Build It Anyway: How He Turned a Broken Automation Into a Chrome Extension for $135 and 19 Days of Work

Build to Launch Friday: Meet Dheeraj, the AI systems teacher who solves his own problems first and turns every build into a lesson worth sharing

February 20, 202633 min read

Build to Launch Friday: Meet Dheeraj, the AI systems teacher who solves his own problems first and turns every build into a lesson worth sharing

Published: February 20, 2026 URL: https://buildtolaunch.ai/p/vibe-coding-builders-substack-notes-chrome-extension-claude-code-135-dollars Engagement: 52 likes, 13 comments, 9 restacks Word count: 6135

Welcome to Build to Launch Fridays, where we meet the builders turning domain expertise into AI-powered products.

Every Friday, I’m spotlighting someone from the vibe coding builders collection who’s doing exactly what I believe is the future: using AI not as just another tool, but as a true collaborator to transform curiosity, passion, and years of professional knowledge into something scalable and ownable. No VC funding, no technical co-founders, no permission required, just domain experts who decided to build.

*Today, meet * — a teacher who couldn’t resist building the very thing he was trying to automate.

Chrome extension scheduling dashboard from Dheeraj Sharma, listed in vibecoding.builders and featured in Build to Launch Friday at buildtolaunch.ai - how domain experts build Chrome extensions with AI collaboration and practical automation expertise.

Have you ever spent hours setting up a workflow only to watch it break the moment you rely on it?

Dheeraj had a perfectly good automation running — an n8n pipeline that scheduled his Substack Notes like clockwork. Until authentication issues killed it. Most people would have searched for an alternative tool. Dheeraj opened Claude Code and started building.

Three days later, he had a working Chrome extension. Nineteen build days after that, it was approved on the Chrome Web Store. Total cost: $135.

Dheeraj has been in enterprise software since 2006 and currently leads a Cloud FinOps practice. He also runs GenAI Unplugged, where he teaches solopreneurs how to build real AI systems — not theoretical ones, but the kind he deploys for himself first. His n8n course alone has 42 lessons across 31 hours. Everything he teaches, he’s built and used. And SubflowAI is the latest example of that cycle in action: a problem he solved for himself that turned out to solve it for others too.

What I love about this story is how naturally it happened. A real problem, a tool he already knew, and the confidence to just start building. That’s the energy I want more builders to feel — if it would solve your problem, build it anyway.

You can see all of Dheeraj’s projects on his Vibe Coding Builders profile.

Quick Stats

Time to MVP: 3 days (December 1–4, 2025) Time to Chrome Web Store: 19 actual build days, 238 commits Launch Date: January 13, 2026 Current Version: 1.0.11 (11 releases in ~4 weeks post-launch) Users: 20 active beta users, couple of paying customers Build Cost: ~$135 total ($5 Chrome Web Store fee + ~$100 Claude Code Max + $4.70 in AI API credits) Ongoing Cost: Cloudflare Workers free tier, $4–10/month in AI credits AI Stack: Claude Code (development), Google Gemini (in-product AI features) Coding Background: Software engineer since 2006, but hadn’t built Chrome extensions or production JavaScript recently

👉 Visit SubflowAI | View on Vibe Coding Builders

Origin & The Problem

You run GenAI Unplugged with courses on n8n, MCP, Claude systems — and now you’ve built SubflowAI. How does the product-building fit alongside the teaching?

These two things feed each other. Everything I teach including n8n, AI automation, Claude Code, AI Agents, workflow design, I used to build SubflowAI. And everything I learn building SubflowAI goes back into my courses.

My n8n course on YouTube has 42 lessons on AI automation. SubflowAI started because I was trying to automate Substack Notes with n8n and kept hitting walls — no public API, fragile reverse-engineered endpoints, and authentication that kept failing randomly. That frustration became the product. And the story of building it with Claude Code became course material.

I don’t separate “teacher” from “builder.” The teaching is credible because I’m building real products or real AI systems for myself first, then share it with others. I become the first user with a problem. The products are better because I’m always thinking about how to explain the decisions.

This is the loop I respect most in any creator: build it, use it, teach from it. No distance between what he makes and what he teaches.

What’s your background? What was your coding experience before AI tools?

I started my career as a software engineer in 2006 and now lead a Cloud FinOps practice for an IT service provider. It’s been about 7–8 years since I’ve coded officially, but in my passion projects I get my hands dirty often, especially with my WordPress-based 17-year-old travel blog which often required me to code one thing or another.

