DEV Community: The Great AI Adventure

I Created a Vibe Coding Checklist Nobody Asked For

The Great AI Adventure — Fri, 27 Mar 2026 18:08:02 +0000

And put 101 checkpoints in it, that nobody would ever be able to check off.

In case you are not a reader and want to checklist right away, here is the link .

There comes a point in your life when you remember that you can do things just because.

I started learning Spanish. Why? Just because.

I went without speaking for 3 hours. Why? Because I wanted to.

I put together a Vibe Coding Checklist that is longer than my resume even though I have been Vibe Coding for only 3 months. Why? Because I wanted to.

And because I can.

And of course, because I like to research.

In case this is the end of your attention span, here is the link to your too long checklist for vibe building .

So here is how it works.

It has 5 sections and (I hope) walks you through the complete journey of Vibe Coding a solution you have in mind.

And that means the foundation which where you actually put things down like what exactly are you trying to solve?

Because I realized a lot of people don’t know their why.

And as Simon (Sinek) says, know your why.

And then there are detailed things like what features do you want?

What exactly is you think you are building or you know some technical stuff that half of which I don’t even understand.

I don’t even know if it’s worth reading anymore. Here is the link to your vibe coding checklist.

AND It’s Inter… Wait for It… Active, Interactive

So like a propa vibe coder, So I went ahead and put it on GitHub so nobody can ever look at it and it goes down the black hole of the internet.

Now I have put some features on it so that it could look shiny but not usable at all.

If you are a procrastinator like me, you will start with adding the project name and summary and look at the long list of checkpoints and never come back to that checklist

Play things really!

but hey, I could do it with codex and so I did. If you are still reading here are some shiny things you will see:

Project name & Project summary (I know I already said it.)
Interactive checklist & Checkboxes with a Progress bar that will never be completed.
Phase grouping, Phase progress summary (So that at least I feel good about myself.)
Generic vs project-specific labels, Item filters, Show/hide completed toggle (You know, the generic vibe coding stuff.)
The best one: DOWNLOAD as a CSV (Ohhh, how exciting because you couldn’t ask your AI to make the checklist, so you need my checklist in CSV which nobody ever uses.)

How to Use It?

Gitpage

If you are a vibe coder, then I am not sure if you can do this whole GitHub repository thing where you have to clone the repository locally and then you can open the file in your browser locally and then so on.

For us uneducated builders, I just created a gitpage.

You just open it and start using it. Hope then it’s very easy for you to use it.

As for the GitHub page, if you haven’t noticed, I have it linked everywhere in this blog.

Sounds easy right?

I mean try doing it. It is 100% not easy.

Cloned Repo

And if you are someone who understands GitHub repo, congratulations!

You can just do the cloning thing, get it on your laptop and start using it.

Bonus (Not really)

Plus like I said, there is a CSV download feature so you could just probably download it if you don’t like the shiny things I added.

So that’s how you use it.

But one of the things I do want to point out is to use the security checklist. So you don’t miss stuff like exposing your tools user-base to some random person. Who can ask you to pay them $10,000 to just take care of things AI already could.

Not sure how you made it here, but try this useless vibe building checklist

Happy building-or time wasting-you will know soon!

I Gave Myself 2 Hours to Find AI’s Limits. Here’s What Happened.

The Great AI Adventure — Wed, 18 Mar 2026 16:48:44 +0000

Spoiler: I found more than I bargained for.

The Experiment

It started, like most rabbit holes do, with a dumb question.

I was messing around with a build, nothing dramatic, just trying to get a feature to work, and I asked Claude to help me debug it.

It did.

In about forty seconds. And instead of moving on like a normal person, I sat back and thought: okay, what, what actually can’t this thing do?

So I gave myself two hours. No agenda. Just pull the thread and see where it goes.

Two hours later, I had seventeen browser tabs open, a notes doc full of half-sentences, and absolutely no clean answer to my original question.

What I did have was something better, and something harder: a genuine sense of standing at the edge of something enormous, and a creeping feeling that we’re not fully ready for it.

Progress doesn’t walk. It teleports.

We talk about AI like it’s a graph going steadily up and to the right. But that’s not how it actually works.

AI progress plateaus, then something cracks open, an algorithmic insight, a weird idea someone tested, and suddenly you’re in a completely different paradigm.

We jumped from “predict the next word”

to

“Follow instructions”

“Reason through problems step by step.”

Each one felt like a wall breaking.

The most recent jump: models that actually think before answering. Not pattern-match. DeepSeek-R1 learned advanced reasoning through pure trial and error, no human examples, no supervised training. The researchers described it as the model having “aha moments.”

*It taught itself. *(Tbh some humans can’t even do that) **

Pull the thread on why models stopped just getting bigger, and you find something unexpected. The entire internet’s supply of high-quality human-written text, estimated at 9 to 27 trillion tokens, is projected to be completely exhausted by 2026 to 2028.

Not depleted. Exhausted. We are, in a very real sense, running out of internet to learn from.

I did not have “running out of internet” on my rabbit hole bingo card.

This is why the whole industry pivoted almost overnight: smaller models, better data, synthetic training. Microsoft’s Phi-4, 14 billion parameters, now matches the reasoning of models many times its size, purely because of how it was trained. The ceiling hit, so they built a different ceiling.

You are the whole product team now

A solo operator rebuilt a public service platform in 14 days. The original version took 15 engineers nine months.

