💊 Week #1: Digital Diet

📄 Overview

• Experiment Name: Digital Diet

• Week: #1

• Time Spent: ~2.5 hours

• Status: In-progress

• AI Tools Used:

  • ChatGPT

  • Claude

---

📝 Context

• Problem

  • I’m curious about many topics, but I don’t know what to focus on without getting overwhelmed with information.

  • When I try to keep up with the news, I either get stuck doomscrolling or feel down because of constant stories about murders, wars, etc.

  • I want to know what’s happening but also control how often and how much I consume.

  • There are new areas I want to explore (like global politics), but when I start reading, it feels like I’m jumping in mid-way without context.

  • I want to stay updated on topics that directly impact my work (E.g. AI and learning) and not just know what is happening but understand so what? (how does it affect me or my work?).

  • I’d like to connect different pieces of information so I can build unique insights instead of consuming isolated facts.

  • I want all of this to be easily accessible in one place, rather than scattered across bookmarks, newsletters, and notes.

• How I Solve it Right Now

  • I limit my structured information intake to newsletters and set aside 30 minutes each day to go through them.

  • In the past, I used to visit news websites directly but found that overwhelming.

  • When I find interesting posts or articles on LinkedIn or Twitter, I bookmark them and later move them into Notion, tagging them by source and topic (like AI, health, etc.) at the end of the week.

  • I’ve realized that making notes helps me actually remember things — otherwise, the information is quickly forgotten.

  • I prefer reading text over watching long videos, though visuals can be a helpful add-on. I don’t have the patience for very long articles or videos.

• Expected Output

  • A working system design for broad (L1) and deep (L2) layers of information intake.

  • Prototype daily summaries (Daily Doses) to check if this approach helps me stay updated without overwhelm.

  • First tests of an AI-powered tool that could eventually automate the workflow from ingestion → summary → insight generation.

---

🔧 Process

• 1/ Set the context of the problem, how I currently solve it, and the expected output

  • I described feeling overwhelmed by the amount of information I’m interested in and wanting a system that helps me balance breadth (knowing what’s happening) with depth (understanding why it matters).

  • I shared my existing workflow: limiting intake to newsletters → spending 30 minutes daily reading → bookmarking interesting things → tagging them in Notion weekly.

  • I clarified that I want to stay updated, form unique insights, connect ideas, and make everything accessible.

• 2/ Brainstormed with ChatGPT on what an AI-powered system could look like

  • We explored a two-layered approach:

    • L1 (Push): daily broad nudges across my interest buckets

    • L2 (Pull): deep dives into topics I choose, with context and relevance

  • I defined my focus areas as Work (AI, learning, business, etc.), Personal (politics, health, etc.), and General awareness (news, climate, tech shifts).

  • I shared that I want a daily Notion page as the main output, with potential future visualizations to see how ideas connect over time.

• 3/ Broke it down into what I thought was realistically possible to do in one week

  • I clarified that I have ~3 hours per week to dedicate to building this system.

  • I aligned with ChatGPT on designing each week as a mini-MVP sprint, where I get something tangible and usable by the end of every week.

  • Together we outlined a 5-week plan to build the system step by step.

• 4/ Aligned on a template & asked ChatGPT to produce an output

  • I decided that a Notion Daily Dose template would be my first MVP.

  • I planned to have ChatGPT help me generate a sample L1 entry manually to see how the daily push could look and feel.

• 5/ Mapped the full iterative build with ChatGPT

  • I created a detailed 5-week plan with a clear outcome for each week:

    • Week 1: Manual Daily Dose in Notion

    • Week 2: Deep dive workflow for selected topics

    • Week 3: Semi-automation of the daily feed

    • Week 4: Insight linking and weekly reflection

    • Week 5: Full system test of Vitamin AI in practice

• 6/ Started working on Week 1. Selected sources for each interest bucket and shared with ChatGPT

  • I listed preferred sources:

    • Work: The Rundown AI, Offbeat, Anne-Laure Le Cunff, Ross Stevenson

    • Personal: Huberman Lab, The Pump Club, The Oatmeal, Marcus Filly, Curiosity Chronicle, Dan Go, Justin Welsh, James Clear, Liz & Mollie

    • General: 1440, The Ken, Vox World

  • I mentioned I’d curate this list further but agreed it was enough to start testing in Week 1.

