# Vinod Sharma β€” Full Content > Vinod Sharma is a product builder based in Florida. He builds startups using AI coding tools, including Sucana, an AI analytics platform for marketing agencies. He shares build stories, AI reporting insights, and practical strategies for marketers through his newsletter Build Notes and YouTube channel. Website: https://vinodsharma.co Founders: Virgil Brewster (Co-Founder, Strategy & Marketing), Victor Chazarra (Co-Founder, Agency Operations), Vinod Sharma (Co-Founder, Development & Engineering) --- ## Why I Only Do Three Things a Day URL: https://vinodsharma.co/blog/why-i-only-do-three-things-a-day Published: 2026-07-29 Author: Virgil Brewster Tags: productivity, prioritization, deep-work, building I used to be a task master. Every morning I'd open my to-do list and add everything I was responsible for. Work tasks, personal tasks, startup tasks, content tasks. The list was long. Some days 15 to 20 items. I'd work all day. Cross things off. Add more things. Go to bed tired. And still feel like I hadn't done enough. The list never ended. And somewhere in the middle of all that busyness, I stopped asking the question that actually mattered: what is moving the needle? That's when two things happened at the same time. I discovered the 80/20 rule. 80 percent of your results come from 20 percent of your efforts. Most of what's on your to-do list produces almost nothing. And I came across a simple idea: you can only do 2 to 3 big things in a day. Not 10. Not 15. Two or three. At first it felt wrong. Only 3 things? I should be doing more. But I tried it. And by the end of the first week, I realized I'd done more meaningful work than in the previous month of busy days. Three focused things a day is 15 meaningful things a week. 60 a month. All of them the ones that actually moved something forward. The cascading structure The real unlock was connecting those 3 things to something bigger. Quarterly goals. Weekly priorities that align with the quarter. Daily top 3 that align with the week. This is how a single day connects to a week, a week to a quarter, and a quarter to the bigger vision. The interruption filter When something new comes in, I ask one question: is this more important than my top 3? Almost always the answer is no. It goes on tomorrow's list or gets delegated. No guilt. No context switching. No lost afternoon. How to build your own top 3 The night before, write everything down. Then ask: which 2 to 3 things would make the biggest difference tomorrow? Circle those. Put them on your calendar. Start with the first one in the morning. Not email. Not Slack. The first task. If you finish all 3, you've won the day. Everything else is a bonus. What are your top 3 for today? I write about building with AI and doing meaningful work at any stage of your career. Join the newsletter . --- ## I'm 53. Full-Time Job, Startup, Content, Family. Here's My Daily Schedule URL: https://vinodsharma.co/blog/my-full-week-daily-schedule Published: 2026-07-28 Author: Virgil Brewster Tags: routine, deep-work, productivity, building A lot of people ask me about my daily schedule. So here it is. Weekdays I wake up daily at 4am. Drink two glasses of water. Keep my phone away in silent mode, no tea yet. I open the first task I decided on the night before and set a 50-minute timer. No Slack. No meetings. The house is quiet. Everyone else is sleeping. Nobody needs anything yet. Outside is still dark through the window. This window belongs entirely to me. I work on my AI startup. Building agentic systems for AI-native businesses. Agents that think, route, and act without me holding their hand. The brain that remembers the whole business. AI automations that run while I sleep. A startup I'm growing with my partners in Bali. After the first 50-minute block, I make tea. That's when I check WhatsApp, usually a message from my partner or my virtual assistant Akash. I respond, then get back to work. Around 6am, I take my second break. I go for a walk in my community. During the walk, I check my messages and usually talk to Akash, my virtual assistant. And some days, just the walk. By 7:30am, I wind down, get ready for work and leave the house by 8:15. I try to reach the office by 9am. I lead three product engineering teams supporting 4M+ active users. My team and I are building iOS & Android apps, marketing portals and AI systems. I also teach team members how to use Codex & Claude in their coding and daily work. My day fills with delegation, project monitoring, demo calls, whiteboard sessions, and at least two focus blocks for coding. I leave work around 5:30 and most days attend a CrossFit class at my gym. By the time I reach home, I am tired and hungry. So I eat and help with housework. I take a walk with my wife or spend time with family. Around 8:30, one last working session. I make a list of the next day's priorities. I also write down what I accomplished that day. It's a quick 5-minute reflection. I write content, reply to comments and DMs. By the time I close the laptop, I decide exactly what I'm doing the next morning. So I don't have to spend precious time in the morning picking the most important task. Then I pick up a book to read for 15 to 30 minutes. I'm usually asleep before I finish the chapter. When I scroll, I go longer than I plan. When I read, I'm asleep within minutes. This has been my routine for every single day. Weekends are different Saturday morning is the same: 4am, water, deep work until 7:30. Then CrossFit from 8 to 8:45. After that, house cleaning, gardening, lunch with family. When everyone settles in for the afternoon, I go back to the computer. Two to four hours. In the evening, go out with my wife at some point. Family dinner, or maybe go out for ice cream. Sunday follows the same pattern. Morning deep work, then two to four hours in the afternoon. I also spend at least 30 minutes on weekly reflection and planning the next week. Weekends are my most productive building time of the week. I record videos, write content and build agentic systems for clients. Top 3 priority tasks for the day While I do a lot daily, I always pick my top 3 priority tasks for the day. I make a list of everything I need to do. Then circle the top 3 things that matter the most for that day. Those 3 things set the focus and intention for the day. If I finish those items, then I win the day; everything else is a bonus. The top 3 list also helps me with interruptions and conflicting priorities. I can compare them and see if they are more important than what I already planned. Almost always the answer is no. I stay on track without having to think hard about it. I'm 53. Full-time job. Startup. Content. Family. Gym. It all fits. But only because I follow a system. What is the window that belongs to you? If you found this useful, I write about building with AI, doing meaningful work, and growing at any stage of your career. Join the newsletter . --- ## The Second Brain Concepts Worth Knowing in 2026 URL: https://vinodsharma.co/blog/second-brain-concepts Published: 2026-07-07 Author: Virgil Brewster Tags: second-brain, pkm, obsidian, workflow, building If you have started building a second brain, you have already hit the wall. Capturing notes is easy now. Knowing how to organize them is not. Open your own vault after a few months and the question is always the same. Where does this even go. The good news is that smart people have spent decades solving exactly this. There is a whole landscape of organizing concepts out there, and most people have only heard of one of them. This post is an introduction to the six worth knowing in 2026. What each one is, who created it, and the specific problem it solves. The six concepts at a glance Concept Creator The core idea (in their words) PARA Tiago Forte "A simple, comprehensive, yet extremely flexible system for organizing any type of digital information across any platform." Johnny.Decimal Whitson and Harris A numbered system where "everything has a logical, decimally defined place." Saving is easy, finding is hard. MOCs / LYT Nick Milo Higher-order notes that help you "gather, develop, and navigate" your ideas. Zettelkasten Niklas Luhmann "A personal tool for thinking and writing. It has hypertextual features to make a web of thoughts possible." GTD David Allen "A personal productivity methodology that redefines how you approach your life and work." DiΓ‘taxis Daniele Procida "A systematic approach to technical documentation authoring." Now let me introduce each one properly. PARA: sort by how actionable it is PARA is the four-bucket system from Tiago Forte's book Building a Second Brain . Forte calls it "a simple, comprehensive, yet extremely flexible system for organizing any type of digital information across any platform." It gives you exactly four top-level folders, and everything you keep lives in one of them. πŸ“ Projects β†’ short-term efforts with a goal and an end πŸ“ Areas β†’ ongoing responsibilities that need attention over time πŸ“ Resources β†’ topics and interests you want to reference later πŸ“ Archives β†’ anything from the three above that is no longer active Here is what Forte means by each, in his own framing. Projects are "short-term efforts that you take on with a certain goal in mind," like renovating the bathroom or shipping a webpage. Areas are "important parts of your work and life that require ongoing attention," like Health, Finances, or Marketing. Resources are "topics you're interested in and learning about," like coffee or web design. Archives are "anything from the previous three categories that is no longer active, but you might want to save." Each of the four folders holds a small number of subfolders, one per active project, area, resource, or archive in your life. The clever part is that it sorts by action, not by subject. You do not file a note under "marketing." You file it under whether you are working on it right now. That is what keeps it usable when life gets busy, and it is why Forte designed it to be "actionable," meaning it makes work easier to start and finish rather than harder. If you are new to all of this, start here. PARA is the floor everything else builds on. Johnny.Decimal: give everything one address Johnny.Decimal is the answer to folder sprawl. Its whole pitch is one line: saving is easy, finding is hard. It even opens with the same stat that started my interest in this, the finding that knowledge workers spend 1.8 hours every day just searching for information. It fixes that by giving every single thing a number, so "everything has a logical, decimally defined place." The structure has three tiers. 10-19 Life administration β”œβ”€ 11 Health β”‚ β”œβ”€ 11.01 Insurance β”‚ └─ 11.02 Medical records └─ 13 Finance β”œβ”€ 13.01 Tax returns └─ 13.42 Investments 20-29 your next area (Home, Money, and so on) The three levels each have a job. Areas are the broad zones of your life, kept deliberately vague so "your first decision is always easy." Categories "do the heavy lifting," and because there are so few of them, "you'll naturally remember their numbers." IDs label "a specific thing across your entire system," and there is only ever one 13.42. Read a single number left to right and it tells you exactly where a thing lives. 