Shipping enterprise software solutions has helped me develop a strong product sense. I know how to scope features, think about UX, prioritize ruthlessly, and ship. The gap was always execution speed. I could spec something in a day but it would take weeks to build, even with a developer.

Claude Code closed that gap. Not by replacing the thinking but by handling the implementation while I focused on decisions. The first time I described a feature and had working code in 20 minutes, I knew life had changed.

I have never been more excited working than now. Claude Code is an addiction for a builder like me.

I feel this so deeply. Claude Code closed that gap for me too — and not just for apps. For personal workflows, automations, content systems, all of it. The number of things you can build once that execution bottleneck disappears is insane. If you’re looking for ideas on what’s possible, I put together 12 Claude Code projects from beginner to advanced. And a 17-year-old travel blog? This is someone who builds things and sticks with them.

There are already several Substack Notes schedulers — WriteStack, Substack Pro Studio, others. Why build your own?

Honestly, I only knew that WriteStack existed but never used it or any other app. I remember had pinged me and I told him I was happy with my n8n pipeline for scheduling notes.

I had solved the problem with an n8n workflow thanks to Jakub Sly’s Substack Community Node. However, after a few days it stopped working due to authentication issues. One thing led to another, and while exploring solutions with Claude in Desktop, one suggestion was to solve it using a Chrome Extension. I was like, hmm... let’s try that out. SubflowAI was born over the next three days.

I wanted something that works with my overall content OS that I’d built around Notion and Claude. I’m a batch thinker. I want to sit down once, create a week’s worth of content, see it on a visual calendar, and walk away. That specific combination — AI repurposing + visual calendar + batch scheduling + customizable time slots — didn’t exist even in my n8n workflow. So I built one.

I also think this is why builders keep rolling their own versions of tools that already exist. It’s not about reinventing the wheel. It’s about building the wheel that fits your car. When the cost of building drops to a weekend and $25, “just use what exists” stops being the obvious answer.

I can so relate. I used to just tolerate friction in productivity apps — the ones that almost work but never quite fit. Now I build my own, because I can, because it saves me hassle, and because it teaches me AI along the way. More and more people are doing this. Some are fine with subscriptions, and that’s great. But for others, building your own private, customizable tool is the better path — and the possibilities are endless.

Building with Claude Code

Is Claude Code your main tool for SubflowAI?

Yes, I’m literally on Claude Code for everything I do now. Every line of SubflowAI was written collaboratively with it — the Chrome extension frontend, the Cloudflare Worker backend, the license validation system, the testing framework, the marketing content, even the privacy policy.

I focus on system architecture and UX decisions. Claude Code builds it. It’s not “AI wrote my code.” It’s pair programming with a patient senior developer who never gets tired and knows every API. I make the product decisions, Claude handles the implementation details.

I’ve used Google Gemini 2.5 Flash for the AI features of SubflowAI, including content repurposing and AI editing capabilities.

I love that — “pair programming with a patient senior developer who never gets tired.” That’s exactly how I feel about Claude Code, and the knowledge it teaches you between builds is the best thing I’ve ever experienced. I also split different AI tools for different jobs. We’re all becoming AI managers now.

Walk me through the timeline. How long from idea to working extension?

Three days to a working MVP. Nineteen actual build days to Chrome Web Store approval.

Days 1–3 (Dec 1–4): Built the core — visual calendar, rich text editor, AI repurposing via Gemini, smart scheduling with time slots, Cloudflare Worker backend. 38 commits. By day 3, you could paste an article URL, generate 5 Note variations, and schedule them across the week.

Days 4–11 (Dec 8–18): Monetization and polish — Lemon Squeezy integration, license validation, free trial flow, batch scheduling, onboarding, testing framework. 168 commits across 7 active days. December 17th alone was 58 commits.

9 days off for Christmas.

Days 14–17 (Dec 28–Jan 5): Non-code work — privacy policy, terms of service, Chrome Web Store screenshots, marketing copy.

Days 18–19 (Jan 6–13): Chrome Web Store submission. Two rejections, then approved.

43 calendar days total. 19 with actual work. 238 commits from first line of code to approval.

58 commits in a single day. That’s not a typo. And what I appreciate about this breakdown is how much of the total time was non-code: legal docs, store screenshots, marketing copy. The AI closes the coding gap, but the everything-else gap is still real.