• 7/ Planned the Notion database structure for Week 1 output of the Digital Diet and shared with ChatGPT

  • I decided to use a linked database in Notion to enable filtering, tagging, and future insight connections.

  • We defined database properties:

    • Date

    • Work headlines

    • Personal headlines

    • General headlines

    • Tags

    • Notes to self

    • (Later) Deep Dive links

  • I shared my preferences: simple bullet points, clear buckets (health, learning, business, etc.), emojis only for section titles, and a clear information hierarchy.

• 8/ Attempted ChatGPT - Notion integration and failed

  I invited ChatGPT (as notion-invite@openai.com) to my Notion workspace with full access but learned that ChatGPT can’t actually create or edit Notion pages directly — outputs from chat can’t be pushed automatically into Notion.

• 9/ Generated an entry instead on ChatGPT only with manual pasting on Notion

  • I chose to proceed by having ChatGPT generate Daily Doses here in chat for me to paste into Notion myself, and plan to explore automation or integration later once I’m confident the manual workflow is valuable.

  • ChatGPT created a Daily Dose for June 30, 2025, structured into Learning & Work, Health & Habits, and World & General on the chat only

  • I realized the article links provided were placeholders, not live URLs, and asked for real articles to read.

The copy of my daily digest - a summary of all the things I’m interested in on ChatGPT. None of the links work :/

• 10/ Moved to Claude to see if it was better

  • Repeated step 1 to 3 again with Claude

• 11/ Designed the solution with Claude - an AI-powered Personal Intelligence Dashboard

  • Smart Content Processor for summaries, categorization, and "So What?" analysis of uploaded/pasted content.

  • Connection Engine to identify patterns and links between different pieces of content over time.

  • Insight Generator providing immediate actionable takeaways and broader strategic implications for work.

  • Personal Knowledge Graph for visual representation of topic connections and knowledge development.

  • Mood-Aware Filtering to balance information diet with content warnings and nutrition tracking.

• 12/ Identified an MVP - a Smart Content Processor as foundational starting point

  • Developed React web app (Claude Artefact) with clean interface for content input and intelligent analysis dashboard.

  • Integrated Claude AI for automated content processing, and insight generation.

  • Implemented local browser storage for privacy and search/filtering capabilities for processed content.

The quick and dirty artefact I made with Claude - you can copy paste text inside it and it processes it, gives a summary, adds tags and connects it to my context

• 13/ Realised I was trying to do too many things and I need to better define the REAL problem

  • I articulated my current workflow of how I process information

  • I identified key issues I face in the workflow

  • I tried to articulate different products/a solution stack for this

---

🪞 Learnings

✅ What Worked

• Starting with a brain dump of the problem, expected output, and my existing workflow helped clarify what I actually wanted from this experiment.

• Defining a roadmap for the 5 weeks right at the start was helpful — it broke the big problem into smaller, more manageable chunks.

• Experimenting with multiple tools (ChatGPT, Claude) and seeing how each approached the problem gave me insights into different workflows and possibilities.

• Time-boxing the experiment helped me pull together this synthesis, and in the process, I learned a lot about my needs and preferences.

❌ What Didn’t Work

• The problem itself was too broad and vague at the beginning — I realized it needed to be further defined (without AI) before jumping into solutions. I only did this after spending almost 3 hours feeling stuck.

• My first MVP attempt with ChatGPT was overly complex, and I wasted time trying to build something perfect instead of quickly testing a simple version.

• After the first failed ChatGPT integration, I lost motivation and felt like I hadn’t made enough progress, which slowed me down for the rest of the week.

---

🚀 Next Steps

• Use the insights from this week to better define the problem — start fresh with a clearer scope.

• Identify which parts of my workflow can realistically be automated with AI to optimize time and reduce manual effort.

• Distinguish the parts of the process where AI can unlock new capabilities (like summarizing or connecting ideas) that I simply can’t do alone, versus tasks that AI can only speed up.