I’ve felt a version of this. I built Shodh, a research tool, in 33 minutes. Two years ago I wouldn’t have even tried. Not because I lacked the idea, but because the gap between thought and thing-that-exists was too wide to cross without a dev background.

That gap is closing so fast it’s almost disorienting. The bottleneck is no longer writing syntax. It’s your ability to think clearly, make decisions, and shape what you’re building.

Which sounds incredible. And it is. But here’s where the rabbit hole gets complicated.

The part nobody puts in the highlight reel

We all bought into the dream that AI would simply take tasks off our plates. I certainly did.

The reality is messier.

When you use an AI, you’re not eliminating work, you’re shifting it. Instead of writing the code yourself, you’re now in a state of continuous sense-making: figuring out why the model hallucinated a Python library that doesn’t exist, or why it confidently produced something that looked right but wasn’t.

Researchers are starting to call this the AI tax, the cognitive and emotional load of supervising systems that operate at a speed and scale we can barely comprehend.

I tried to automate a simple internal workflow last week. The AI completely choked. Not a hallucination problem. A reality problem.

Because here’s what integrating AI into actual work exposes: most of our workflows aren’t clean, linear tasks. They’re held together by undocumented workarounds, institutional memory, and tacit knowledge that lives entirely in someone’s head.

When you try to map an autonomous agent onto a messy human process, you don’t get efficiency. You get what some early adopters are already calling agentic workslop, AI doing the wrong things really, really fast.

AI won’t fix a broken process. It will execute your dysfunction at the speed of light.

The quiet atrophy of judgment

This is where the rabbit hole gets a little dark.

Because these models are getting so good at logic, we’re starting to trust them with things they have no business touching.

Who gets an interview
Who gets a loan
How to resolve a customer dispute

It’s called moral outsourcing, handing ambiguous, consequential human decisions to algorithms because it’s faster and feels more objective.

But an AI has no stake in the physical world. It doesn’t feel the sting of a rejection or the weight of a denied claim. And when we consistently hand those calls to a machine, our own capacity for judgment starts to atrophy. Like an unused muscle.

We are so focused on making machines smarter that we are quietly deskilling our own humanity.

The physical wall

After all of this, the breakthroughs, the friction, the uncomfortable truths, I finally circled back to the original question. What actually can’t AI do?

The answer is strange. And it has to do with energy.

Your brain and nervous system run on roughly 15 to 20 watts. About the same as a laptop in sleep mode. That 20 watts handles your vision, language, memory, reasoning, and also lets you pick up a coffee cup, fold a shirt, navigate a crowded room without thinking.

Training GPT-4 consumed approximately 6,154 megawatt-hours. Enough to power 570 homes for a year. A 22x increase from the previous generation.

And yet, with all of that compute, AI still can’t fold a laundry shirt.

This is Moravec’s Paradox.

Cognitively hard tasks (graduate-level reasoning, complex mathematics, writing architecture at scale) are computationally cheap for AI.
Trivially easy tasks (physical dexterity, real-world navigation, presence) are computationally nightmarish.

The current limits of AI are, in the most literal sense, the most human things. Presence. Touch. Physical reality. Judgment with skin in the game.

Both things are true

My two hours ended. I closed about eight of the seventeen tabs. The others are still open.

Here’s what I came out with:

The wonder is real.
The friction is real.
You don’t get to pick just one.

By 2028, 15% of day-to-day work decisions are projected to be made completely autonomously by AI agents, up from essentially zero in 2024.

The people building right now, learning to think alongside these tools, figuring out where the AI tax is worth paying and where it isn’t, they’re the ones who will understand what’s actually happening when it arrives.

The floor underneath us keeps moving. What felt ambitious last month is obvious now. What feels impossible this week will be a template by Q3. The ceiling keeps moving too, just not always in the direction we expect.

I went looking for what AI can’t do.

I found a much more interesting question: what are we going to do with it?

I Vibe Coded a Tool to Find New Vibe Coding Ideas (in 33 Minutes) And I Think It Works

The Great AI Adventure — Fri, 13 Mar 2026 14:15:00 +0000

Shodh is live on GitHub pages (which I did learn from Gen-AIs). Try it on any subreddit where people complain about things. That’s most of them including me of course.

I started with a stopwatch and a plan to find three ideas on Reddit.

That’s it. Simple enough. Go on Reddit, find problems people are complaining about, pick three that feel interesting to build. Shouldn’t take long.

Forty-five minutes later I’d read seventeen comment threads, gotten into a debate about SaaS pricing, watched two video recommendations I did not ask for, and had zero ideas worth anything. Classic Reddit. The algorithm doesn’t want you to leave with something useful. It wants you to stay.

So I stopped. Closed the tab. Opened ChatGPT.

And I said: I need a distraction-free way to do this.

That conversation lasted maybe ten minutes. I sketched out what I wanted, input a subreddit, get back posts framed as problems or challenges, skip the ones that don’t interest me, save the ones that do. Simple triage. No feeds, no comments, no rabbit holes.

ChatGPT gave me a basic skeleton. I took it to Antigravity, started building, kept improving. Thirty-three minutes from that first message to a deployed, live tool.

I had the stopwatch running the whole time. Originally I’d started it to time how long it took me to find ideas on Reddit. It ended up timing how long it took me to build the tool to find them better.