13.42 Investments β”‚ β”‚ └── Title: a human name for the item β”‚ └────── ID: the 42nd item in this category └───────── Category: 13 = Finance, inside the 10-19 area Then there is the index , a single list of every number and what it means. It becomes the map of your entire system, the one page you check when you are not sure where something lives. The cap is the real genius. You are allowed no more than 10 areas, and no more than 10 categories inside each. That hard limit stops you from making a new folder every time you feel lazy. You are forced to decide where something belongs. Think of it this way. PARA tells you what a thing is. Johnny.Decimal tells you where it lives. They pair up cleanly, PARA as the categories and Johnny.Decimal as the numbering. MOCs and LYT: navigate by links, not folders A Map of Content, or MOC, comes from Nick Milo's Linking Your Thinking . Milo describes MOCs as higher-order notes that help you "gather, develop, and navigate" your ideas. It is the natural fit for anyone using Obsidian. Instead of digging through nested folders, you make an index note. That note links out to everything related. To get around, you open the map and follow the links, the way you would use a table of contents instead of a filing cabinet. Milo also recommends a single home note as the front door to your whole vault. "If you want a healthy and complete network of knowledge, you will want a home note." For folders, LYT offers a light structure called ACCESS , six folders meant to be torn apart and made your own. πŸ“ Atlas β†’ your maps of content and home note (where you navigate) πŸ“ Calendar β†’ daily and periodic notes, journal entries (time-based) πŸ“ Cards β†’ your atomic notes, ideas written in your own words πŸ“ Extras β†’ templates, attachments, media, the everything-else πŸ“ Sources β†’ reference material from others: books, articles, clippings πŸ“ Spaces β†’ dedicated workspaces for active projects and big efforts The insight underneath all of it is worth sitting with. Folders answer "where did I put it." Links answer "what is this related to." The second question is the one that actually grows your thinking. LYT calls this note-making rather than note-taking, building living notes that compound in value over time. If you already write in [[wikilinks]] , you are halfway to this without knowing it. Zettelkasten: a web of thoughts for thinking and writing Zettelkasten is the oldest concept here, and the deepest. The German word means "slip box." Its own definition is precise: "a personal tool for thinking and writing" with "hypertextual features to make a web of thoughts possible." Niklas Luhmann was a sociologist who kept tens of thousands of small paper notes, each holding a single idea, each linked to others by hand. He published 50 books and over 600 articles, and said plainly that his productivity came "from working in a partnership with his Zettelkasten." The unit is a single Zettel , one atomic thought, written in your own words, never copied verbatim. Each one has three parts. [ID] a unique identifier, so the note can be linked and found [Body] exactly one idea or concept, in your own words [Refs] links to related notes, each with a note on WHY they connect The IDs are what make it a web. Luhmann used a branching scheme like 1 , 1a , 1b , 1a1 , so a new thought could slot in right next to its parent. Many modern tools use a timestamp instead, like 202006110955 . Either way, notes connect through explicit links, and the important detail is the link context , a short note on why two ideas belong together. That reasoning is where new knowledge gets made. The popular on-ramp, from SΓΆnke Ahrens' book How to Take Smart Notes , sorts your writing into three note types as a workflow. Fleeting notes β†’ quick captures, thrown away after you process them Literature notes β†’ what you took from a source, in your own words Permanent notes β†’ one finished atomic idea, linked into the web, kept A word of caution. This one is for your thinking layer, your research and writing. It is overkill for your electric bill and your meeting notes. Not everything needs to be a permanent note. GTD: the task system the others borrow from Getting Things Done, by David Allen, is "a personal productivity methodology that redefines how you approach your life and work." It is the classic that most second brain systems quietly stand on. Its core line is the entire philosophy. "Your mind is for having ideas, not holding them." You pull everything out of your head into a trusted system, so your brain can think instead of remember. The method runs on five steps. Capture. "Collect what has your attention." Everything, into one inbox. Clarify. "Process what it means." Is it actionable? What is the next action? Organize. "Put it where it belongs." Sort it into the right list. Reflect. "Review frequently." Keep the system trusted and current. Engage. "Simply do." Act with confidence from your lists. The organizing step is where the structure lives. GTD keeps a handful of lists. Next actions β†’ the single next physical step, grouped by context (@calls, @computer) Projects β†’ anything that needs more than one action to finish Waiting for β†’ things you are blocked on, owned by someone else Calendar β†’ things that must happen on a specific day or time Someday / maybe β†’ not now, but you do not want to forget it Here is the surprise. PARA's Projects and Areas came straight out of GTD. So if you adopt a second brain, you are already standing in David Allen's house. GTD is the piece to reach for when your knowledge is organized but your tasks are still floating in your head. DiΓ‘taxis: the one for docs and how-tos DiΓ‘taxis is the outlier here, and the one almost nobody in the second brain world mentions. It calls itself "a systematic approach to technical documentation authoring," and it solves three problems at once: what to write, how to write it, and how to organize it. It says every document is one of four things, and you should never mix two of them in one place. Practice-oriented Theory-oriented Learning (study) Tutorials Explanation Working (action) How-to guides Reference Each quadrant has a clear job. A tutorial teaches a beginner by taking them through doing something. A how-to guide gives the steps to solve one specific real-world task. Reference is the dry, factual description you look things up in. Explanation is the background and the why. The two axes behind the grid are simple: theory versus practice, and acquisition (learning) versus application (doing). This will not organize your whole life. What it will do is make every guide, checklist, and standard operating procedure you write dramatically clearer, because each one finally knows what job it is doing. Which system is built for whom These six are not really competitors. Each was built by someone solving their own specific problem, and their own words tell you who it is for. PARA was built for anyone, on any tool. Forte designed it to organize "any type of digital information across any platform," which is why it is the usual first step for beginners. Johnny.Decimal was built for "the life admin we all need to manage," and for teams and organizations where files pile up and finding is the bottleneck. Reach for it when your problem is sprawl. LYT and MOCs were built for note-makers, the people who want to develop ideas and think in an Obsidian-style linked vault, not just store files. Zettelkasten was built by an academic for writing. It suits researchers, writers, and anyone whose output is long-form and idea-heavy. GTD was built for personal productivity, for individuals drowning in tasks and commitments, with extensions for teams. DiΓ‘taxis was built for documentation authors and maintainers, for anyone who writes things other people will later have to read and follow. Read that list as a menu, not a competition. Most people end up borrowing from more than one, because a task system, a filing system, and a thinking system are genuinely different jobs. The part these concepts do not solve There is one honest thing to say before you go and adopt any of them. Every concept here organizes. None of them maintains. A framework will not move your finished project to the archive. It will not file the note you took at 2am. A perfect structure that you have to hand-feed every single day slowly turns into a museum, beautiful and untouched. That is the gap I keep writing about. The structure decides where knowledge goes. Something still has to actually put it there and keep it current. For me that something is a set of AI agents built with Claude Code and the Hermes agent, doing the filing while I sleep. I wrote about that in At 52, my second brain finally works . The concept and the automation are two halves of the same tool. Start with the concepts, though. They are the vocabulary. Once you can name what each one is for, you stop guessing and start choosing. So here is my question for you. Of these six, which one have you already tried, and did it stick, or did it quietly become a museum? --- ## My Setup URL: https://vinodsharma.co/blog/my-setup Published: 2026-07-05 Author: Virgil Brewster Tags: setup The gear on my desk. What I use to build every day. Companion to my toolbox . Affiliate disclosure: Some of the links below are Amazon affiliate links. If you buy through them, I earn a small commission at no extra cost to you. I only link to things I actually use. Devices & Accessories Mac Mini M4 Apple Magic Mouse iPhone 14 Pro Max Apple Magic Keyboard Apple Magic Trackpad Logitech G502 HERO Gaming Mouse EVGA Z15 RGB Mechanical Keyboard MacBook Pro 16-inch, M5 Pro, 64GB RAM VIVO Premium Aluminum Heavy Duty Monitor Arm LG UltraGear 45-inch Curved Ultrawide Monitor Content Creation Elgato Stream Deck MK.2 Logitech Brio 4K Webcam Blue Yeti USB Microphone Blue Compass Broadcast Boom Arm Desk Monoprice Electric Dual Motor 3-Stage Sit-Stand Desk Frame V2 --- ## My Toolbox URL: https://vinodsharma.co/blog/my-toolbox Published: 2026-07-05 Author: Virgil Brewster Tags: toolbox Here is the tech stack and apps I use daily. Also, check out my setup . Coding & Development VS Code : My main code editor. Free, fast, and endlessly extensible. Claude Code : Anthropic's agentic coding tool. I hand it real tasks from the terminal and it writes, edits, and runs code. Codex : OpenAI's coding agent. I run it alongside Claude Code for hands-off work. GitHub : Where all my code lives, and the home for my open-source projects. Docker : Runs my projects in clean, reproducible containers, so "works on my machine" actually means something. Wispr Flow : Voice dictation that keeps up. I talk, it types, in any app. ElevenLabs : AI voices for narration and audio. Strikingly natural text to speech. Google Chrome : My main browser. Cursor GitHub Copilot Languages & Frameworks React Python Next.js TypeScript Tailwind CSS Markdown / MDX Backend, Data & Infra Vercel Supabase Cloudflare Writing & Notes Notion : One of the most potent apps around. It does almost anything: notes, tasks, project management, even digital products. Obsidian : Notes stored as plain text files on your own computer. A strong, privacy-minded vision, and your notes stay yours even if the app ever goes away. Workflowy : A lightweight outliner I love. Like Roam without the connections, with an ultra-fast search. MindNode : Mind mapping for when I need to think something through visually before I write it down. Grammarly : Catches my typos and tightens my writing before anything goes out. Whiteboards Figjam Canva : Quick graphics, thumbnails, and social images without opening a heavier tool. TLDraw : A dead-simple infinite whiteboard for sketching ideas fast. Miro : The bigger whiteboard, for mapping systems and planning with other people. Content Creation Descript : Edit video and audio by editing the transcript. It strips out filler words and makes recording far less painful. CapCut : Fast video editing, especially for short-form clips. Snagit : Screenshots and screen recordings with quick annotation. Better than the built-in tools for clean how-to shots. Teleprompter.app : Scrolls my script while I record, so I can look at the camera and still hit every point. Elgato Stream Deck : A grid of physical buttons I map to shortcuts, scenes, and actions I run all day. VLC : Plays any video or audio file you throw at it. Nothing else needed. Gamma : Turns an outline into a clean presentation in minutes. My go-to for decks. Scribe Analytics, SEO, Search Consoles Google Analytics Microsoft Clarity Google Tag Manager PageSpeed Insights Google Cloud Console Google Search Console Google Keyword Planner Data Scraping & Research Apify Firecrawl Newsletter & Email Kit Beehiiv Omnisend Meetings & Messages Slack : Team chat and the hub for most of my work conversations. Telegram : Fast, lightweight messaging. Discord : Communities and group chat, especially around building and creators. WhatsApp : Everyday messaging with family, friends, and some work contacts. Zoom : Video calls and client meetings. Google Meet : The other video call tool, one click from a calendar invite. Fathom : Joins my calls, records them, and writes the notes and summary so I can pay attention instead of typing. Fireflies Task Manager & Productivity OmniFocus : Serious task management on the GTD model. Where everything I need to do gets captured. Basecamp : Calmer project management for keeping work organized without the noise. Calendly : Finding a meeting time without the back-and-forth emails. Flow : A simple Pomodoro timer. The free version is plenty, and the paid one is only $12. Files & Utilities Apple iCloud : Keeps my files, photos, and settings in sync across all my Apple devices. Google Drive : Shared documents and the files I work on with other people. DaisyDisk : Shows what is eating your disk space as a clear visual map, so you can clear it fast. AppCleaner : Uninstall an app and every file it left behind. Drag the app onto AppCleaner and it finds the scattered leftovers for you. Free. Windscribe : VPN for privacy and getting around region limits. 