Chrome extensions have quirks — manifest files, content scripts, browser APIs. How much did AI handle versus what you figured out yourself?

Chrome extension development is genuinely weird. Manifest V3, content scripts, service workers, message passing between contexts, Chrome storage APIs, alarm systems — it’s a different world from regular web development.

Claude handled about 70% of the implementation. It knows Chrome APIs deeply. When I said “I need scheduled posts to survive browser restarts,” Claude set up the Chrome alarms system, the onInstalled listener, the onStartup re-initialization — all the plumbing.

The 30% I handled was mostly debugging things that only show up in real usage. Like: content scripts inject into Substack’s page to post Notes, but Substack uses ProseMirror for its editor, which has a very strict JSON schema. Empty paragraphs cause HTTP 500 errors. That’s not in any documentation. I found it by posting a Note with a blank line and watching it fail. You need to work with AI as a partner, not as a sole doer.

The other thing AI can’t fully handle is the Chrome Web Store submission process. Google’s review is opaque. Our second rejection was because the reviewer couldn’t test the extension without a license key — it looked non-functional to them. No amount of AI could have predicted that. I had to add magic test keys and detailed reviewer instructions.

The blank line causing HTTP 500 errors made me laugh — I’ve hit similar surprises with Substack’s API. And his Chrome Web Store rejection experience is funny to compare with mine. I built my own Chrome extension too, but my submission went smoother because my extension links to a live website with Google sign-in, so the reviewer could just test it directly. Same store, very different experiences. But I agree on both findings: Substack’s API is cloudy and messy, and Google’s review process is unwieldy.

Has AI ever completely failed you during a build?

The Substack formatter was the hardest problem. Substack uses ProseMirror internally, and their Note format is undocumented JSON. You can’t just send HTML — it has to be their specific document structure with exact node types, marks, and attributes.

I spent hours on this. Claude would generate a converter. I’d test it. It would work for plain text but break on bold text. Fix bold, links break. Fix links, lists break. Fix lists, nested formatting breaks. Even after that, during beta, Pinkie from faced issues with the formatting of her notes as I couldn’t test all possibilities.

The issue was that Claude was working from general ProseMirror documentation, but Substack’s implementation has custom node types and specific validation rules that aren’t publicly documented. I had to reverse-engineer the format by creating Notes in Substack’s editor, inspecting the API payload in Chrome DevTools, and feeding those real examples back to Claude.

It wasn’t that I “did it manually.” I did the research manually and let Claude do the coding. That’s the actual workflow. AI is incredible at implementation, but sometimes you need a human to go figure out what the correct implementation even looks like.

In both the formatter and the Lemon Squeezy integration, Claude was stuck in a loop — solving the issue in similar patterns and going back to the first incorrect pattern again and again. Unless I as a human intervened, it would have kept burning my usage limits.

The hard part here isn’t AI failing. It’s that real platforms are messy and undocumented. You won’t hit Dheeraj’s exact problem, but you’ll hit something like it, guaranteed. The trick is you don’t just hand it off to AI. You explore together, discover different scenarios, and end up understanding the whole system better. That broader picture is how you help AI make the best decisions. (More on this kind of back-and-forth with AI coding tools.)

SubflowAI — The Product

What does SubflowAI actually do?

Both writing and scheduling. That’s the whole point.

Writing your own Notes: You open the dashboard, write a Note in the rich text editor (bold, italic, links, lists, images — all supported), pick a time slot, and schedule it. Simple.

AI repurposing: You paste a URL to an article or blog post — yours or someone else’s you want to riff on — and SubflowAI fetches the content, sends it through Gemini, and generates 5 standalone Note variations. Different angles, different hooks. You edit the ones you like, discard the rest, and schedule.

Batch scheduling: Generate 5 Notes from AI, hit “Schedule All,” and they spread across your week at optimal times — 7am, 10am, 1pm, 4pm, 7pm. Or customize which slots you use.

The calendar ties everything together. You see your entire month at a glance. What’s scheduled, what’s posted, where the gaps are. It’s the bird’s-eye view Substack doesn’t give you.

