Life Manager: The Build Log
A vibe-coding case
March 4, 2026 — Build Your Own Software with Claude Code
It's been some time now that I have been hearing about Codex, Copilot, Replit, or Claude code. They are all coding-versions of LLMs, which basically means they can transform sentences into code. I'm not going to go too much into details here, but long story short: what once required years of technical training to build software is now mostly accessible to everyone.
"And so what?", you might ask. Well, what if we could use it for self-improvement?
In early February, I got myself a Claude Pro membership, downloaded the app, and started my first digital project: my Life Manager.
Before we go further, I recommend you think for a couple minutes and try to come up with habits you'd like to improve or create in your life. Once that's done, let me run through this use-case of AI-enabled vibe-coding.
My journey started with two ideas: track my contacts (a direct consequence of living abroad last year) and start journalling. Having an updated and complete view of people I know, and creating a healthy journalling habit was the backbone of this project.
Quickly, I figured out that there were a bunch of things I wanted to improve in my daily life, and that understanding how I, Noe, usually build my habits was the first step. I wanted to build a gamified platform gathering different features like meditation, daily learnings written with Claude's API, activity and nutrition tracker with integrated coaches to help me reach my fitness and health objectives, portfolio management and news generator. I even created a specific wordle mini game with the League of Legends universe (I don't recommend it, it's hard).
What started with a CRM software for my contact transformed into a platform, locally hosted on my computer, that would motivate me to login daily, achieving badges, earning XP and grinding levels.
Claude Code would turn my ideas into plans and plans into code. I would create AI specialized subagents in user interface, testing, and designing. Claude was orchestrating that team depending on my requests. Plan limits would slow down the progress, but give me time to reflect on what I wanted to get out of this app.
This is highly personalized and I don't want to claim that you need the same app. But I think reflecting on your goals, looking ahead, and how to build your future self is a worthy exercise. And Claude Code enables you to build something that will help you reach those goals!
If I feel like it's necessary, I might write a complete guide on how to technically build the app, with guidelines and structure, and document how I created mine. For now, I'll leave you with a few pieces of advice, while still being reachable if you have any questions.
- It may seem obvious with a step back but create virtuous cycles and ask Claude (or any other AI you want to use) how to best use that specific AI. It generates report on what you did right, what could be improved and how to act on those recommendations. Keep the feedback loop. I'm gradually improving my use of AI not by seeing others use it for their personal use-cases, but by understanding how the AI can help me, in my specific use-case.
- I've already made the mistake of using the default model while latest models were significantly better, don't do that. Use the latest models. Models are coming left and right, day and night, and new models are getting exponentially better. Make sure you use the latest models and that you understand the strengths of each one of them.
- Don't try to one-shot a project, iteration will do the trick so start with one or two features, then keep improving the feeling on the app and keep using it. Take notes of your ideas and the issues you've stumbled upon while using it, and when you have time, explore them with Claude.
- Go crazy, the app is entirely yours, the only limit is your creativity, it just needs to be perfect for you, no one else.
The key takeaway is to be curious, to read, experience and have fun with this new tool. It is likely to spread across most office jobs and "knowledge" jobs within years, so you better start paying attention to who's leading the way, and whether their values hold under pressure.
Also, Anthropic just released some free classes, worth having a quick look.
March 26, 2026 — The App Improved, and I've Lost 20 Days of Journaling
Good news and bad news, let's start with the good, right?
I'm glad about the interest some of you have shown in building a similar app for yourselves. So please keep sending me updates on your vibe-coding experiences!
Since the last issue, I've integrated my current fitness plan and my sports watch directly into the app. This allows me to feed my training data directly into a Claude-powered training coach. It then has all the keys in hand to generate and adapt my plan based on my performances. I've also built a visual overview of the training plan calendar to help me plan my activities. Discussing with that "coach" about my recent runs without any manual entry and watching it adapt my upcoming sessions is probably the most satisfying feature to see come together.
I'll slightly adapt the adage "a picture is worth a thousand words" and tell you that a demo website is probably worth a couple hundred words. Yes, you've got that right! I've deployed a demo version of my Life Manager's current state, with obvious fake data (to avoid oversharing with you guys) that you should be able to access with this link. Quick disclaimer, I've taken out most of the actions possible like writing, editing, or calling the APIs, which means you might run into issues with some inputs.
Building that demo version was probably the biggest vibe-coding related learning experience so far: make sure to always keep an updated backup file somewhere, especially when you're using fake information to create a demo version of your app. Yeah, you guessed it, Claude overwrote my whole database with fake data while creating this demo version, right before I noticed that I did not have an up-to-date backup. I got Claude to write me an apology letter, so we're good now.
So yeah, even after a couple hours of fetching different files, changes in history and cache, I ended up losing a lot of data, and most importantly about 20 days of daily journaling (about 500 words every day). I have now created a script that creates a backup of the database every couple days, preventing this sort of error from happening in the future.
With a step back, that database disaster isn't really Claude being bad at coding, but the result of unclear instructions and lack of safeguard from my side. Vibe coding does lower the barrier to building, but it certainly does not lower the cost of not thinking. Mistakes still happen, and they happen faster and more thoroughly now.
May 13, 2026 — Go Slow Before You Go Even Slower
Go slow before you go even slower.
Traditional update on my Life Manager App:
I've done little to no work on the app, other than focusing on creating a proper Investing feature.
My current use of the app on a day-to-day basis could be rated as a 3/5. I'll report back when I've done some significant improvements.
June 13 2026 — Activity and Money, Centralized
Managing my activity and my money, good enough for now.
While trying to use my app as much as possible, I quickly thought about other softwares or services I was using outside of the Life Manager. So why not centralize everything?
As many of you have seen recently (my dear Strava followers), my training has been intensifying over the past weeks. My Life Manager helps me to stay up-to-date, with a consistent and flexible training program. From sleep score, to resting heartbeat and running data, it has everything needed to guide me towards my sports goal: a marathon in less than 4 months.
On the other side, I've implemented a thorough and complete investing hub feature, to help me in my investment journey. And also, it's preparing me for the next tax season, now that realized profits on financial assets will need to be appropriately tracked, declared and taxed in Belgium starting this year.
The last major change was creating a frontend to turn my newsletter into a proper blog: have a look here, it should be live!
Day-to-day usage rating of the Life Manager this month: 3.3/5 (last issue: 3/5).