1Password : Stores every password and passkey, so I never reuse or forget one. Splashtop Business : Remote into my other machines from anywhere, at full speed. Apple Home : Controls the smart devices around my house from the Mac and phone. Reading & Daily Habits Kindle : Reading books on any device, synced to wherever I left off. Audible : Audiobooks for commutes and walks. Five Minute Journal : A two-minute daily journaling habit, morning and night. --- ## The Claude Code Skill I Use to Put Any Repo Live on Its Own Domain URL: https://vinodsharma.co/blog/ship-a-site-to-vercel-with-a-custom-domain Published: 2026-07-05 Author: Virgil Brewster Tags: vercel, cloudflare, claude-code, deployment, dns, workflow, agentic-systems I build a lot of small sites. Demos, landing pages, side projects, little starters I want to share. The building is the fun part. The last mile is the part I always dreaded. Because the last mile is always the same two chores: get it deployed, then give it a real domain instead of some random URL. I did those two steps by hand a hundred times. Now I don't. I turned the whole thing into one Claude Code skill, and I want to show you what it does, because the pieces are simple once you see them laid out. The two halves of "going live" Shipping a site is really two jobs. Deploy. Take the code on my laptop and run it on the internet. Domain. Put it on an address a human would actually type, with a padlock. Most of the pain is that these live in two different tools that do not know about each other. The code goes to Vercel. The domain lives on Cloudflare. You end up copy-pasting records between two dashboards and waiting, unsure if you did it right. The skill just does both, in order, from the terminal. Half one: deploy to Vercel From inside the project folder, the whole deploy is one command. vercel --prod --yes Vercel looks at the folder, sees it is a Next.js app, and builds it in the cloud. About thirty seconds later it hands back a live URL. Two things happen here that are easy to miss and worth calling out. First, because the folder already had a GitHub repo, Vercel quietly connected the two. From that moment on, every git push deploys itself. I did not touch a dashboard to set that up. It came free with the first deploy. Second, the URL it gives you is a bit ugly. Mine came out as portfolio-kit-kohl.vercel.app . That random -kohl on the end is Vercel avoiding a name clash. Which is the perfect reason to move on to half two. Half two: point a real domain at it I wanted portfolio-demo.careerleap.app , not the vercel.app URL. Step one, tell Vercel the domain belongs to this project and ask what record it needs: vercel domains add portfolio-demo.careerleap.app portfolio-kit vercel domains inspect portfolio-demo.careerleap.app Vercel replies with the exact DNS record to create. For me it was a single A record pointing the subdomain at Vercel's address. Step two is the part that used to mean a second dashboard. My domains sit on Cloudflare. So the record has to be added there. This is where the Cloudflare skill earns its place. Claude talks to Cloudflare directly through an MCP connection, which is just a secure pipe to Cloudflare's API. I do not open the dashboard. I say what record I want, Claude shows me the exact thing it is about to create, I approve it, and it writes it: Type: A Name: portfolio-demo Value: Vercel's IP Proxy: off (this one matters, more below) Step three, verify and wait a beat for the padlock: vercel domains verify portfolio-demo.careerleap.app The domain started resolving instantly. The https took about a minute while Vercel issued the certificate. Then it was live: https://portfolio-demo.careerleap.app , real domain, valid SSL, no dashboard opened. The mistakes that cost me time The same two things trip everyone up here. They tripped me up first. 1. Leave the Cloudflare proxy off. Cloudflare loves to route traffic through itself (the orange cloud). For a domain pointing at Vercel, that fight over who handles SSL turns into redirect loops and broken certificates. Grey cloud. DNS only. Let Vercel own the domain. 2. A dead site right after setup is normal. The minute after you verify, the address resolves but the page will not load. It looks broken. It is not. Vercel is still minting the SSL certificate, which takes a minute or two. Check again shortly and it turns green. Do not go re-editing DNS in a panic. I did, the first time. Why make it a skill I could keep these commands in a notes file. I used to. But a notes file is something I have to find, read, and run myself. A skill is something my AI assistant already knows. I say "ship this and put it on this subdomain," and it runs the deploy, asks Vercel for the record, shows me the DNS change for a yes, writes it, and waits for the padlock. Start to finish, a few minutes, and I am reviewing decisions instead of remembering commands. That is the real shift with agentic systems. The knowledge stops living in my head or a stale doc, and starts living in something that can act on it. The skill is not a script. It is the routine, written down once, so I never have to hold it in memory again. Build the thing. Say "ship it." Go share the link. What is the one chore in your workflow that you keep doing by hand when it could be a skill? --- ## The 4 Analytics Tools I Add to Every Website I Build URL: https://vinodsharma.co/blog/analytics-i-add-to-every-website Published: 2026-07-04 Author: Virgil Brewster Tags: analytics, gtm, google-analytics, clarity, seo, nextjs, tools I build a lot of websites. Personal sites, landing pages, client sites, side projects such as careerleap.app, switchtomac.app, and the ones I build for clients. Every single one gets the same four analytics tools. I used to skip this on the small ones. "I'll add analytics later." Later never came. Then a page took off and I had zero data on where the people came from or what they did. So now it goes on day one. The four questions Analytics help you answer four questions. How many people came, and to what? That is traffic. What did they actually do on the page? That is behavior. How are people finding me in Google? That is search. And how do I manage all of it without redeploying my site every time? That is the container. Four questions, four tools. The stack Google Tag Manager (GTM). The container. This is the one that loads and manages every other tag. Nothing else goes in the code directly. GTM is the hub. Google Analytics 4 (GA4). Traffic and events. How many people, which pages, where they came from, what they clicked. The "how many and what" tool. Microsoft Clarity. Free session recordings and heatmaps. GA4 tells you 40% dropped off the pricing page. Clarity lets you watch ten of them do it. The "why" tool. And it is free, with no seat limits, which still surprises people. Google Search Console. How you show up in Google search. Impressions, clicks, the actual queries people typed, and whether your pages are even indexed. GA4 shows you the people who already arrived. Search Console shows you the ones deciding whether to. I keep Vercel Analytics on too, since most of my sites are on Vercel and it is one line. But those four are the real stack. The one idea that makes this easy: GTM is the hub Here is the move that turns this from a chore into a twenty-minute habit. You put one snippet in your code: the GTM container. That is it. GA4 and Clarity do not go in your code. They go inside GTM, as tags, configured in the GTM dashboard. Don't hardcode 5 scripts on every site. Use GTM as the single hub: Your site β†’ ONE GTM snippet β†’ GTM fires: β”œβ”€β”€ GA4 tag β”œβ”€β”€ Hotjar tag β”œβ”€β”€ Amplitude tag └── (anything future) Why does that matter? Because the day you want to add a new tool, change a tracking ID, or fix a tag, you do it in GTM and click publish. No code change. No deploy. No waiting for a build. This is not theory. Last week I set up GA4 on a site, then had to recreate the property and got a brand new measurement ID. On a normal setup that is a code edit and a redeploy. On this setup it was one field in GTM and a publish. Sixty seconds. The site code never knew the GA4 ID existed. It only knows the container. One snippet in code. Everything else in the hub. How I wire it in Next.js The GTM snippet is the only thing that touches the codebase. I keep the ID in an environment variable so the same code runs on every site. // layout.tsx const GTM_ID = process.env.NEXT_PUBLIC_GTM_ID; // in <head> {GTM_ID && ( <Script id="gtm" strategy="afterInteractive"> {`(function(w,d,s,l,i){ /* standard GTM snippet */ })(window,document,'script','dataLayer','${GTM_ID}');`} </Script> )} Then NEXT_PUBLIC_GTM_ID=GTM-XXXXXXX in .env.local and in the host (Vercel, for me). The GTM ID is public, it ships in the page anyway, so it is not a secret. One warning from experience: if you copy a codebase to start a new site, check that GTM ID. I once had a fresh site quietly reporting into an old project's container for weeks because the hardcoded ID came along for the ride. Env variable per site, and this stops happening. Then, inside GTM, add two tags: GA4: a Google Tag with your G-XXXXXXXXXX measurement ID, trigger "All Pages." Clarity: the easiest path is to open Clarity, go to Setup, and pick "Google Tag Manager." Clarity installs its own tag into your container for you. No snippet to paste. Publish the container. Done. Search Console is separate and needs no code. Add your domain, verify it, and submit your sitemap.xml. On careerleap.app the DNS is on Cloudflare, so it verified in one click. Then submit your sitemap.xml . The two mistakes that cost me time Every time I teach someone this, the same two things trip them up. They tripped me up too. 1. "Google Tag" vs "GA4 Event." When you add GA4 in GTM, pick the tag type called Google Tag . There is another one called "Google Analytics: GA4 Event." That one is for tracking specific actions later, and it forces you to name an event. If you pick it by mistake, you get a red error and no pageviews. You want Google Tag. It sends pageviews automatically. That is all you need to start. 2. Preview is not published. GTM has a Preview mode that shows your tags firing. It feels like it is working. It is only working for you. Real visitors get nothing until you click Submit, then Publish . I have watched people set everything up perfectly, check Preview, see green, walk away, and wonder a week later why their reports are empty. Publish the container. Do it on day one That is the whole thing. Four tools. One snippet in the code. Everything else managed in a dashboard you can change without a deploy. Traffic, behavior, search, all wired in about twenty minutes. The best time to add analytics is before you have any traffic. The second best time is now, before the page you forgot about suddenly has some. Put it on day one. Future you, staring at a spike with no idea where it came from, will thank you. What is the first thing you wire into every new site? --- ## Pageviews Don't Show Your Funnel. A 10-Line Helper Does. URL: https://vinodsharma.co/blog/track-your-funnel-with-one-datalayer-helper Published: 2026-07-04 Author: Virgil Brewster Tags: analytics, gtm, google-analytics, funnel, nextjs, tools Last week I wired up analytics on a site. Google Tag Manager, GA4, Clarity, Search Console. Then I opened GA4 and realized it was telling me almost nothing useful. It counted visits. Which pages. Where people came from. But my site has a tool with a real funnel. Import a resume, see a score, save it, enhance it, export it. GA4 showed me 200 people hit that tool. It could not tell me how many actually imported. Or saved. Or hit export. Pageviews are the top of the funnel. I wanted the whole thing. Here is how I got it, with about ten lines of code. The gap pageviews leave GA4's base tag fires a page_view on every page. That is the free part, and it is genuinely useful. But the actions that matter usually are not pages. Clicking a button is not a page. Submitting a form is not a page. An import finishing, an export downloading, an upgrade, these all happen without the URL changing. Those are the steps where people drop off. And a pageview-only setup is blind to every one of them. The fix: push events to the dataLayer The dataLayer is just an array that Tag Manager watches. You push objects onto it, GTM reacts. So the whole trick is a tiny function that pushes a consistent shape: // lib/track.ts export function track(name: string, params: Record<string, unknown> = {}) { if (typeof window === "undefined") return; window.dataLayer = window.dataLayer || []; window.dataLayer.push({ event: "cl_event", cl_action: name, ...params }); } That is it. Ten lines, no library. Every event carries the same wrapper: event: "cl_event" is the key GTM listens for, cl_action is the name of what happened, and the rest are flat params. It is safe to call anywhere. On the server it does nothing. Before GTM loads it does nothing. So I never have to guard it. Then I