What Shodh Actually Does

The tool is called Shodh (an idea scout really). Sanskrit for search, or inquiry.

It pulls posts from any subreddit you give it, filters by timeframe, hot, top today, this week, this month, all time, and surfaces them one at a time as challenges. You skip or accept. No noise. If something catches your eye and you want context, there’s a link to the original thread. Otherwise you just move through.

Under the hood it’s hitting Reddit’s API directly, filtering posts for friction-heavy signals, and surfacing each one with upvotes, comment count, and engagement ratio. The accepted tab lets you track status, add notes, and move ideas through a simple pipeline from Accepted to Started to Completed.

It broke a few times during the build.

The first version had a hardcoded list of subreddits and just kept looping the same posts. Had to add a proper input.

Then every time I added a new subreddit, instead of fetching posts it would print the instruction prompt back at me, literally outputting the code comment that said “scan this subreddit manually.”

Reddit’s API limitations meant some workarounds were messier than expected. But nothing stopped the build. Each break was a ten-minute fix, not a wall.

---

What Actually Surprised Me

Not that the tool worked. But how close it felt.

Before, and I mean even six months ago, taking an idea from my head to something functional and live felt like kilometers away. You’d need to find a developer, brief them, wait, iterate, wait more. Or you’d learn to code yourself, which is its own years-long detour.

Now it’s centimeters. Not zero distance. But the gap has collapsed.

There’s still judgment involved. Still back-and-forth. Still moments where the AI confidently gives you something broken. But the distance between “I have this idea” and “this thing exists in the world” has shrunk in a way that most people haven’t fully felt yet.

Does It Actually Work Though?

That was the real question I sat with after building it.

I’ve used it since. Pulled ideas from several subreddits I care about. Accepted somewhere around ten or twelve. Two of them have genuinely changed what I’m building right now, not because they were revolutionary insights, but because they reminded me to start from what people are actually frustrated about, not from the product I imagine they want.

That reorientation matters. It’s easy to build for the idea in your head. Harder to build for the problem in someone else’s day.

It’s not just an idea collector. It’s a grounding mechanism.

That’s the use case I didn’t fully articulate when I was building it.

The Bigger Thing

A few years ago, if you needed a tool that did something specific for your workflow, you had two options. Find a generic app and spend weeks trying to bend it to your needs. Or hire someone to build what you actually wanted.

Now there’s a third option: build it yourself, in 33 minutes, tuned exactly to how your brain works.

Shodh doesn’t have features I don’t need. It doesn’t have a pricing page or a waitlist or a growth loop. It has one job: help me find ideas without getting lost. It does that job. For me. Because I built it for me.

That’s the shift. Not that AI can build things. That you can build things, faster than it used to take to find the right search term.

I think that’s what “I think it works” means. The tool works. But more than that, the approach works.

Now I just have to build the ten things Shodh told me to build.

Happy International Women's Day 🌸 To every woman who's ever felt like the room wasn't built for her and walked in anyway. The AI revolution doesn't have a prerequisite. It just has a window. And it's open right now. Pull up a chair. There is room 💛

The Great AI Adventure — Sun, 08 Mar 2026 17:05:48 +0000

Why Every Woman Should Care About the AI Revolution?

The Great AI Adventure — Sun, 08 Mar 2026 16:03:54 +0000

Not because someone told you to. Because this one’s different. | Women’s Day edition | 8 min read

Nobody handed me a seat at the AI table.

I came to it sideways, through years of marketing work, through curiosity, through a vague but persistent feeling that something enormous was happening and I didn’t want it to happen without me.

I didn’t have a computer science background. I didn’t have a technical co-founder or a startup or a plan. I just started building. Experimenting. Paying attention.

And the more I built, the more one question kept surfacing: where are the other women?

Not because this is a women’s issue. Because this is a power issue. And power, historically, goes to the people who show up early.

We’ve seen this before. It didn’t go well.

Every major technological shift in modern history has created enormous wealth and opportunity. The internet boom. The mobile revolution. The rise of platforms. Crypto.

And every single time, women were underrepresented at the table where the decisions were made, and overrepresented among the people those decisions affected.

AI is no different.

Right now, women hold between 22% and 30% of global AI jobs. In data science, that drops to 12%. In cloud computing, 15%. In the UK, women account for just 14% of STEM roles.

At the leadership level, the people actually deciding how these systems work, what they prioritize, whose needs they serve, women hold just 10% of CEO and top tech positions across AI organizations.

That’s not just an equity problem. That’s a design problem. When the people building a technology don’t reflect the full range of people who will use it, the technology reflects their blind spots. Hiring tools that discriminate.

Medical diagnostics that underperform on women’s health data. Voice assistants that default to female-coded subservience. These aren’t accidents. They’re the output of rooms that lacked the right people.

Curious about the numbers?

Checkout the interactive version to take a closer look: Women & AI: The Numbers

I built a small interactive tool to go alongside this piece. 12 statistics, all connected, click any bubble to see the full story behind the number and follow the threads to related data points. Some of these stats will surprise you. The one in the middle surprised me the most.

The risk nobody is talking about loudly enough

The International Labour Organization analyzed 436 specific occupations globally to understand real AI exposure. And here’s what they found:

29% of female-dominated roles, clerical support, translation, administrative work, face high exposure to AI displacement. For male-dominated roles, that number is 16%. Almost half. And zooming out further: 57% of all jobs currently at risk of automation are held by women.