I have to be honest here. My own app, Quick Viral Notes, does almost the same things: AI repurposing from articles into 18 notes of different purposes, storing and scheduling Notes via Chrome extension. It even connects to MCP so any AI tool can work with your notes. We’re practically direct competitors. But I’m genuinely glad, because the whole purpose of building is to learn from each other and make the next thing better. I already got ideas from his batch scheduling approach. And Substack writers having more options that suit their needs? That’s a good thing.

When SubflowAI generates content, how do you make sure the output sounds like the user and not like AI slop?

The AI Studio repurposes your existing content — articles, newsletters, YouTube transcripts, LinkedIn posts. As long as it’s your own content, it’s going to sound like your brand voice in most cases.

Currently, a carefully tuned system prompt tells Gemini exactly what a good Substack Note looks like. Not generic social media content — specifically Notes. Each Note should read like something a thoughtful person would actually type using their own existing content.

The bigger lever is constraints. It generates 5 variations, not 1. Each variation takes a different angle — a hot take, a practical tip, a question, a contrarian view, a story. Users pick the ones that sound like them and edit the rest.

The goal was never “AI writes your Notes.” It’s “AI gives you 5 starting points so you’re not staring at a blank screen.” I mostly edit those repurposed Notes before scheduling. That’s by design. The AI saves you from the blank page problem, not from the editing.

This is the real edge. But even with your own content, AI can still be unpredictable and fall into formulaic patterns. I’m seeing more and more content where I can tell it’s AI-generated, and it’s not limited to one tool. Even if you’re just using ChatGPT to generate notes, it’s still not guaranteed to sound like you. So that’s the reality: keep trying, keep iterating, keep editing. The blank page problem is real, but so is the “sounds like AI” problem.

Security & Shipping

Chrome extensions interact with people’s Substack accounts. How do you handle credentials and security?

This was a deliberate architectural decision from day one: SubflowAI never touches your Substack password.

I was deliberate about this with Claude too. Every time it suggested storing credentials or tokens, I pushed back. The architecture rule was: if we don’t need to store it, we don’t store it. The fewer secrets we manage, the fewer things that can go wrong.

I executed two security audits with Claude subagents during a “Phase 1: Code Freeze & Audit” for production readiness. Three parallel agents were spawned: a security audit for XSS, injection, API key exposure, permissions, and CSP; a performance test for 1000+ notes handling, storage limits, and UI lag; and a technical debt cleanup for dead files and unused code.

In the second audit, I had Claude spawn 4 parallel security review agents — one auditing every chrome.tabs.create call, one checking all permissions and content script scopes, one analyzing network requests, and one reviewing every interval and loop. The report came back: 100% clean.

I didn’t ship a single line of code until that review was done.

I was debating whether to even ask this question because I built a Chrome extension too and I know the answer. But I’m so glad I asked, because the security audit part was not what I expected. Spawning parallel subagents to audit your code from different angles? That applies to way more than Substack. Any Chrome extension that touches social platforms needs this kind of review. If you’re curious about how AI agents work under the hood, I broke down the one pattern behind every AI tool recently.

AI hallucinates. When you’re shipping something that touches users’ accounts, how do you catch errors?

Testing. A lot of testing.

I built a full testing framework. Playwright for end-to-end tests, Jest for unit tests. There’s a 14-scenario manual testing checklist I run before every Chrome Web Store submission. Things like: schedule a Note, wait 2 minutes, verify it actually posted. Expire the license, try to schedule and verify it’s blocked. Restart Chrome, verify alarms resume.

But the most important safety net is the architecture itself. SubflowAI doesn’t auto-post silently. Every scheduled Note shows up on the visual calendar. You see exactly what’s going out and when. If something looks wrong, you catch it before it fires.

For the AI repurposing feature specifically, the app never auto-schedules generated content. The AI gives you 5 variations. You review them. You edit them. Then you explicitly schedule. There’s always a human in the loop before anything touches your Substack account.

The one thing I’m still paranoid about: the Substack API is reverse-engineered. It could change without notice. So SubflowAI has retry logic, error handling, and a “failed queue” — if a post fails, it doesn’t disappear. It moves to a visible failed state so you know to retry.

Human in the loop is the way to go for a long time. Everyone is already using AI in some ways, whether they realize it or not. The most important part is that WE set the restraints and WE work with AI together, not just hand it off to do things on its own. These kinds of decisions are what make your product production ready or not. It’s the 90% that isn’t writing the first draft of the code.

The Business of Building

What did SubflowAI actually cost to build? Real numbers.