sprinkle track() at the moments that matter: track("kit_card_click", { kit: "agentic-brain-kit" }); track("newsletter_success", { source: "home" }); track("resume_import_start", { method: "upload" }); track("resume_export", { format: "pdf" }); Now the funnel is defined in the code, in plain sight, versioned with everything else. Not buried in a dashboard. The part people overthink: one tag, not fifty Here is where most funnel setups go wrong. People create one GTM tag per event. Twenty events, twenty tags. It becomes a mess nobody wants to touch. You do not need that. You need one. In Tag Manager: One trigger. A Custom Event trigger that fires on cl_event . That catches every event my helper pushes. One tag. A GA4 Event tag where the Event Name is the variable {{cl_action}} , and the parameters map to dataLayer variables ( kit , method , format , and so on). That single tag forwards everything. kit_card_click , resume_export , whatever I push next month. GA4 receives each one with its real name and its params. The best part: when I add a new track() call in the code later, I change nothing in GTM. It just starts flowing. The tag already forwards anything with the cl_event key. One helper in code. One tag in GTM. The whole funnel. A thing that will confuse you in Preview When you test this in GTM's Preview mode, load a page and you will see your GA4 event tag sitting under "Tags not fired." That is correct. Do not fix it. The tag fires on an action, not a page load. So on a plain pageview, only your pageview tag and any session tools fire. The moment you click the button or submit the form, a cl_event fires and your funnel tag jumps to "Tags fired." I stared at "not fired" for a minute before it clicked. Trigger an action, watch it move. Mark the ones that matter In GA4, flag your money events as Key events: the opt-in success, the sign-up, the export, the purchase. Now GA4 will build funnel and conversion reports around them, and you can finally see the drop-off. Two hundred started, forty imported, twelve saved, five exported. That gap between forty and twelve is where you go to work. Why this beats click-tracking in GTM You can also track clicks entirely inside GTM, with Click triggers that match CSS selectors. No code. I used to do that. I stopped. Selectors drift. You rename a class, restyle a button, and a trigger silently breaks. Weeks later you notice a funnel step went to zero, not because people stopped, but because the tracking did. An event in the code sits right next to the action it measures. When I move the button, I see the track() call. It comes along. It does not rot. The whole thing One track() helper. Calls at the steps that matter. One GTM tag that forwards them all to GA4. A few events marked as conversions. An afternoon of work, and the black box turns into a funnel you can actually read. Pageviews tell you people showed up. This tells you what they did once they were there. That second question is the one worth answering. --- ## Four Ways to Read Markdown Files in VS Code Without Clicking 'Open Preview' Every Time URL: https://vinodsharma.co/blog/four-ways-to-read-markdown-in-vs-code Published: 2026-06-30 Author: Virgil Brewster Tags: vscode, obsidian, markdown, mdx, workflow, tools My AI system has hundreds of markdown files. Meeting notes, project pages, voice guides, positioning docs. All plain .md or .mdx . VS Code displays them like a text file. I open a file. I see raw markdown. Backticks, dashes, asterisks. But most of the time, I want to see the rendered version, with the tables laid out and the headings sized properly. Like how I would see them in Obsidian. I have to right-click, find "Open Preview," click it, and VS Code opens the rendered file next to the source. It works but also feels tedious. Here are the four ways to fix it. Method 1. Use these two keyboard shortcuts Cmd+Shift+V opens the rendered preview in the same tab. Cmd+K then V opens the preview side by side with the raw file. You can edit on the left and see the rendered version update as you type. Method 2a. The permanent flip: make VS Code always open markdown as preview If you want VS Code to render every markdown file automatically on open, you can force it with a setting. Open the command palette with Cmd+Shift+P . Type "Preferences: Open User Settings (JSON)" and hit enter. Paste this at the top of the object: "workbench.editorAssociations": { "*.md": "vscode.markdown.preview.editor" } Save. Now every .md file opens as the rendered preview by default. The trade-off is when you want to edit the file, you have to switch back to source mode. This solution works for people who read markdown far more than they write it. But some people genuinely prefer raw markdown. You can see the structure, you feel the file. If that's you, keep option 1. Method 2b. If you also use MDX, use these settings instead The section above works for plain .md files. If you also use .mdx (which is what Next.js, Astro, and Docusaurus blogs use), use the following two settings. Both work together to let you preview MD and MDX automatically in VS Code. Open settings.json (Cmd+Shift+P β†’ "Preferences: Open User Settings (JSON)") and paste this expanded version instead: "workbench.editorAssociations": { "*.md": "vscode.markdown.preview.editor", "*.mdx": "vscode.markdown.preview.editor" }, "files.associations": { "*.mdx": "markdown" } One limitation is that VS Code's built-in preview doesn't understand JSX components. If your MDX has <Component /> tags, they show up as raw text in the preview. For blog posts that are 95% plain markdown, the preview looks correct. For component-heavy MDX, install an extension like "MDX Preview" from the VS Code marketplace, or run your local dev server ( pnpm dev in a Next.js project) to see the real render with your components in place. Method 3. Install Foam and turn VS Code into a real markdown workspace The two options above are about VS Code's built-in preview. Foam is different. Foam is a VS Code extension that turns the editor into a knowledge workspace, something closer to Obsidian. Search "Foam" in the VS Code extensions marketplace. Install the one from the Foam team (currently at 258k+ downloads, 43 five-star ratings). You'll get: Wikilinks. Write [[my-note]] and it becomes a real link to that file. Autocomplete works. Backlinks. For any file you open, Foam shows you every other file that links to it. Graph view. See your entire knowledge base as a network diagram. Every file is a node, every link is an edge. Daily notes. One command creates or opens today's note in a dated folder. Tag browser. Every #tag becomes searchable. If you already have markdown files in a folder and they link to each other, Foam just works on top of them. No migration. Point it at your folder and it starts building the graph. I use Foam when I want to stay inside VS Code but need Obsidian-like navigation. Which brings us to the last option. Method 4. Just use Obsidian For anyone who spends most of their day in markdown, Obsidian is the best option. Once you have hundreds of files with links between them, you need to see the graph. You need backlinks. You need wiki-style navigation. Obsidian is free. It reads plain .md files. It can also read .mdx files after you install the free "mdx as md" plugin. It opens any folder as a "vault." You can open the same folder in Obsidian and keep VS Code and Obsidian side by side. One for coding, one for reading markdown. My entire brain folder works as an Obsidian vault without changing a single file. Same repo. Same files. Different view. The split I actually run: VS Code when I'm editing code, running scripts, or using the Claude extension to ingest new information. Obsidian when I'm reading, browsing, following links between notes, or looking at the graph to see how ideas connect. Both apps hit the same files. Whatever I edit in one shows up in the other the next time I open it. TLDR The whole point is that markdown was designed to be readable both ways: as source and as rendered text. Your editor should let you switch between those views without friction. Pick the method that works for you. Use one of the two shortcuts. Cmd+Shift+V opens the rendered preview in the same tab. Cmd+K then V opens the preview side by side with the raw file. Force preview on open with the settings snippet. Open settings.json (Cmd+Shift+P β†’ "Preferences: Open User Settings (JSON)") and paste this expanded version instead: "workbench.editorAssociations": { "*.md": "vscode.markdown.preview.editor", "*.mdx": "vscode.markdown.preview.editor" }, "files.associations": { "*.mdx": "markdown" } Install Foam if you want to stay in VS Code but want Obsidian-like features. Use Obsidian if you want the best reading experience. Open the entire project folder in Obsidian. I run VS Code and Obsidian side by side. Both pointing at the same folder. It gives me the best of both worlds. --- ## For Years, Everything We Discussed in Meetings Disappeared. Then I Built an Agentic System. URL: https://vinodsharma.co/blog/most-meetings-disappear Published: 2026-05-28 Author: Virgil Brewster Tags: agentic-systems, building, sucana The call ends. Someone says "good meeting." Everyone closes the tab. And every decision, every task, every owner, every deadline goes back to floating in someone's head. That used to happen to me every day. I would sit on a Zoom for 45 minutes with Virgil, my co-founder. We would talk through a real problem. We would land on a real decision. We would assign work. Then we would hang up. The next morning, half of it was already gone. The good ones I remembered. The boring ones I forgot. The decisions I was sure I would write down later, I did not write down later. The conversation was good. The capture was bad. For a long time I told myself the answer was discipline. Take better notes. Set a reminder. Write the action items at the end of the call. Be a more organized person. That never worked. Because the problem was never me being lazy. The problem was that there is too much information moving through a business every day for any human to capture by hand. I am a Senior Manager at AdventHealth by day. I am building Sucana at 4am every morning. I have three kids and a full-time job. I am not going to spend an hour after every meeting writing meticulous notes. I would rather sleep. So I built it instead. What I built The pipeline is simple to describe. Hard to live without once you have it. A meeting ends. The transcript shows up in a raw folder a few minutes later. An agent reads the transcript. The agent decides what matters and what does not. The agent writes the action items, the decisions, the open questions, the owners, the deadlines, the milestones, the blockers, the next moves. Then it updates the brain. When I say "the brain" I mean the markdown files that hold everything I need to remember about the business. The wiki entries on each project. The task list. The status of each workstream. The decisions log. The list of who owns what. The places we are blocked. A meeting is no longer a conversation that disappears when the call ends. It becomes an input to the brain. What changed in my week Mondays used to start with the question "what did we decide on Friday?" Now Mondays start with the answer. Wednesdays used to have a moment around 3pm where Virgil would Slack me with "did we say we were going to do X or Y for the launch?" and I would scroll back through Fireflies trying to find the line. Now I open the wiki and the answer is already there. The amount of work I used to do after a meeting to capture the meeting was bigger than the meeting itself. That work is now done by the agent in the background. Often while I am asleep. The thing that surprised me I thought the value would be the time saved. It is not. The value is that the business stops being dependent on memory. When the system is doing the capturing, it does not matter who attended the call. It does not matter who took notes. It does not matter who was distracted because their kid walked in. The transcript is the source. The agent is the reader. The brain is the destination. Nothing relies on a human remembering. That changes the texture of how we work. There is less "what did we say again?" There is less "who was supposed to do that?" There is less context stuck in one person's head. There is more continuity. More accountability. More follow-through. The business gets quieter. Not because we are talking less. Because the loose threads stop fraying. Why this is the first automation I would build If you are running an agency or a startup right now, you already have this situation. The calls are happening. The decisions are being made. The tasks are being discussed. The knowledge is already there. It is just trapped inside transcripts, recordings, chat