This isn’t because AI is targeting women. It’s because generative AI is entering a labor market that has been gender-segregated for decades.

Female-dominated professions heavily rely on text processing, scheduling, and routine communication, the exact capabilities that large language models are engineered to replicate at scale.

Meanwhile, male-dominated fields index higher in physical trades and hands-on work. A language model cannot fix a broken pipe.

The shockwave is hitting office-based economies first. In high-income countries, 41% of total employment is exposed to this disruption. If you are mapping your career for the next decade, this is not a distant concern.

The algorithm judging you right now

Modern fintech platforms have moved beyond traditional credit scoring, your repayment history, your debt-to-income ratio, to what researchers call learning algorithms.

These systems analyze alternative data: your mobile phone usage patterns, your social media connections, even which games you play in augmented reality. All to determine your financial credibility.

The problem is that this alternative data doesn’t exist in a vacuum. It reflects the same structural inequalities that traditional scoring was supposed to move past. Women in precarious employment, or those who took career breaks for caregiving, get flagged as high risk, not because they’re bad borrowers, but because their life patterns don’t match the model’s idea of financial reliability.

They get approved. But with APRs ranging from 300% to 700%. It’s not credit access. It’s a debt trap engineered by machine learning. And it’s completely invisible to the end user.

We have a tendency to trust machines as impartial. The computer says so, so it must be objective. But these models were built by humans, trained on human data, and they inherit every bias baked into that history. The gap is just hidden in the code now.

The confidence gap is real. And it’s costing us.

A Paris School of Economics study gave university students access to ChatGPT and watched how they used it:

Male students were 34% more likely to get the correct answer from the tool. Not because they were smarter, when you looked at the top-performing female students, they matched their male counterparts exactly. The gap wasn’t ability. It was persistence and confidence. When ChatGPT didn’t give the right answer, 71% of male students tried again. Only 55% of female students did.

When asked how confident they felt about their prompts, over 40% of male students said very or extremely confident. Only 18% of female students said the same.

I recognize that feeling. The hesitation before asking the dumb question. The assumption that someone else probably knows better. The sense that the tool was built for a different kind of person.

But here’s what the data also shows: the gap closes completely at the top. The most capable women perform just as well as the most capable men. This isn’t a ceiling. It’s a confidence tax. And AI is just the latest place it shows up.

But here’s what’s actually happening

If the story so far sounds heavy, this is where it turns.

In the past year, the proportion of US women adopting generative AI has tripled. That outpaces men, who grew at 2.2 times. The gender gap in daily AI use is closing, fast.

In Europe, 77% of female founders are actively using AI to scale their businesses, contributing to 5.76 billion euros in funding in 2024 alone, a third of which went into deep tech. Not basic consumer apps. Advanced AI, quantum computing, synthetic biology. Women building the fundamental infrastructure of the future.

In India, women-led AI startups grew 300% between 2020 and 2024, from around 520 to an estimated 2,100 companies. They’re building in health tech, ed tech, e-commerce, solving deeply entrenched problems that the mainstream tech industry has historically ignored.

And at a recent 72-hour vibe coding build-a-thon, women with no computer science background turned raw ideas into functioning software. Not prototypes. Not mockups. Working products.

Because when you remove the gatekeeping of traditional coding, what’s left is just: can you think clearly about what you want to build?

Turns out, a lot of women can.

What getting involved actually looks like

This is the part I want every woman who thinks “AI isn’t for me” to read slowly.

You do not need a computer science degree. You do not need to know how to code. What AI actually needs, what it is genuinely bad at and desperately requires humans for, is empathy, judgment, communication, creativity, and the ability to understand how real people actually live. The skills that have been historically undervalued in tech.

The skills that a lot of women have been quietly developing for their entire careers.

If we want AI systems that are less biased, that serve the full range of human experience, we don’t just need more computer science PhDs. We need anthropologists, psychologists, designers, storytellers, and people who understand what it actually means to live in a body in the world.

The Council of Europe’s new AI treaty, the first legally binding international framework on AI, is built on exactly this argument.

Getting involved looks like using AI to move faster in the job you already have. It looks like building something small just to see if you can. It looks like automating the mental load, the family logistics, the scheduling, the endless coordination, so there’s more space for the work that actually matters.

It looks like learning to prompt well, which is just another way of saying: learning to ask clearly for what you need.

None of that requires permission.

This one’s different

The internet required infrastructure. Mobile required hardware. Crypto required a particular kind of risk tolerance and tribal belonging. AI requires curiosity and willingness to try.

The barrier has never been lower. The upside has never been higher. And the window, where getting in early actually means something, is open right now.

The data is clear that diverse teams build better AI, less biased, more creative, more attuned to the full range of human experience. Which means women being at the table isn’t just good for women. It’s good for the technology. It’s good for everyone who’s going to live with whatever gets built.

And here’s the question I keep sitting with, the one that came up at the end of a research deep dive I did recently: if we eventually build AI that is perfectly equitable and endlessly patient, will we start preferring the empathy of our machines over the messy reality of each other?

The better these systems get at understanding us, the more it might change how we connect with each other as humans.

That question doesn’t have an answer yet. But the people who will shape it are the ones building right now.

I’m one woman who came to this sideways, who started building before she felt ready, who found the table and pulled up a chair without waiting for an invitation.