This is my favorite part.

One-time costs:

Chrome Web Store developer fee: $5

Domain (genaiunplugged.com — already had it): $0 incremental

Monthly costs:

Claude Code (Max plan): ~$100/month (but I use this for everything, not just SubflowAI)

Cloudflare: $0 (free tier — 100K requests/day)

Google Gemini API: $4–$10

Lemon Squeezy: ~6–7% of revenue (only pay when I earn)

Total direct build cost: about $135 — the $5 store fee plus a month of Claude Code that I was already paying for.

The infrastructure is essentially free at current scale. Cloudflare’s free tiers are generous. Gemini’s cost is under $5 for current load. The only variable cost is Lemon Squeezy’s commission on actual sales.

If I had hired a developer to build this? Conservatively $5,000–$10,000. Probably more, because the iterative back-and-forth would have taken months of calendar time.

The Claude Code subscription paid for itself before SubflowAI’s first customer.

Most builders I’ve interviewed land somewhere in the low hundreds of dollars for a near-production app. But the cost isn’t the full story. Handing it off to a developer is one thing. The context switching, the constant discussions, the feedback loop back and forth? That’s a time sink I can’t afford. And honestly, for the 19 days Dheeraj spent and all the tweaking in between, if you were paying someone else to build this, $5,000–$10,000 sounds about right. He created a $10K-value app himself. But can you guarantee a developer delivers exactly what you have in your head? It’s not that their quality is bad. It’s the specifics, the requirements, the patience, the communication. All of that friction exceeds the dollar cost. And that’s the beauty of building with AI: you spend as much time as you need because you’re passionate about it, and the low cost keeps pushing you forward.

What does SubflowAI do that existing tools don’t?

Honestly, I haven’t used those tools, so I have no idea how SubflowAI compares. I just built it for myself without thinking about releasing it to the public. Thanks to feedback from friends, family, and especially some Substackers — including , who’s a pro at publishing Chrome extensions — their advice pushed me over the edge and I released it.

I know it solves my problem very effectively, so I was sure it would solve others’ problems as well. And the beta user feedback has been very pleasant — extremely effective, great user experience, highly productive in terms of time saved, and a deal worth every penny for the first 50 users.

SubflowAI is a byproduct of what GenAI Unplugged is all about: build it, learn from it, share it, repeat. I built it because I had a problem. I learned a ton solving it. Now I’m sharing the tool AND the lessons including my build log series. That cycle is the whole point. The product feeds the teaching, the teaching feeds the product.

It’s proof that the approach works.

*I love seeing this. Karen is one of my favorite AI builders in the community. She brought such boldness and willingness to try things that are supposedly really hard. She also built her own Chrome extension for scheduling Substack Notes, and her attitude was just: why not, let’s build it, no need to charge. And * was the first one I know who came up with the Chrome extension scheduling idea, with a brilliant lean approach and tons of paying users to show for it. Now Dheeraj joins them. None of them built to beat anyone. They all built to solve their own problem. And what I cherish most is seeing community members helping each other ship. Substackers have more options, builders have more passion. That’s what this community is for.

The Teacher-Builder Flywheel

You teach AI automation to solopreneurs. What’s something AI still can’t do well that students always expect it to handle?

Taste. Product decisions. Knowing what to build, what to skip, and when to stop.

Students come in expecting AI to tell them “build this feature next” or “this is the right UX pattern.” But AI gives you every option. You ask “should I add a settings page?” and it’ll build you an elaborate settings page with 30 toggles. The question was never whether it could — it’s whether you should.

In SubflowAI, I deliberately removed features Claude suggested. It wanted to add a full theme customizer. I said no, just a dark/light toggle. It suggested notification badges, floating widgets, keyboard shortcuts panels, image support, link cards support. All technically correct suggestions. All wrong for v1.

The thing AI can’t do is sit with your users, feel their frustration, and know which ONE thing to fix first. told me she only posts 3x/day and the 5-slot system was forcing her into slots she didn’t want. No AI would have prioritized that over the 50 other things on the backlog. But it was the right call.

When AI removes the friction of execution, what’s left for us? Taste. Judgment. Personal preferences. AI surfaces what looks like the best option, but it doesn’t have opinions of its own. We bring the nuances, we make the call. Most of the time you have to be the person making the decision, and that instinct is the one thing AI won’t replace. (And Elena — she’s also a builder in our community. Love seeing her name come up here.)