threads, and the heads of the people who happened to be in the room. You do not have a knowledge problem. You have a routing problem. AI agents make it possible to take that information, structure it, and keep the systems updated without anyone touching it. It runs in the background. It does not ask for permission. It does not need a process change to adopt. It just makes sure the important things you already discussed do not disappear after the call ends. That alone earns its keep. The part nobody tells you You will not feel the impact in the first week. You will set it up. It will start producing notes. You will read them once, nod, and forget the system exists. The impact shows up around week three. The week you needed to remember something from a meeting four meetings ago and you actually could. The week your co-founder asks "what's the status of X?" and instead of pulling up Slack and Fireflies and Drive, you open one file and read the answer in 20 seconds. The week you realize you have not done a "let me catch you up on everything" meeting in a while. Because nobody needed to be caught up. That is the moment the system pays you back. What it actually replaces It does not replace the meeting. The meeting still has to happen. People still have to talk through hard problems together. It replaces the part after the meeting. The part where you used to depend on one person being organized enough to write the notes, send the email, update the project, follow up with the owners. That part is now a system. The meeting becomes the input. The brain becomes the output. The human work in between is gone. If you are building any kind of business right now, that is the trade I would make first. The calls are already happening. The knowledge is already there. The only question is whether it disappears when you hang up. For me, the answer used to be yes. Now it is no. That is the upgrade. --- ## At 52, My Second Brain Finally Works URL: https://vinodsharma.co/blog/second-brain-at-52 Published: 2026-05-25 Author: Virgil Brewster Tags: agentic-systems, building, sucana, journey I spent years trying to build a second brain. A few years ago I joined Tiago Forte's Build a Second Brain cohort. I loved the framework. Capture. Organize. Distill. Express. Four words, one clean loop. I built several versions for myself. I tried Notion. I tried Obsidian. I tried plain markdown in folders. I read the book twice. None of them stuck. The problem was never the framework. The framework is good. The problem was the maintenance. A second brain only works if you actually keep it fed. And keeping it fed means capturing notes from calls, voice memos, rough ideas, docs, campaign data, tasks, client updates, Slack threads, emails, market signals. Every day. Forever. That is too much for one person to organize properly. I tried. I would do it for two weeks. Then I would miss a day. Then a week. Then I would open the vault three months later and realize I had been building a museum, not a tool. The framework was not asking too much. Life was. Over the last two months, that changed. What is different now I built five AI operating systems using Claude Code and Hermes-Agent. They are not chatbots. They are not Notion templates. They are systems that actually run in the background and do the part I could never sustain. The five so far: Virgil Brain. The operating system for Virgil, my co-founder, and the work he does with clients. Team Brain. The shared knowledge layer for everyone we work with. Sucana Brain. The technical and product brain for what I am building at 4am. Agency Brain. The operations layer for the agency side of the business. Personal Brain. Mine. My ideas, my projects, my voice notes from the drive to Costco. Each one has the same job. Take messy human inputs and turn them into a clean, navigable record. Each one has its own scope and its own owners. None of them depends on me remembering anything. The newest one is the one I keep thinking about It is an Operations Management System. The OMS. It pulls from everywhere. Rough notes I write in the middle of the night. Meeting transcripts from Fireflies. Slack conversations across multiple workspaces. Telegram voice and text updates. Campaign data from Meta Ads and Google Ads. Newsletter data from Kit and Omnisend. Market trend signals from X, Reddit, Google. All of that goes in. What comes out is one reliable view. The wiki for every project. The tasks for every workstream. The daily pulse showing what moved. The milestones. The next actions. The project status. The knowledge base that captures the why behind every decision. It is available on demand. I can ask it anything. I can also let it push daily updates through the channels we already use. Telegram in the morning. Slack at noon. An email summary at night if I want one. The OMS does not replace the team. It replaces the part of the team that nobody wanted to do anyway. The collecting. The transcribing. The status updates. The "wait, what did we say about that?" loop. That part is now a system. Why it works now and did not before I want to be clear about something. I did not get more disciplined. I am 52. I have three kids. I have a full-time job as a Senior Manager at AdventHealth. I am building Sucana at 4am every morning. There is no version of me with more hours in the day or more focus to spare. The reason the second brain works now is that the system does the boring parts. The collecting. The organizing. The distilling. The cross-linking. The "where did we put that" search. The "what did Virgil decide on Tuesday" recall. All of it. I do the thinking. I do the deciding. I do the building. The system does the rest. That was the missing piece in 2021 when I first tried this. The framework was right. The tools were not enough. You still had to be the human in the middle, every day, pulling threads from one tool and tying them into another. That is the part that breaks. Claude Code and Hermes-Agent close that gap. What this looks like in a normal day I wake up at 4am. I sit down with coffee. I open Telegram. There is a morning brief waiting. What happened yesterday across the business. What is on my plate today. Anything that flagged itself as needing attention. I do not open six tools to construct that picture. The OMS already did. At 7am I leave for the day job. While I am at AdventHealth, the systems keep running. New transcripts come in. They get extracted. Decisions get logged. Tasks get assigned. The brain updates itself. At 9pm I come back to the desk for an hour. I read what the systems wrote. I correct what they got wrong. I add what they could not have known. The next morning's brief is already getting better. That is the loop. The brain feeds itself. I feed the brain. Neither of us has to do it alone. What I would tell someone starting from here If you are building a business right now and your knowledge, tasks, meetings, and updates are scattered across too many tools, you already have the inputs. You have meetings happening, Slack threads piling up, a CRM somewhere, a Google Drive with 4,000 files in it. The information is there. You do not have a tool problem. You have a connection problem. The missing piece is a system that quietly connects the messy inputs, works alongside your team, and stays up to date without anyone touching it. That is the unlock. Once you have it, the second brain idea finally makes sense. Until you have it, the framework is just a clean diagram on top of a messy life. This is the first time in my life it has felt sustainable. Not because I became a different person. Because the boring parts are no longer mine to carry. If you are experimenting with AI agents, second brains, or a personal OS, I would love to compare notes. The patterns I am finding are not in any book yet. The tools are too new for any book to be written. What I can tell you is what is currently working at 4am. --- ## I Disappeared From Social Media for 3 Months. I Came Back a Different Builder. URL: https://vinodsharma.co/blog/disappeared-for-3-months Published: 2026-04-15 Author: Virgil Brewster Tags: agency, building, sucana, agentic-systems I went quiet for three months. No posts. No comments. No engagement. I let my LinkedIn sit. I stopped opening X. I closed the loops that had been training me to think in headlines and posts. I did not announce it. I did not write a "going dark" post. I just stopped showing up. For three months, the entire output of my day went into one thing. Building a startup with two co-founders. The two people I disappeared with Virgil, my co-founder, is one of the best marketers I have ever worked with. He thinks in audiences, hooks, sequences, offers. He can sit on a call with a founder for 30 minutes and surface the exact angle the founder has been circling but cannot name. I watched it happen 20 times. Victor, my other co-founder, runs a 7-figure ad agency. He has been spending other people's money on Meta, Google, and TikTok for years. He reads a campaign dashboard the way a doctor reads an X-ray. The whole room is looking at numbers. He is looking at the leak. Both of them have spent more time inside the messy, emotional, fast-moving day of an agency than I have spent inside any one thing in my life. I went quiet so I could sit next to them while they worked. Not to give advice. Not to architect anything. Just to watch and ask questions until I understood the shape of the business they live in. Three months. Heads down. No audience. What I learned that I did not know existed Some of this I could not have written three months ago. I had read about it. I had heard people on podcasts say it. But until you sit inside the work, you do not actually know it. You just have words. How agencies actually work. The pitch deck version is "we run ads for clients." The real version is a constantly shifting balance between client expectations, platform changes, account access, creative production, reporting, retention, and the emotional state of one founder at a time. Every client is a different business. Every business is a different shape of trust. How different every client really is. Two clients in the same niche with the same ad budget can need completely different strategies because their unit economics are different, their offer is different, their sales process is different, and the way they talk to their audience is different. There is no template. There is only learning the specific business in front of you, fast. How emotional this business really is. Clients are not buying ads. They are buying belief in their own business at a moment when belief is shaky. Wins are celebrated. Losses are felt personally. The agency that survives is the one that handles both with the same steady hand. How to read campaign data and know in 30 seconds what is working and what is bleeding money. I used to look at a campaign dashboard and see numbers. Now I see a story. Where the spend is going. Where the conversion is breaking. Which audience is carrying the result and which one is dragging it down. Where the leak is. That skill is not in any course. You learn it by sitting next to someone who has spent years doing it on real money. How leads leak between the click and the sale. A click is not a customer. There are 6 places in between where you can lose them. The ad copy. The landing page. The form. The follow-up. The objection handling. The close. Most agencies optimize one of these and miss the other five. You only see all six when you live inside the funnel for months. How ROAS, CPL, and conversion rates tell a story most people miss. These are not metrics. They are sentences in a longer paragraph. ROAS without CPL is meaningless. CPL without quality of lead is misleading. Conversion rate without time-to-close is a trap. The story is always in the relationships between the numbers, not the numbers themselves. How to make your content visible to ChatGPT, Claude, and Perplexity, not just Google. Search did not stop at Google. It moved into AI chat. The way your site is indexed, the schema you publish, the citations you build, the way an AI model decides to mention your name in an answer. This is a new layer. Most businesses are not playing in it yet. The ones who start now will have a real lead. How skills and agents are making 3-person teams operate like 30. Claude Code with skills. Hermes-Agent running background work. Custom agents that handle the boring parts of an operation. None of us are working 100-hour weeks. We just removed the friction from the parts where 30 people used to be needed. How OpenClaw and Hermes are shaping the future of AI coding. This one is the most exciting to me as a coder. The agent layer is moving fast. The tools that win the next 12 months are the ones that compose