There’s room. There’s always been room.
The question is whether we’re going to take it.

Happy International Women’s Day.

If Your Marketing Stack Doesn’t Include AI, Now Would Be a Great Time

The Great AI Adventure — Fri, 06 Mar 2026 13:51:00 +0000

A provocation from someone who waited too long

Let Me Paint You a Picture

It’s sometime in the early 2020s. You’re a marketer. You have a tool for email, a tool for SEO, a tool for social, a tool for data, and none of them talk to each other (a nightmare subject sometimes). You are living inside a very unglamorous version of Everything Everywhere All At Once. Just without the googly eyes.

Twenty browser tabs. Ten logins. One overwhelmed human trying to make it all look intentional.

I know this because I was that human.

And for a long time, I told myself the chaos was normal. That this was just… marketing. That everyone was operating this way. That the tools would eventually catch up.

They didn’t catch up. Something else happened instead.

The Part Where I Have to Be Honest With You

Before AI became part of my daily workflow, a single blog post took me 8 to 10 hours. Not because I’m slow. Because the process was genuinely broken, research in one tab, outline in another, brief in a Google Doc, writing in something else, SEO keywords, and optimization somewhere else entirely.

When I finally admitted that to myself, it was embarrassing. Not because I’d been struggling, but because I’d been struggling and calling it normal.

That same post now takes under two hours. But here’s the thing — the time isn’t even the real story. The real story is where the saved time went.

AI didn’t replace my thinking. It cleared space for it.

The blank page paralysis, the first draft dread, the repetitive structuring that ate hours before I’d written a single real sentence, gone. What was left was the part I actually went into marketing for. The strategy. The judgment. The creative decisions that required a human to make them.

The Confidence Gap Nobody Talks About

Here’s a something that should make every marketer uncomfortable.

86% of marketers say they understand AI. But when asked to rate their expertise as genuinely strong, that number drops to 32%.

Are more than half faking it?

And I don’t say that judgmentally. I was in that majority for longer than I’d like to admit. I was using AI, but I was using it the same way you’d use a faster version of Google. Ask a question, get an answer, close the tab. Copy generation. Basic summarization. The occasional email draft.

That’s not a stack. That’s a party trick.

AI has reached critical mass, everyone has it installed somewhere. But almost nobody has it embedded anywhere.

The research calls this the “critical depth problem.” Awareness is everywhere. Actual transformation is rare. Most usage in 2025 and into 2026 is still surface level — doing the same old tasks slightly faster. Efficiency, sure. But not a different way of working.

What the Marketers Actually Winning Are Doing Differently

There’s a small group of marketers seeing something remarkable: plus 13% revenue growth and plus 13% cost savings simultaneously. If you know anything about how marketing budgets work, you know that’s not supposed to happen at the same time.

So what are they doing differently?

They’ve stopped treating AI as a tool and started treating it as a teammate. Not metaphorically but architecturally. The difference between a tool and a teammate is that a tool waits for instructions with zero context every time you open it. A teammate remembers. It knows your brand guidelines. It understands your audience. It can act, not just respond.

This is what people mean when they talk about agentic AI. And yes, it sounds like a buzzword. But the underlying shift is real.

Generative AI creates. Agentic AI acts.

One writes a blog post when you ask. The other monitors your pipeline, identifies which leads haven’t been followed up, and triggers a personalized sequence, without being asked.

A Concrete Example That Changed How I Think About This

Imagine you’re a hotel marketing manager. You have two customer segments, one loves the beach, one loves the pool. You have one master hotel image.

With AI in your stack, you generate a beach version and a pool version of that image. Each segment gets the version that speaks to them. Personalized creative, delivered at scale, in seconds.

But here’s what I want you to notice: the AI did the execution. It swapped the background. It rendered the images. It sent the right version to the right person.

The strategy, knowing that beach lovers are a high-value segment, knowing that personalized creative drives conversion, understanding the why behind the decision, that came from a human.

You are the director. AI is the special effects team.

The technology can’t invent the insight. It can only execute it faster than was ever previously possible.

Start Here-The Boring Stuff That Already Saves Hours

Before we talk about agents and automated pipelines, let’s talk about your Tuesday morning.

Because the biggest misconception about AI in marketing is that you need to overhaul everything to feel the difference. You don’t. The gains start small and they start immediately.

Here’s what I mean:

Your spreadsheets. Google Sheets and Excel both have AI built in now. You can describe what you want in plain English, “calculate month on month growth and highlight anything above 20%,” and it builds the formula. No more Googling syntax. No more copying formulas and breaking the whole sheet.
Your inbox. AI can draft reply emails from a single bullet point, summarize long threads before you respond, and flag what actually needs your attention. If you are spending more than 30 minutes a day on email, this alone is worth the experiment.
Your meeting notes. Tools like Granola or Notion AI sit in your meetings, transcribe everything, and pull out action items automatically. The follow-up email that used to take 20 minutes writes itself.
Your content calendar. Describe your audience, your topic, and your tone, and AI will give you a month of post ideas in two minutes. Not all of them will be good. But you’ll find three you actually want to write, which is three more than the blank page gave you.
Your ad copy variations. Instead of writing five versions of the same headline manually, describe your product and your audience and ask AI to generate 20 variations. Test the ones that feel right. Kill the rest. The whole process takes 15 minutes instead of an afternoon.
Your keyword and search research. AI changes how organic research works. You can drop a competitor URL into Claude or ChatGPT and ask it to identify content gaps, cluster keywords by intent, or explain why a page is ranking. What used to take a full afternoon with multiple SEO tools now takes 20 minutes of focused conversation.
Your on-page and content strategy. AI won’t replace your SEO thinking, but it will do the structural and repetitive work faster than any human. Meta descriptions, title tag variations, internal linking suggestions, content briefs built around search intent. All of it. In minutes.
Your answer engine presence. Search behavior is shifting. People are asking AI chatbots instead of typing into Google. If your content isn’t structured to answer questions clearly and directly, you’re invisible in that new layer. AI can help you audit and reformat existing content to show up in AI-generated answers, what people are calling Answer Engine Optimization.
Your Reddit and community research. Organic growth starts with understanding what your audience is actually saying. AI can help you analyze threads, surface recurring pain points, and identify the language real people use, which is also the language that ranks.
Your workflow handoffs. Tools like Zapier or Make let you connect your apps without code. A new lead fills out a form, AI summarizes their info, adds them to your CRM, sends a personalized intro email, and pings you on Slack. All automatically. All while you’re doing something else.

None of this requires a new strategy. It just requires starting.

This is the layer most marketers skip because it feels too simple to count. It counts. Start here, build the habit, and the bigger stuff follows naturally.

Where AI Belongs in Your Stack The Bigger Picture

Once the basics are running, here’s roughly how a more complete AI-augmented marketing stack looks:

Ideation and thinking partner: ChatGPT, Claude, Gemini. Your always-available brainstorming collaborator that starts fresh only if you let it.
Workflow automation: Gumloop or n8n to connect your tools without writing code. Scrape competitor data, feed it into your CRM, trigger Slack alerts. All automated.
Content and SEO: Surfer SEO or Semrush AI for strategy backed by real data, not gut feel.
Creative at scale: AdCreative.ai for test-ready ad variations in minutes instead of days.
Distribution: Zapier to route leads, trigger sequences, and remove the manual handoffs that eat your afternoons.

The point isn’t to use all of these. The point is to stop doing manually what can be systematized, so your actual thinking time goes to the things only you can do.

The Real Cost of Waiting

25% of organic search traffic is projected to shift to AI chatbots by 2026. If you aren’t thinking about how your content shows up in AI answers right now, you are already behind on a trend that is not going to reverse.

Marketers with genuine AI skills are commanding a 43% wage premium. Not a small bump, nearly half again as much.

And perhaps most telling: 95% of B2B marketers are using AI weekly. 65% daily. The question is no longer whether your competitors are using it. The question is how deeply.

The cost of waiting isn’t theoretical. It’s compounding.

The Mistake Most Marketers Make When They Do Start

They use AI to cut corners.

They pump out content because it’s cheap and fast. They chase efficiency, saving money, instead of impact, making money. And the output reads exactly like what it is: content shaped by a machine, with no human judgment applied to it.

We’ve all felt it. The LinkedIn post that is clearly AI-generated. The newsletter that sounds like it was written by someone who has never met their own audience. The uncanny valley of content that is technically correct and completely hollow.

That destroys trust faster than silence would.

AI used without intention creates noise. AI used with strategy creates leverage.

The marketers winning aren’t the ones generating the most content. They’re the ones using AI to handle the repetitive 80%, so they can pour their actual human judgment, empathy, and creativity into the 20% that builds something real.

One Thing You Can Do This Week

Don’t overhaul your entire stack. Don’t spend a week researching tools. Don’t wait until you feel ready.

Pick one bottleneck. One thing that eats your time and requires no genuine human judgment. Hand it to AI and see what happens.

That’s how it starts. Not with a strategy. With a single experiment.

The tools are ready. The technology is not the question anymore.

The question is whether you are structured to use them.

And if you’re reading this and thinking “I’ll get to it,” that’s exactly what I told myself. For longer than I’d like to admit.

How I (A Non-Coder) Found My Vibe Coding Stack. And Why You Probably Shouldn't Copy It?

The Great AI Adventure — Wed, 04 Mar 2026 15:25:31 +0000

How I (A Non-Coder) Found My Vibe Coding Stack

And Why You Probably Shouldn’t Copy It

A story about overcomplicating things, breaking stuff, and accidentally finding what works

Everyone Told Me I Needed a Stack

Before I wrote a single prompt, I was already overwhelmed.

I’d been lurking on Twitter, watching people build apps over weekends. I wanted in. So I did what any reasonable person does , I asked the internet where to start.

Big mistake.

The internet had opinions. Strong ones. You need Cursor. No wait, use Windsurf. Actually Bolt is better for beginners. Have you tried Lovable? Replit is underrated. Just use Claude. No, ChatGPT is better for code. Don’t forget GitHub Copilot.

I had a spreadsheet of tools before I had a single idea built.

I was treating finding a stack like a research project. Like there was a correct answer out there and I just needed to find it before I could start. This is, I now know, completely the wrong way to think about it. But it felt very responsible at the time.

So I Just Started Wandering

I didn’t pick my first tool scientifically. I picked it because someone mentioned it in a Reddit thread I was reading at 11pm.

That’s how this actually starts, by the way. Not with a plan. With curiosity and a tab open.

The first few tools felt wrong in ways I couldn’t fully articulate. Some felt like they were built for developers who already knew what a terminal was and were comfortable talking to it. Others felt like a maze, layers of options with no clear path for someone who just had an idea and wanted to build it.