How has building your own products changed how you teach?

Small correction — I teach solopreneurs how to build AI systems. AI automation is one part of that, but it’s the whole stack: agents, workflows, tooling, infrastructure, the systems thinking that ties it all together.

Building products hasn’t changed how I teach. It’s how I’ve always taught. SubflowAI just made it more visible.

I’ve always built first, taught second. The 42-lesson n8n course — 47 videos, 31 hours — most of the teaching came from a real problem I was solving for myself as a solo content creator with a day job and no team.

The difference with SubflowAI is that I actually released it as a product for the first time. Before that, I was building all of this but keeping it internal — solving my own problems, then teaching from those solutions. 2026 changed that for me. Now I complete the cycle by releasing the product as well.

I only teach what I’ve built and used myself. No jazzy demos, no jargon. If it didn’t solve a real problem for me first, it doesn’t make it into a course or a video.

“I only teach what I’ve built and used myself.” That’s the sentence that separates credible teachers from everyone else. And what changed in 2026 wasn’t the teaching method, it was the release. I’m looking forward to the whole AI teaching landscape to follow the same paths: from “build internally, teach externally” to “build, release, AND teach.” The full cycle.

Reflections & Builder’s Taste

You built in a crowded category. What did that teach you about your own taste as a builder?

I didn’t know it was a crowded category when I started. I’m fairly new to Substack, just about 3 months old here.

So the lesson wasn’t “how do I differentiate in a crowded market.” The lesson was that when you build for yourself first, you end up with something opinionated. And opinionated turns out to be the differentiator.

I wanted a visual calendar as the primary interface because that’s how I think about content. I wanted AI repurposing that goes from article URL to scheduled Notes in one flow. I wanted batch scheduling where one click spreads 5 Notes across the week.

Those choices came from my workflow, not from studying competitors.

What I learned is that your taste as a builder shows up in what you include AND what you leave out. SubflowAI doesn’t have analytics on post performance. No multi-account support. It doesn’t try to be a growth tool. It tries to be the fastest way to get from “I have content” to “it’s scheduled.” That focus came naturally because I was building for one user first: me.

Turns out, when you build something that fits your workflow perfectly, other people with similar workflows find it and go “finally, someone built this the way I think.”

Ha, fair enough — sorry I put him into the “crowded category” scenario. But this kind of tension tells you exactly what building is like. Whether you joined Substack today or a year ago, it’s still going to feel like a brand new day. That’s a reminder for myself too. The philosophy here is what I love: build for yourself, stay opinionated, and the right people will find you.

What question would you love someone to ask about SubflowAI that almost never comes up?

“What did you remove?”

Everyone asks about features. Nobody asks about what I deliberately cut. And the cuts are what made the product good.

Claude Code suggested a full theme customization system. I cut it. It suggested notification badges and floating widgets. Cut. It suggested a YouTube transcription mode that would turn video content into Notes. I actually built it, then hid it because it wasn’t reliable enough.

The 58-commit day where I built batch scheduling produced the most important feature in SubflowAI. But the following week, I removed three features that were making the UI cluttered. The undo/redo buttons went away (Ctrl+Z still works). The image upload button got simplified. A “suggested topics” panel got deleted entirely.

Shipping is choosing. And choosing what to remove is harder than choosing what to build. AI will happily build everything you ask for. Your job is to not ask for everything.

This really resonated with me. The 58-commit day produced his most important feature, and the week after, he had the courage to cut three things that came out of that same sprint. That takes real maturity as a builder. I struggle with this myself, once I’ve built something, I want to keep it. Dheeraj is better at letting go than I am.

I have a full-time day job. GenAI Unplugged — Substack, YouTube — all of it happens in the margins. Early mornings, late nights, weekends.

My system has three layers: automate, batch, compress.

Automate: I built an AI-powered content operating system called PubFlow OS that runs the entire GenAI Unplugged pipeline. Research, drafting, auditing, social distribution. The system tells me what’s next. I just follow. When you’re running content across a newsletter, two YouTube channels, LinkedIn, and a travel blog, you can’t carry all of that in your head. The system carries it.

Batch: I don’t context-switch daily. Product features get focused sprints. Course content gets created in blocks. SubflowAI is literally a product built around batching.