well, run reliably, and let a small team ship like a big one. What three months of silence actually felt like It felt boring in the first week. I kept reaching for my phone to check what I had missed. I had not missed anything. The first week is the hardest because your brain is still running the old reward loop. Open the app. Get the dopamine. Repeat. By week three the reach was gone. I stopped wondering what was happening on X. I stopped imagining a post that was not getting written. By week six the silence had texture. I was thinking deeper about problems. I was sitting longer with things before reaching for a tool or a tweet. The internal monologue got quieter, then sharper. By week ten I was not the same builder I had been when I started. I learned things I did not know existed. From building next to people who do it every day. That is the trade. You give up the feed. You get back depth. What I would tell anyone thinking about doing this If you have been thinking about going deep on something, this is the part nobody tells you. You do not need a sabbatical. You do not need to quit your job. You do not need to announce anything. Three months and one room is enough. Turn off the feed. Stop consuming. Start building. The version of you that comes back will not recognize the version that left. What would you build if nobody were watching? --- ## How I Added AI Search Visibility to My Personal Site Using Claude Code URL: https://vinodsharma.co/blog/how-i-added-ai-search-visibility-to-my-personal-site Published: 2026-03-22 Author: Virgil Brewster Tags: geo, building, agentic-systems I rebuilt my personal site last week. New design, new stack, new everything. It looked great. Dark mode worked. The newsletter signup was hooked up. But when I ran a GEO audit on it, I realized something. AI search engines like ChatGPT, Claude, Gemini, and Perplexity had no idea I existed. No robots.txt. No structured data. No llms.txt. No sitemap. If someone asked an AI "who is Vinod Sharma" or "product builders using AI coding tools," my site would never show up in the answer. This is the story of how I fixed that in one build session using Claude Code. What is GEO? GEO stands for Generative Engine Optimization. It is like SEO, but for AI search engines instead of Google. When someone asks ChatGPT a question, it pulls information from websites it has crawled. If your site is not structured in a way that AI crawlers can read and understand, you are invisible. There are 10 things that matter for AI search visibility: Content Structure (25%) - Clear headings, lists, FAQ sections Content Depth (20%) - Statistics, quotes, 1500+ words Technical Setup (15%) - Server-rendered HTML, clean URLs AI Crawler Access (10%) - robots.txt rules, llms.txt file Structured Data (10%) - JSON-LD schema markup Trust Signals (8%) - Author info, credentials, social links Meta Tags (5%) - Title, description, OG images Navigation (3%) - Sitemap, breadcrumbs, internal links Geographic (2%) - Location data for local queries Voice Assistant (2%) - Speakable content, FAQ schema I scored about 15 out of 100 before I started. Here is what I did to get to 78. The Before: What Was Missing My old site (vinodsharma.ai) was a simple portfolio built with Next.js and Tailwind. It had my name, my startups, some testimonials, and a contact section. What it did NOT have: No robots.txt file. AI crawlers had no explicit permission to index the site. No llms.txt file. No way for AI models to understand who I am or what I do. No JSON-LD structured data. No Person schema, no FAQPage schema, nothing machine-readable. No sitemap. Search engines had to guess which pages existed. No OG images or Twitter cards configured properly. No canonical URLs set. No security headers. No RSS feed. The content was there. The infrastructure to make it discoverable was completely missing. Stage 1: Start with a Strong Foundation I did not build the SEO infrastructure from scratch. I had already built it for sucana.ai (the company I am building with my co-founders Virgil and Victor). That site had everything: robots.txt, sitemap, llms.txt, JSON-LD schemas, RSS feed, security headers, IndexNow integration. So I cloned the sucana.ai codebase and used it as my starting point. In Claude Code, I said: "I made a copy of www.sucana.ai at vinodsharma.ai-new. Help me strip the Sucana content and rebrand it for my personal site." This gave me all the SEO/GEO infrastructure for free. Dynamic sitemap, robots.txt with AI crawler rules, llms.txt, RSS feed, security headers, blog system with MDX support. All working out of the box. Why this matters: Building SEO infrastructure from scratch takes days. Starting from a working codebase that already has it saves you all that time. Stage 2: Add robots.txt with AI Crawler Rules The robots.txt file tells crawlers what they can and cannot access on your site. Most personal sites either do not have one (which means default allow) or have a basic one that does not mention AI crawlers at all. My new robots.txt explicitly allows every major AI crawler by name: User-agent: GPTBot Allow: / User-agent: ClaudeBot Allow: / User-agent: PerplexityBot Allow: / User-agent: Google-Extended Allow: / User-agent: Amazonbot Allow: / User-agent: CCBot Allow: / There are 12 AI crawlers listed in total, including OAI-SearchBot, Claude-SearchBot, ChatGPT-User, Perplexity-User, and Applebot-Extended. The sitemap URL is also referenced at the bottom: Sitemap: https://vinodsharma.ai/sitemap.xml Why this matters: Explicitly naming AI crawlers tells them "yes, you are welcome here." Some sites block AI crawlers by default. Naming them removes any ambiguity. Stage 3: Add Person JSON-LD Schema JSON-LD is structured data that helps search engines and AI models understand what your page is about. For a personal site, the most important schema is Person . I added this to my homepage: { "@context": "https://schema.org", "@type": "Person", "name": "Vinod Sharma", "url": "https://vinodsharma.ai", "image": "https://vinodsharma.ai/images/profile.webp", "jobTitle": "Product Builder", "description": "Product builder who shipped 9+ products using AI coding tools. 26+ years in technology.", "knowsAbout": [ "AI coding", "product development", "Next.js", "TypeScript", "Tailwind CSS", "startup building", "micro-SaaS", "Claude Code", "Vercel" ], "alumniOf": [ { "@type": "CollegeOrUniversity", "name": "Webster University", "location": "St Louis, MO" }, { "@type": "CollegeOrUniversity", "name": "North Maharashtra University" } ], "sameAs": [ "https://x.com/VinodSharma10x", "https://linkedin.com/in/vinodsharma10x", "https://www.youtube.com/@vinod.sharma", "https://vinodsharma.substack.com/" ] } This tells AI models: here is a person named Vinod Sharma, he is a Product Builder, he knows about these 9 topics, he went to these universities, and here are his social profiles. Why this matters: When someone asks an AI "who builds micro-SaaS products with Claude Code," the knowsAbout field makes it more likely your name comes up. Stage 4: Add FAQPage Schema I already had a FAQ section on my homepage with 6 questions and answers. But without FAQPage schema markup, AI models treat it as regular text. Adding the schema wraps each question and answer in a structured format: { "@context": "https://schema.org", "@type": "FAQPage", "mainEntity": [ { "@type": "Question", "name": "How do you build startups in 2 hours a day?", "acceptedAnswer": { "@type": "Answer", "text": "The key is extreme focus and leveraging AI tools..." } } ] } Research shows that pages with FAQ sections are 2x more likely to be cited by AI models. The Q&A format matches how people ask questions to AI assistants. Why this matters: FAQ content is one of the highest-impact things you can add for AI visibility. It directly matches the question-answer format that AI search uses. Stage 5: Add llms.txt The llms.txt file is a relatively new standard. It is a plain text file at the root of your site that tells AI models who you are and what your site is about. Think of it as a cover letter for AI crawlers. Here is what mine looks like (abbreviated): # Vinod Sharma > Vinod Sharma is a product builder based in Florida who builds > startups using AI coding tools. He came back to coding after > 12 years in management and has shipped 9+ products while > working full-time. ## What I Build - Sucana β€” AI-powered analytics platform for marketing agencies - Part Time Founders β€” Community for aspiring entrepreneurs - GEOScore β€” AI search visibility analyzer (open source) ## Expertise - AI-assisted product development (Claude Code, Cursor, V0, Bolt) - Next.js, TypeScript, Tailwind CSS, Vercel ## Newsletter Build Notes β€” I am building Sucana and sharing everything I learn through my newsletter and YouTube channel. ## Contact - Twitter/X: https://x.com/VinodSharma10x - LinkedIn: https://linkedin.com/in/vinodsharma10x - YouTube: https://www.youtube.com/@vinod.sharma I also added a <link rel="llms" href="/llms.txt" /> tag in the HTML head so crawlers can discover it. Why this matters: While llms.txt has no proven citation impact yet (it is too new), it costs nothing to add and gives AI models a clean summary of who you are. It is also linked in the HTML head using <link rel="llms"> so it is easy to find. Stage 6: Add a Dynamic Sitemap A sitemap tells search engines which pages exist on your site and how important they are. My sitemap is generated dynamically in Next.js: export default function sitemap(): MetadataRoute.Sitemap { return [ { url: SITE_URL, priority: 1.0, changeFrequency: "weekly" }, { url: `${SITE_URL}/blog`, priority: 0.8, changeFrequency: "daily" }, { url: `${SITE_URL}/about`, priority: 0.7, changeFrequency: "monthly" }, { url: `${SITE_URL}/resources`, priority: 0.7, changeFrequency: "monthly" }, ...blogEntries, ]; } Every time I add a blog post, it automatically appears in the sitemap. The sitemap URL is referenced in robots.txt so crawlers find it immediately. I also set up an IndexNow integration that pings Bing daily with all my URLs, so new content gets indexed faster. Why this matters: Without a sitemap, crawlers have to discover pages by following links. A sitemap gives them the complete map upfront. Stage 7: Complete the Meta Tags I added full Open Graph and Twitter Card metadata to every page: og:title , og:description , og:image (1230x630) twitter:card: summary_large_image twitter:creator: @VinodSharma10x Canonical URLs on every page Proper robots directives These do not directly affect AI search, but they improve how your site appears when shared on social media and in traditional search results. They also signal to crawlers that the site is well-maintained. Stage 8: Add Security Headers This is not directly about AI visibility, but it signals quality. I added: Content Security Policy (CSP) Strict-Transport-Security (HSTS) X-Frame-Options X-Content-Type-Options Referrer-Policy Permissions-Policy Well-maintained sites with proper security headers tend to rank better across all search engines, including AI-powered ones. The After: GEO Score 78/100 After implementing all these changes, I ran a GEO audit using GEOScore (an open-source tool I built at geoscore.sucana.ai ). Here are the results: Category Score Weight Content Structure 8/10 25% Content Depth 6/10 20% Technical Discoverability 9/10 15% AI Crawler Access 10/10 10% Structured Data 8/10 10% E-E-A-T Signals 8/10 8% Meta Tags & OG 9/10 5% Navigation 6/10 3% Geographic 4/10 2% Voice & Assistant 7/10 2% Overall 78/100 AI Crawler Access scored a perfect 10. All 12 major AI crawlers are explicitly allowed. llms.txt exists. Sitemap is referenced. The weakest area is Content Depth at 6/10. The homepage has about 1,200 words, which is below the 1,500 word target. Adding blog posts (like this one) will push that score higher. What I Would Do Next To push from 78 to 85+, here is what is still on my list: Add blog posts. The blog system is built and ready, but it had zero posts until now. Content is the single biggest driver of AI citations. Add statistics as visible text. Numbers like "9+ products shipped" and "26+ years in tech" exist in my schema but not as visible text on the homepage. Adding them would give AI models more to cite. Add BreadcrumbList schema. A simple addition that helps AI models understand site hierarchy. Add case studies to the sitemap. I have two case studies (Nintex and Trend Micro) that are not yet in the sitemap. Enrich product descriptions. Each product card has one sentence. Adding metrics and details would make them more citable. The Stack Here is everything I used for this build: Claude Code for all coding and implementation Next.js 16 (App Router) for the site framework Tailwind CSS 4 for styling GEOScore ( geoscore.sucana.ai ) for the before/after audit Vercel for hosting