I kept switching. I kept starting over. I kept wondering if I was too stupid for this or if I was just using the wrong tools.

The turning point came quietly. No dramatic moment. I just noticed one day that I’d stopped switching tools and started actually building things. Something had clicked , not because I’d found the perfect stack, but because I’d stopped looking for one.

The Thing I Got Wrong From the Start

I assumed the stack was the thing. That once I had the right combination of tools, building would feel natural.

What I actually needed to learn had nothing to do with tools.

It was how to think with AI. How to describe an idea clearly enough that something could be built from it. How to break a problem into pieces small enough to hand off one at a time. How to read an output and know whether it was right or just looked right.

Those are communication skills. Problem-definition skills. Marketing skills, honestly.

The tools are almost beside the point. Once you know how to think this way, you can pick up most tools in an afternoon. The thinking is the hard part. The stack is just furniture.

Fine. Here’s What I Actually Use

Since you’re going to ask: my stack emerged through experimentation and it’s embarrassingly simple. Six layers. Nothing exotic.

Idea Generation

Where the ideas come from before I write a single prompt. For me this is personal notes and Reddit, reading real conversations about real problems and noticing what keeps coming up.

AI Model , The Thinking Partner

The core of everything. This is what generates the code, explains what’s happening, and troubleshoots when things break. I use ChatGPT, Gemini, Claude, and NotebookLM depending on the task. No single one does everything best.

Editor / Environment , Where the Code Lives

The place where I actually interact with what’s being built. VS Code for most things, Antigravity when I want something lighter. Think of it as your workbench.

Where Things Run

How the thing actually becomes a thing you can open and use. Locally on my machine for experiments, Live Server for quick previews, GitHub for anything I want to keep or share, Vercel when I need it live on the internet fast.

Database , Where Information Gets Stored

Only needed when your tool has to remember something. Pocketbase for simpler projects, Firebase when things get more complex. I ignored this layer completely until I actually needed it.

That’s it. A typical session looks like this: I write out what I’m trying to build in plain language. The AI helps me generate a starting point. I test it. Something breaks. I describe what broke. We fix it. Repeat until it works or until I understand why it can’t.

It’s less like engineering and more like a conversation with a very patient collaborator who never judges you for not knowing what a dependency is.

Ok Then, Why You Shouldn’t Copy It?

My stack works for me because of how I think. I came from marketing. I’m comfortable with ambiguity, with testing, with not knowing the right answer before I start. I like breaking things and seeing what happens. I evaluate tools the way I evaluate campaigns, does this actually produce results, or does it just feel productive?

If you’re a different kind of thinker , someone who needs structure before they can move, someone who gets frustrated by things not working as expected, someone who wants to understand how something works before they use it , my stack might feel like chaos to you.

And there’s a practical problem too. The AI ecosystem is moving so fast that whatever I’m using today might be obsolete in six months. New tools appear constantly. The specific combination I’m recommending could be the wrong recommendation by the time you read this.

So copying my stack gets you my tools. It doesn’t get you the process I used to find them.

Steal This Instead

Before you pick any tools, ask yourself three questions.

First: how comfortable are you with breaking things? If the answer is “not very,” you probably want a more guided, visual environment to start. If you don’t mind chaos, something with more control and flexibility will serve you better.
Second: what are you actually trying to build? A quick prototype to test an idea is a different job to a full application with real users. Match the tool to the scope, not to what someone on Twitter says is best.
Third: what’s your budget for exploration? Most tools have free tiers. Start there. Don’t pay for anything until you’ve broken the free version.

But underneath all three questions is a more important one: how clearly can you describe what you want to build?

Because that’s the real barrier now. Not the tools. Not the code. Not the stack.

The barrier is your own clarity. Your ability to take a fuzzy idea and make it specific enough to build. Everything else is just furniture you can rearrange.

The Accidental Architecture

Nobody plans their first stack. You stumble into it. You try things, break things, abandon things, and one day you notice that something is working.

My stack isn’t a recommendation. It’s a record of what survived my experimentation. The tools that made it through are the ones that fit the way my brain works , a curiosity-driven, break-it-and-see-what-happens, marketing-brain way of building things.

Your stack will look different. It should.

The goal isn’t to find the right tools. The goal is to start building something and let the right tools find you.

What’s the first tool you’re going to try?

WTF Is Vibe Coding? A Marketer’s Honest Take

The Great AI Adventure — Tue, 03 Mar 2026 13:51:01 +0000

And why the people who understand problems might be the best builders of all

I Was only a Marketer, a Consumer, and Not a Builder

If you started your career in sales or marketing over the last decade, you know the pain. You buy a tool, try to bend the data to fit your specific situation, and quickly realize that doing the manual work yourself would have been faster. The tools are built for someone else’s problems.

For most of my career, there was a very clear line: marketers think, and developers build. I stayed on my side of it. I wrote briefs, sat in developer queues, followed up, compromised, and eventually got something close to what I had originally imagined.

Three years ago, I got so frustrated that I decided to learn Python, then JavaScript, then HTML. I quickly got lost in a sea of libraries and jargon. I basically just gave up (true story 😊).

Then 2025 happened.

The Moment the Rules Changed

Being chronically online, I started noticing something on Twitter (yes twitter not X). People were building apps. Fast. Not just developers but founders, writers, marketers, people who had no business shipping software were shipping software.