Compress: Making small windows count. Claude Code turns a 2-hour evening session into a shipped feature. Before AI tools, 2 hours meant “read the codebase and remember where I left off.” Now it means “describe, iterate, test, commit.”

That combination is what makes the multi-project life possible with a day job.

This speaks to me so much. Full-time job, early mornings, late nights. My weekends are occupied with kids in the house, but pretty much everything else is the same. And seeing how he runs Substack, YouTube, courses, a personal hub, AND LinkedIn? That gives me more courage to actually step out to YouTube and into building my own courses. Thank you for sharing all that so openly, Dheeraj.

For someone thinking about building a tool even when alternatives exist, what’s your advice?

Build it if you’d use it yourself. That’s the only test that matters.

Don’t build it to compete. Don’t build it because you think you’ll make money. Build it because the existing tools don’t fit the way your brain works, and you now have the ability to fix that in days, not months.

Three practical things:

Ship the ugliest version first. SubflowAI v1.0 wasn’t pretty. “Works” beats “beautiful” every time, especially at the start. You can’t get user feedback on code sitting in your repo unshipped.

Talk to users immediately. My best features — customizable time slots, missed notes recovery, link preview cards — all came from users, not from me staring at a roadmap. Ship, ask, build what they say. Repeat.

The cost floor has collapsed. I built a monetized Chrome extension with AI features, serverless backend, payment processing, and image hosting for about $135 and 19 days. The math on “build vs. buy vs. use what exists” is completely different now.

If the tool you wish existed is 3 days away, why wouldn’t you build it?

I totally agree on all three. If I build something, it has to be something I want to use daily and something I can afford to keep running, even if no one pays. That’s the bottom line. More and more people are going to adopt this mindset, and I think it’s what makes building sustainable. If you’re curious about building apps, just step in and try it. Day one.

Connect & Explore

Dheeraj’s Platforms:

🚀 SubflowAI — One of the fastest ways to schedule Substack Notes
📰 GenAI Unplugged Newsletter — Practical AI systems for solopreneurs
🌐 GenAI Unplugged — Courses on n8n, Claude Code, AI agents, and MCP
👤 Dheeraj’s Vibe Coding Builders Profile — See all his projects and journey

Courses & Content:

🎓 n8n AI Automation Course — 42 lessons, 31 hours
🌏 Discover with Dheeraj — His 17-year-old travel blog

Final Thoughts

Dheeraj is one of those people who makes building look natural — not because it’s easy for him, but because he’s wired to turn every problem into something he can share. A broken automation becomes a Chrome extension. A Chrome extension becomes course material. The course material helps someone else build their own thing. The cycle never stops, and he genuinely loves every part of it.

What I keep thinking about after this conversation is his framework: automate, batch, compress. He has a full-time day job, runs a newsletter, teaches courses, maintains a 17-year-old travel blog, and still shipped a product in 19 days. Not because he has more time than the rest of us, but because he’s built a system where every small window of focus counts. Claude Code turns his evening hours from warm-up time into shipping time. That’s the real transformation here.

And his generosity stands out. He’s not hoarding what he learns — he’s turning every build decision, every debugging session, every feature cut into lessons for his students. The build log series for SubflowAI is a masterclass in transparent building. You don’t get that from someone who’s just trying to ship a product. You get it from someone who genuinely wants to help others build too.

If Dheeraj’s approach resonates with you — building from personal need, teaching from real experience, sharing everything openly — connect with him. He’s the kind of builder who makes everyone around him better.

[

GenAI UnpluggedSimple AI automation systems for solopreneurs who are done doing everything by hand.By Dheeraj Sharma](https://genaiunplugged.substack.com?utm_source=substack&utm_campaign=publication_embed&utm_medium=web) If you’re turning your expertise into products, building with AI, or helping others do the same, you belong here. Join the vibe coding builders community and get featured on Build to Launch Friday. Curious why it all started? Here’s the full story behind Vibe Coding Builders.

Your turn:

Have you ever built something just because you needed it — and then realized other people needed it too? What was that moment like?

Dheeraj teaches AI systems by day and builds them by night. His n8n workflow broke, and three days later he had a Chrome extension. What would you build if you gave yourself permission to just start?

Dheeraj went from a broken automation to a Chrome Web Store product and beta users who love it — all for $135. What will your builder story be?

— Jenny

This is part of my series profiling domain experts who are building AI products. If you’re building something with AI, I’d love to hear your story.

← All articles