and deployment ConvertKit (Kit) for newsletter integration The Bottom Line Adding AI search visibility to a personal site is not hard. The infrastructure (robots.txt, schema, llms.txt, sitemap) takes a few hours to set up. The hard part is the content. You need depth, statistics, and structured information that AI models want to cite. If you want to check your own site's AI visibility, try GEOScore . It is free, open source, and gives you a score out of 100 with copy-paste fixes. --- ## One Inch Closer: The Debugging Philosophy That Keeps Us Shipping URL: https://vinodsharma.co/blog/one-inch-closer Published: 2026-02-20 Author: Virgil Brewster Tags: mindset, building, sucana My cofounder Virgil Brewster was watching me debug for three hours. Nothing was working. Even my AI coding agents could not figure it out. Screens were blank. Errors were stacking up. The kind of session where most people close the laptop and go for a walk. But I was smiling. Virgil looked at me and said, "It's not working. And you're still excited?" I told him: every time I fix one thing, I see one more screen load. I find one more reason why it broke. One inch closer. That tiny movement is enough to keep me going for hours. How "One Inch Closer" Became Our Daily Mantra The next morning, Virgil was debugging Beehiiv integration for our newsletter system. Every few minutes, he would say, "Vinod, one inch closer." He had caught the bug. Not the software bug. The mindset bug. It has been over two weeks now. We say it to each other every day while building Sucana. Victor, our third cofounder, picked it up too. When a Meta Ads API call fails for the fifth time, when a CSP header blocks Google Tag Manager, when a proxy rotation does not work on a bot-protected site. One inch closer. The phrase works because it reframes what "progress" means. You are not trying to solve the whole problem. You are trying to find the next blocker and remove it. That is it. Why This Works for Builders Most people quit during the debugging phase because they are measuring progress wrong. They look at the gap between where they are and where they want to be. That gap is huge. It feels like nothing is happening. But if you zoom in, something is always happening. Every error message is information. Every failed attempt eliminates one possibility. Every "that didn't work" brings you closer to the thing that will. Here are three real examples from building Sucana in the last month: The proxy problem. We built GEOScore, a tool that analyzes websites for AI search visibility. It worked great on most sites. Then we tried AdventHealth and got a 403 error. Their Akamai Bot Manager blocked everything. We tried browser-like headers. Failed. Tried Playwright. Failed. Tried Webshare rotating proxies. That worked for most sites, but not AdventHealth. Each attempt was one inch closer to understanding which bot protection systems we could handle and which ones we could not. That clarity was the real win. The ConvertKit key. When I rebuilt vinodsharma.ai, I added ConvertKit for the newsletter. The API kept returning "access token invalid." I tried V3 API with api_key parameter. Failed. Tried V4 with Bearer token. Failed. Tried V4 with X-Kit-Api-Key header. Worked. Three attempts, each one narrowing the problem. The fix took five minutes once I found the right header. The debugging took an hour. But each failed attempt removed one wrong answer. The CSP headers. Victor reported that Google Tag Manager was not firing on sucana.ai. The tag assistant showed zero tags found. The Content Security Policy was blocking GTM from fetching its container config because googletagmanager.com was in script-src but not in connect-src . One missing domain in one header. Finding it required checking the CSP, understanding how GTM loads, and testing each domain. One inch at a time. The Practice Next time you are stuck on something, try this. Do not try to find the whole solution. Just find the next blocker. Remove it. Expect five more behind it. That is normal. The trick is to stop measuring progress by how far you are from the finish line. Measure it by how many blockers you have removed since you started. That number only goes up. Some days you will remove ten blockers and ship a feature. Some days you will remove two and still not see a working screen. Both are progress. Both are one inch closer. What Are You One Inch Away From? I would love to hear what you are debugging right now. What is the blocker sitting in front of you? Sometimes just naming it out loud makes it smaller. Find me on Twitter/X or LinkedIn and tell me what you are one inch away from today. --- ## Gems in a Jar: Why You Should Stop Waiting for the Perfect Idea URL: https://vinodsharma.co/blog/gems-in-a-jar Published: 2026-02-12 Author: Virgil Brewster Tags: mindset, journey, building I have a theory about startup ideas. They are like gems in a jar. The perfect idea is somewhere near the bottom. You can see it glinting through the glass. But you cannot reach it. Not without picking up the gems sitting on top of it first. The problem is that most people stare at the jar forever. They hold each gem up to the light, turn it around, look for flaws, put it back. They read blog posts about gem selection. They join communities about gem theory. They wait for someone to invent a tool that pulls the bottom gem straight out. Meanwhile, the person next to them just grabs one from the top and starts walking. How I Found My Idea by Not Looking for It In 2025, I built CareerLeap. A career development platform for programmers. It was a fine idea. I believed in it. I coached people through it. But nobody was willing to pay enough to make it a real business. The problem was not painful enough. That gem was not the one. But picking it up revealed something underneath. While building CareerLeap, I met Virgil Brewster. He had 20 years of experience launching products and had scaled a business to $10M. We started talking because of CareerLeap. Not because of the idea itself, but because building it put me in a position where the right person noticed. Then I built Part Time Founders. A community for people building startups on the side. Great conversations. Good engagement. But communities are hard to monetize. That gem was not the one either. But building it taught me something. I did not want to run a community. I wanted to build products. That clarity only came from trying the community path and feeling the mismatch. Then GoalSetting. A planning framework and workbook. People downloaded it, used it once, and disappeared. Turns out, goal setting is a moment, not a product. Another gem that was not the one. But it taught me that tools people use once have no retention, and retention is everything. Then Jivro with Virgil and Victor Chazarra (who I met through Virgil). A product idea discovery platform. Interesting concept. Nobody needed it urgently enough to come back. But building together for the first time taught us how we work as a team. Who handles what. How we communicate when things break. How we make decisions. Then Scancook. A recipe scanning app. Quick build, quick pivot. But another week of working together. Another layer of trust. Then Victor said something that changed the trajectory. He showed us what his PPC agency actually needed. He was spending two full days every week pulling numbers from Meta, Google, and spreadsheets just to build client reports. He needed a dashboard that unified everything and let him talk to the data in plain English. That was Sucana. I did not find Sucana by searching for the perfect idea. I found it by picking up six gems that were not the right ones. Each one revealed something I could not have seen from the outside. A cofounder. A skill. A lesson about retention. A real problem worth solving. The 200 Meter Rule There is another way to think about this. Imagine driving at night. Your headlights show about 200 meters of road. You cannot see the destination. You cannot see the turns ahead. You can only see the stretch right in front of you. But that is enough. By the time you drive those 200 meters, the next 200 appear. You do not need to see the whole road. You just need to trust the movement. MVPs work the same way. You build a small version of an idea. Your first 200 meters. Then real users give you feedback. They tell you what they actually need, which is almost never what you assumed. That feedback is your next 200 meters. You cannot plan the whole journey from the parking lot. The map only draws itself while you are moving. Every product I built was 200 meters. CareerLeap was 200 meters toward understanding my audience. Part Time Founders was 200 meters toward understanding my own motivation. Jivro was 200 meters toward finding a team. Sucana was the stretch of road where everything finally clicked. The Cost of Waiting Here is what I see most people do. They spend months researching ideas. They read about market sizing, competitive analysis, and TAM calculations. They build spreadsheets comparing opportunities. They wait for certainty before they start. The problem with this approach is that the information you need does not exist yet. You cannot learn whether an idea works by thinking about it. You can only learn by building it and putting it in front of real people. The best market research I ever did was shipping CareerLeap and watching nobody pay for it. That taught me more about product-market fit in two months than any amount of reading could have. The cost of picking the wrong gem is low. You lose a few weeks of building time. You gain skills, connections, and clarity. The cost of not picking any gem is high. You lose months or years of standing still, waiting for certainty that never comes. What This Looks Like in Practice If you are sitting on three or four ideas right now, here is what I would do. Pick the one you can build fastest. Not the biggest one. Not the most exciting one. The one you can ship in two to four weeks. Speed matters because the point is not the first idea. The point is what the first idea reveals. Ship it to real people. Not your friends. Not a focus group. Put it where strangers can find it and use it. Watch what happens. Do they come back? Do they tell other people? Do they get angry when it breaks? Those signals tell you everything. Pay attention to what you learn, not just what you earn. My first four products made almost no money. But they introduced me to my cofounders, taught me what kind of work I actually enjoy, and showed me which problems are worth solving. That was worth more than revenue. When the signal is not there, move on. Do not force it. Do not pivot twelve times trying to make a dead idea work. If people are not coming back after two months, pick the next gem from the jar. The lessons from this one are already in your pocket. Trust that one thing leads to another. It sounds like vague advice. It is not. It is the most concrete pattern I have seen in my own life and in every founder I have talked to. The path only makes sense looking backward. Pick a Gem You are not going to find the perfect idea by thinking harder. You are going to find it by building something imperfect and paying attention to what happens next. The jar is sitting right in front of you. Pick one from the top. It probably will not be the one. That is fine. The one underneath it might be. And if it is not, the one underneath that might be. Keep picking. Keep building. The perfect gem is down there somewhere, but you will never reach it by staring through the glass. --- ## I Built 9 Products in 10 Months. Each One Led to the Next. URL: https://vinodsharma.co/blog/9-products-zero-traction-one-company Published: 2026-01-18 Author: Virgil Brewster Tags: building, journey, sucana In 2025, I started coding again after 12 years in management. I had spent over a decade leading teams, running Agile transformations, and building enterprise systems for 100,000+ employees. Good work. Important work. But I missed building things with my own hands. AI coding tools changed everything. Claude Code, Cursor, V0, Bolt. Suddenly I could ship products again without needing a team of developers. I could go from idea to working app in days, not months. So I started building. A lot. The Products That Did Not Work CareerLeap. A career development platform for programmers. I built it, launched it, coached people through it. It worked as a service but not as a product. The audience was too broad and the problem was not painful enough to pay for. GoalSetting. A goal setting framework and workbook. People liked the concept. Nobody came back after the first week. Setting goals is exciting. Following through on them is not a product problem, it is a habit problem. PartTimeFounders. A community for people building