So I went to ChatGPT with an idea I’d been sitting on for a while: a custom marketing experiments tracking tool. I was tired of squeezing my workflow into generic solutions. I explained the idea. I also mentioned, almost apologetically, that I couldn’t code.

It didn’t care. It laid out a complete, code-free plan and started building with me.

Watching an idea move from my head to a working thing on screen felt a little like cheating. Like the rules had quietly changed and nobody had sent the announcement.

To be more specific: I was dizzy and overwhelmed. A few hours later, I was just getting the code and adding it to VS code and one click on “Go Live” text in the footer I could actually see things in action. One of the top-top feelings!

So What Actually Is Vibe Coding?

The term was popularized by Andrej Karpathy, an AI researcher. But the definition that makes most sense to me is simple.

Vibe coding is building software by describing what you want in plain language and letting AI do the technical translation. You are not writing syntax. You are communicating intent.

But it’s worth being precise about where it sits, because “using AI” means a lot of different things right now. There’s a version most people are already doing, generating text, drafting emails, summarizing documents. That’s useful. But it’s operating inside existing tools, inside someone else’s system. Call it Level 1.

Vibe coding is Level 3. You’re using AI assistants like Cursor or Claude to build your own software, custom tools, interactive prototypes, applications that didn’t exist before you described them. It’s not prompting inside a product. It’s using AI to create the product itself.

That distinction matters more than it might seem.

What I Actually Built

My first project was that marketing experiments tracking tool, the one that started everything. I’d been frustrated for a long time with generic solutions that required tedious manual filtering and were clearly designed for someone else’s workflow. So I described exactly what I needed and built it myself.

The second project came shortly after: a Reddit brand analysis tool to help surface new marketing ideas from real conversations happening in my niche.

Neither of these were technically complex in the grand scheme of software. But they were mine, built around my exact problems, my exact workflow. That specificity is something you almost never get from off-the-shelf tools.

What the Process Actually Feels Like

I want to be honest here because most writing about AI leans too hard into the magic.

The good part genuinely feels like thinking out loud. You describe an idea, the AI builds something, and your thinking gets pressure-tested in real time. Sometimes it builds something slightly different from what you imagined and you realize, actually, that’s better. The conversation sharpens your own thinking.

The frustrating part is real too. There were stretches where I was going in circles, describing the same problem four different ways, wondering whether I was being unclear or whether the AI just couldn’t figure it out. Sometimes both.

My biggest early mistake was a painful one. I gave the AI a small, casual prompt to fix an input error. One prompt. It proceeded to break the entire system and the background along with it.

That taught me something important fast. You can’t throw generic prompts at it and expect surgical results. You have to break problems down step by step, like explaining to someone smart but completely new to the context how to build a glass of water. Small increments. Clear descriptions. Constant evaluation.

Where It Hits Its Limits

Vibe coding is real. It is also not magic.

It hits walls with complex backend integrations, database connections, and anything that needs serious infrastructure. The AI has limited memory across sessions, it forgets context, occasionally hallucinates fixes that don’t actually work, and can produce bloated code that becomes unmanageable as a project scales.

And here’s the honest core of it: you still need to know what you’re building. You need to define the problem clearly, write structured specifications, give the AI rich context, and evaluate the output for edge cases. The AI handles the technical translation. The thinking still has to come from you.

Which brings me to the part I find most interesting.

Why Marketers Are Surprisingly Good at This

Here’s a realization I didn’t expect going in.

The skills that make someone a decent marketer, understanding problems deeply, defining audiences, thinking in outcomes, communicating clearly, are almost exactly what vibe coding demands.

You’re not learning syntax. You’re writing briefs. Precise, structured, outcome-oriented briefs. You’re describing behavior, edge cases, what the user should experience. Vibe coding is essentially just writing a really good creative brief, and then watching it come to life.

Marketers spend years developing a feel for what people actually need versus what they say they need. That instinct, that ability to peek into real problems and translate them into something buildable, turns out to be genuinely valuable here.

The paradigm has shifted. The value in software is no longer in knowing how to build a system. It’s in knowing what to build. And that’s a question marketers have been answering their whole careers.

The Bigger Thing Underneath All of This

There’s a question I keep coming back to when I think about vibe coding: who gets to build things now?

For most of software’s history, the gate was simple. You either knew how to code or you depended on someone who did. I spent years on the wrong side of that gate, writing briefs, waiting in queues, compromising on my original vision.

That gate is not gone. But it’s more open than it has ever been.

And I don’t think the people walking through it first are going to be developers looking for shortcuts. They’re going to be domain experts, people who understand a specific problem deeply, who’ve been sitting on ideas for years with no way to build them. Marketers. Founders. Analysts. People who always had the brief but never had the means.

The shift I’m experiencing isn’t just about tools. It’s about identity. Moving from someone who executes tasks inside other people’s systems to someone who builds their own. From task executor to system orchestrator.

What Comes Next

This blog is going to be my building experiment journal. I’m going to document what works, what doesn’t, and what surprises me. The wins, the broken systems, the unexpected moments, and the questions I can’t stop thinking about.

I’m not a developer. I won’t become one. But I’m building things now, and that feels like enough of a reason to keep going.

Here’s the one question I’ll leave you with: What was the first time you heard of Vibe Coding and what was your first impression?