterate

In the truest sense, this replaced Figma for me and became my canvas (also a deliberate tool-switch I wanted to try out). What comes out of my Notion, I refine it here: sketching a few different directions before settling on one, analysing screens against UX principles and writing prompts, sometimes from a rough voice note and sometimes from a screenshot of something that didn't come out right.

Lovable - build, test, ship


For bigger features, I really liked the 'Plan' feature where it reasons through the approach, instead of jumping straight to code. Two features saved real time and tokens: 'Select element', which lets me click directly on a piece of the interface and make direct edits. And 'Text edit' for the small copy fixes that don't need a whole prompt. I also published the test link and shared it with a couple of close friends for feedback.

Lovable - build, test, ship

For bigger features, I really liked the 'Plan' feature where it reasons through the approach, instead of jumping straight to code. Two features saved real time and tokens: 'Select element', which lets me click directly on a piece of the interface and make direct edits. And 'Text edit' for the small copy fixes that don't need a whole prompt. I also published the test link and shared it with a couple of close friends for feedback.

A different (and one of the) process

Process

A snapshot of the process

1 / Thoughts and vision


I spent a couple of hours in Notion, writing down why I wanted this. Friday, the planner I used to love, had been sunset. I wrote down what I missed about it and separately, what I liked about Toggl, the habit of knowing where my hours actually went.

Notion: all my initial thoughts and ideas about Persimmon

I brought these to Claude and asked it to turn my rambling into something clear enough to build from, i.e. PRD.

2 / Ideate and iterate (on the prompt!)


I wanted to see if I could one-shot my way to a decent build (first-pass). Arriving at that prompt took 2-3 passes. Once it was ready, I handed it to Lovable, and a working first version came back. This is where I stopped typing prompts properly and started talking instead. Rambled through an idea out loud and let Claude turn it into something structured. Felt like less work and somehow produced better prompts than when I sat down and tried to write them properly myself.

The prompt that help me one-shot the first version!

3 / First build + Testing


I shared the preview link from Lovable itself, to a few close friends. I wasn't exactly looking for a full-blown, formal user feedback. I just wanted to have another set of eyes looking at it and spot any obvious bugs or flaws. (and also what they think about the planner overall and would they use it as a product :)

This feedback helped me get a fresh opinion on the experience ad fix some bugs.

4 / The loop (+ mentionable moments)


This is the part that shaped Persimmon the most. Instead of Figma, Claude was my canvas. I iterated there with prompts, and only took something to Lovable once it was ready to build, or when something needed debugging, say explorations, UI iterations, color options and quick brainstorming sessions.

/ Writing prompts with Claude
This way has been very effective for me: Dump your thoughts on Claude, say what you want out of the prompt, and let it write it. Just make sure Claude has the context first (here, e.g. a simple web-based planner).

Prompt I wrote with Claude to build Time Tracker's stat section 'one-shot'.

The same technique works even better for tweaking visual and functional details for some features. e.g. here, defining how the 'time blocks' should be rendered.

💡

Tip 1: For this to work effectively, Claude (or any LLM) should have all required context. And what input you give for it to write the prompt, will largely decide the quality of the prompt.

⚠️

Tip 2: It gets ti right 90% of the times, although it's a good habit to review the prompt once before using; as Claude tends to add extra stuff (what no to do, constraints and the the typical "not this, but that" stuff :D)

/ Typing fatigue is real
One downside of prompt-to-build is that you end up typing a lot and it gets tiring. Claude's built-in speech-to-text has been a lifesaver and tools like WisprFlow make it even easier to get your thoughts out without you having to type every word.

🧠

Extra: Claude can also be a good thinking partner when exploring new features or additions. The part where I found it most useful is it flags any potential conflicts with existing architecture and systems, helping me spot things I might have missed before jumping into building.

/ Exploring multiple directions in lesser time
Since it’s well established that LLMs are faster at generating outputs, I deliberately wanted to try using one for rapid UI exploration. It took significantly less time than working on a traditional canvas like Figma, which meant I could explore multiple directions quickly before settling on one. I think this works particularly well for low- to mid-fidelity exploration, where the goal is to iterate quickly, compare options and see what direction is worth pursuing.

I had a quick brainstorming session with Claude to explore different ideas for the landing page.

In some cases, I found it easier to prompt the UI and quickly generate a few options for the “Time Tracker” than to build and explore them manually in Figma.

💬

This might not be the best way to do it, but it definitely saved me time and effort. And isn’t that a win? Next, I want to try prompt-to-UI with Paper, where I am hoping to get better consistency and stronger alignment with the brand.

/ Faster UI iteration loop
Beyond UI exploration, I found Claude useful for refining existing UI as well. I would take a screenshot of the current Lovable output, highlight what I wanted to tweak and give it to Claude. It could then generate a visual mockup and translate the change into a prompt I could take back into Lovable. This made the process much more iterative. I could quickly visualise a change, refine it and then hand it back to Lovable once I was happy with the direction.

💡

Tip: While asking Claude to generate mockups, I also started sharing the rationale behind my design decisions, even when it wasn’t strictly needed for the immediate task. What I found useful is that Claude retains this context, which becomes valuable later when exploring new ideas or changes that might conflict with an existing decision. It's suggestions become more grounded in the existing product context.

Tools are ever evolving and this definitely isn't 'the' process; but it has opened multiple new ways to build stuff and it's very exciting!

I found myself spending less time translating ideas between tools and more time directly exploring them, talking through problems, testing directions, building, breaking and iterating.