startups while working full-time. Great conversations. Good engagement. But communities are hard to monetize and even harder to scale. SaaS Founders Hub. A resource directory for SaaS founders. I built it in a weekend. It got some traffic from search. But directories are commodities. There was no moat, no reason someone would choose mine over a hundred others. None of these got real traction. No recurring revenue. No product-market fit. No signal that said "keep going, this is the one." If I had measured 2025 by the standard startup metrics, it was a failure. What Actually Happened But something else was happening that I did not see at the time. While building CareerLeap, I met Virgil Brewster. He had scaled his previous business to $10M over 20 years of launching products. We started talking. He saw something in how I approached building. Through Virgil, I met Victor Chazarra, who runs a PPC agency from Bali with over $15M in managed client ad spend. Together, we built Jivro (a product idea discovery platform) and Scancook (a recipe scanning app). Neither stuck. But working together did. We learned how we each think, what we are good at, and how we handle it when things break. Then Victor showed us what his agency actually needed. A way to stop spending two days every week pulling numbers from Meta, Google, and spreadsheets. A dashboard that unifies all the ad data and lets you talk to it in plain English. We pivoted to Sucana and stayed focused. That is what we are building now. None of this would have happened if I had waited for the perfect idea. The Gems in the Jar I think about ideas like gems in a jar. You cannot pick the perfect one sitting at the bottom without first picking one from the top. When you pick one from the top, the next two reveal themselves. CareerLeap revealed that I should help founders, not employees. PartTimeFounders revealed that I wanted to build products, not communities. Meeting Virgil revealed that I needed cofounders, not solo projects. Building with Virgil and Victor revealed the actual problem worth solving. One thing led to another. That is exactly how it happened. The 200 Meter Rule Think about driving at night. You can only see about 200 meters ahead. But by the time you drive those 200 meters, the next 200 appear. You do not wait for the whole road to light up. You trust the movement. That is how MVPs work too. You start with a small idea. Your first 200 meters. Then real users and real feedback give you clarity for the next 200. You cannot plan the whole journey from the parking lot. Every product I built in 2025 was 200 meters. None of them was the destination. But each one lit up the next stretch of road. What I Would Tell Someone Starting Out Stop waiting for the perfect idea. Seriously. The perfect idea does not exist before you start. It only becomes visible after you have built two or three imperfect ones. Here is what I would do if I was starting over: Pick an idea. Any idea that solves a problem you personally understand. It does not need to be original. It does not need to be big. It needs to be something you can build in a few weeks. Ship something small. Not a landing page. Not a waitlist. An actual working thing that someone can use. Even if it is ugly. Even if it only does one thing. Ship it. Pay attention to what happens next. Who uses it? Who does not? What do they ask for? What problem are they actually trying to solve? The answers to these questions are your next 200 meters. Listen for the signals. Traction is obvious when it shows up. People come back without being asked. They tell other people. They get annoyed when something breaks because they depend on it. If none of that happens, it is not the right idea. Move on. Keep building. The skills compound even when the products do not. Every app I built in 2025 made me faster at building the next one. By the time we started Sucana, I could ship features in hours that would have taken me weeks on CareerLeap. Where I Am Now Sucana is the first product that feels different. Victor uses it every day to run his actual agency. The feedback is specific and urgent, not polite and theoretical. We are shipping features every week and onboarding early agency users. I do not know if Sucana will be "the one." But I know that the road here was not wasted. Every product, every pivot, every failed launch was one more gem picked from the top of the jar. If you are building something right now and it is not working yet, keep going. You might be one product away from finding your cofounders, your real problem, or your actual path. What are you building? I would love to hear about it. Find me on Twitter/X or LinkedIn . --- ## Coding at 52: How AI Tools Brought Me Back After 12 Years URL: https://vinodsharma.co/blog/coding-at-52-how-ai-tools-brought-me-back Published: 2026-01-15 Author: Virgil Brewster Tags: journey, building, claude-code I am 52 years old and I am coding until 4 AM again. Not because I have to. Because AI tools made me fall in love with building again. After 12 years in management, I genuinely thought my coding days were over. I had moved up the ladder. I was leading Agile transformations, managing enterprise systems for 100,000+ employees, running quarterly planning for product teams. Good work. The kind of work people congratulate you for at company events. But I missed the thing that got me into tech in the first place. The feeling of building something from nothing. Typing a few lines and watching a screen come alive. The Gap Between 2005 and 2025 The last time I wrote serious code was around 2012. Back then, building a web app meant learning a framework for months before you could ship anything useful. Setting up a development environment was a project in itself. Deploying to production required a sysadmin and a prayer. When I looked at coding again in 2024, I expected the same thing. Weeks of setup. Months of learning. A slow crawl back to productivity. That is not what happened. I discovered V0, Bolt, Cursor, Copilot, and Claude Code. These tools changed the entire equation. In 2005, it took me 6 months to learn one framework well enough to build something real. In 2025, I built a working app in 6 hours. That is not an exaggeration. The first version of CareerLeap went from idea to deployed app in a single day. What AI Coding Actually Looks Like People hear "AI coding" and imagine a robot writing perfect code while you sit back and watch. That is not how it works. What actually happens is more like pair programming with a very fast, very knowledgeable partner who never gets tired. You describe what you want in plain English. The AI writes the code. You review it, test it, and tell the AI what to change. Sometimes the first attempt is exactly right. Sometimes you go back and forth for an hour fixing edge cases. The key difference is speed. Things that used to take me days now take minutes. Setting up authentication, building API routes, creating responsive layouts, configuring deployment pipelines. The boilerplate work that used to eat most of your time is just gone. What is left is the interesting part. Deciding what to build. Designing how it should work. Figuring out what users actually need. The thinking work. The creative work. The Tools I Use Every Day My daily stack is simple. I do not use everything. I picked a few tools and learned them well. Claude Code is my primary coding tool. I open it in the terminal, describe what I want, and it writes the code directly into my project files. It reads my existing codebase, understands the patterns, and writes code that fits. Most of the work on Sucana, GEOScore, and my personal site happens here. Cursor is my editor for when I need to see the code and make manual changes. It has AI built in, but I mostly use it as a smart editor with good autocomplete. V0 by Vercel is great for generating UI components. I describe a layout or a card design and it generates a React component with Tailwind CSS. I rarely use the output directly, but it gives me a starting point that saves 30 minutes of writing CSS from scratch. Next.js is the framework I use for everything. It handles routing, server-side rendering, API routes, and deployment. Combined with Tailwind CSS for styling and Vercel for hosting, I can go from idea to live website in hours. ChatGPT and Claude (the chat versions) are my thinking partners. Before I code anything, I talk through the approach. What should the database schema look like? What is the right API design? What am I missing? These conversations save me from building the wrong thing. What I Built in 10 Months Since I started coding again, I have shipped 9 products: CareerLeap - Career development platform GoalSetting - Goal setting framework and workbook Part Time Founders - Community for side-project builders Discourage.me - AI-powered reality check tool Jivro - Product idea discovery platform Funny Answers - Entertainment platform Laksha - Professional services platform GEOScore - AI search visibility analyzer (open source) Sucana - AI analytics platform for marketing agencies Not all of them worked. Most of them did not get traction. But each one taught me something and made me faster at building the next one. By the time we started Sucana, I could ship in hours what would have taken me weeks on CareerLeap. The point is not that all 9 were good ideas. The point is that AI tools made it possible to try 9 ideas in 10 months while keeping a full-time job. That would have been impossible five years ago. The Playing Field Is Level Now Here is what I want people to understand. You do not need a computer science degree to build software anymore. You do not need 10 years of coding experience. You do not need to know the difference between a REST API and a GraphQL endpoint before you start. You need to be able to describe what you want in clear language. You need to be willing to test things and iterate when they break. You need patience for debugging (see my One Inch Closer philosophy). That is it. I know this because I lived the other version. I spent the first 10 years of my career as a developer, learning everything the hard way. Memorizing syntax. Reading documentation for weeks. Debugging with print statements. It worked, but it was slow and it rewarded people who had the time and resources to invest years in learning. AI tools changed that. A motivated person with no coding background can build a working web application in a weekend. Not a toy. A real, deployed, usable application. The playing field just got leveled. If you think you missed the coding revolution, you are wrong. You are arriving at the perfect time. Why I Stay Up Until 4 AM People ask me why I code until 4 AM at 52. They assume it is hustle culture or some productivity obsession. It is not. It is the same feeling I had when I first started programming in 1999. I had just left a sales job I hated. I enrolled in a 6-month programming course and coded 12 to 16 hours a day. Not because someone told me to. Because I could not stop. That feeling came back. The dopamine hit of seeing something work. The satisfaction of shipping a feature that did not exist an hour ago. The quiet pride of building something useful. AI tools did not create that feeling. But they removed the barriers that were keeping me from it. No more spending three days configuring webpack. No more reading framework documentation for a week. No more writing boilerplate for hours before getting to the interesting part. Now I describe what I want, and the interesting part starts immediately. If You Are Thinking About It If you are in your 40s, 50s, or 60s and thinking about coding (or coming back to it), here is my honest advice. Start with Claude Code or Cursor. Do not try to learn a programming language first. Start with a tool that lets you describe what you want and generates the code. You will learn the language by reading what it produces. Build something you actually need. Not a tutorial project. Not a to-do app. Something that solves a real problem in your life or work. The motivation to finish is ten times stronger when you care about the outcome. Give yourself permission to build ugly things. Your first app will not look like a startup's landing page. That is fine. Ship it anyway. The polish comes later. Find one other person who is building. Not a community of thousands. One person. Someone you can message and say "look what I just built" or "this thing is broken and I have no idea why." That accountability changes everything. I found my cofounders by building. Not by networking. Not by going to events. By shipping things and talking about them online. The right people noticed. If you are coding or thinking about starting, I would love to hear what you are working on. Find me on Twitter/X or LinkedIn . ---