It was also a lot: a lot of events, a lot of work, and a lot of fun. I got the opportunity to live right in Manhattan, the heart of New York City, and see if I could survive in the city that never sleeps. I also got a front-row seat to how the dramatic rise of AI both powered and clashed with today’s classroom.

All in all, grad school looked very different from undergrad — six years after graduating from Indiana, a lot has changed. Below are 26 thoughts on graduating Columbia Business School Class of 2026.

1. Let’s start with the most important thing: overall, I loved it. In the words of Frank Sinatra — “Regrets, I’ve had a few, but then again, too few to mention.” That about sums up my two years at CBS.
2. I would encourage anyone to get an MBA if…
3. You have some sort of non-loan financial assistance (scholarships, employer assistance, family help, etc.). Even from an M7 school, jobs are never guaranteed, but debt always is.
4. The ROI on school seems to pay off more in 5-10 years than it does right out of school. While you might get a good job right away, it’s the network that helps you land the dream job down the road. Additionally, as the world gets more “Ai-ified,” your network will matter more and more, which is an immense benefit of school.
5. Doing grad school in a major city was the right choice. Looking back, I was really happy with my order of decisions to do a college-town undergrad (Bloomington) and a major city grad school (NYC). That sequence just made more sense for each stage of my life.

I went to the Met a few times which made me feel like a real New Yorker

6. NYC is the best place I have ever lived. It’s expensive as hell but you get what you pay for. I found myself spending more money on doing things just because there’s always something to do - every block has an amazing restaurant, bar, theater, or most likely, all three.

Christina and I got some great engagement photos in the city

7. Also, New Yorkers aren’t mean, they are just direct.
8. Contrary to undergrad where everyone is growing up together, most people at grad school are at a different stage of their life. You naturally gravitate towards people in the same ‘life-stage’ as you. This can be a challenge for building friendships but also led to a cool diversity of experience.
9. Joining rugby was the best decision I made at school because it brought together all different kinds of people for a common goal. You gain a lot of mutual respect when you put your bodies on the line for one another (especially when you’re closer to 30 than 20), and you cannot manufacture that respect anywhere else in school.

The Fellas

10. Going to an Ivy league school does not mean I’d work with everyone here. There are a lot of very smart people at the school, but just like anywhere else, you have those you’d choose to work with and those you’d choose not to.
11. The clubs at grad school make and grow the community.
12. The hardest, but most rewarding, semester of school is the first - new environment, intense classes, and you’re likely recruiting. Doing all this at once is challenging; however, there is a sense of ‘shared trauma’ quality to those 4 months that bring together the students going through it.
13. CBS, and many other grad schools, have a policy called Grade Non-Disclosure (GND) which means you don’t report your GPA to recruiters. You still earn grades, they just aren’t reported. Ultimately, I thought this was a good policy. While it was misused by some to put little to no effort into classes, for many it was an invitation to try new or challenging classes without the fear of consequences on a transcript. This also reduced the amount of ‘sharp elbowing’ done during recruiting.
14. On a similar point, I found the best use of grad school is to try new things and push yourself out of your comfort zone. I tried a new sport (rugby), became an orientation leader, leaned more into community service (taught at a prison), took and TA’d challenging classes, and even participated in a fashion show. It is a low-stakes environment that encourages you to take risks. I will look back at my time at CBS and think first of doing all these new and interesting things.

Struttin’ my stuff in the CBS Retail Luxury Goods Fashion Show

More rugby

My fellow orientation leaders for one of the clusters at CBS

15. Opportunities at grad school aren’t just handed to you on a silver platter, as some people seemed to think at the beginning. However, there are more opportunities available and presented to you than anywhere else I’ve ever seen. Opportunities to do anything and everything, professionally and socially…they just take a lot of effort to get.
16. The most engaged classrooms had a no-laptops no-phones policy. We were all provided a locked down iPad for notes, which worked well.
17. My biggest classroom pet peeve was when people were on their laptops during guest speakers. It makes them look bad but it also makes everyone else look bad.
18. There’s a bit of an (outdated) idea that people get an MBA to find a partner. I didn’t really see this…probably more than 50% of people come into school already with a partner, and I didn’t see that many students get in relationships with each other.

One of my favorite nights of NYC - bringing some of our best friends from Chicago out to Karaoke at 3AM

19. I used AI a lot in school - specifically, I used tools like Claude and ChatGPT as private tutors to go deeper on concepts, and NotebookLM to make podcast summaries of cases I read. AI enhanced my learning experience.
20. At some point midway through second semester year two (Spring 2026) students started speaking of Claude as their main AI tool instead of ChatGPT, which was the most prevalent tool until then. While I cannot pinpoint why exactly this happened, I think it’s likely some mix of the fact that Claude got a lot of popular press for the refusal of the DOD contract (and vice versa for OpenAI) and then people started feeling like it had better models around that same time (mostly for automating powerpoint presentations).
21. AI is creating a bifuraction of student effort. For one half, it’s enhancing everything they do - better prep, better analysis, better productivity, and even better creativity. For others, it’s simply a bypass. It’s hallowing out the majority middle and pushing students towards either effort extreme - much more or much less. Mark Cuban thinks so too.
22. Schools are still trying to figure out how best to integrate AI into the curriculum. At CBS, while not perfect, steps were made during the two years to add more classes and more AI-tools. The question now is if education as a whole can keep up.
23. AI made standard problem-set homework almost meaningless. AI has gotten so good that you can upload 20 stats questions and it can solve it almost perfectly. We need to rethink what homework even is.
24. Bloom’s Taxonomy describes levels of learning — from the foundational (remembering, understanding) up to the complex (analyzing, evaluating, creating). Historically, a lot of homework lived at the bottom: here are 20 problems, submit them, never discuss them again. AI has made that kind of work trivially easy. The opportunity now is to design learning that lives at the top of the pyramid — live debate, real analysis, building things. That’s where the actual learning is, and it’s also where AI can’t just do it for you.
25. Two of the best classes I took had drastically different AI policies. One had a near-zero AI policy, while the other used it every single class. Both classes had the following characteristics in common: cold callilng, plenty of discussion and debate, and live presentations and feedback.
26. Two years is exactly the right amount of time. There’s no way I could do another year, but I felt like I was just hitting my stride after year one. And from everyone I’ve talked to - most of them loved their time here. I am definitely one of them.

Thanks CBS for an amazing two years!

First day of class

Last day of class. Please excuse the cigarette and beer

That’s a wrap!

From the syllabus…

“The focus of this course is to give you a working knowledge of what it takes to customize and assemble a customized generative AI application.”

That’s about as good as I could describe it. It’s about building AI applications.

Now, the “AI portion” goes two ways - it’s about integrating AI into applications (using OpenAI’s API) but also using AI to build (Cursor, Claude Code, etc.).

The class was only five days, 8 hours a day, and was broken up between course work and project work. Again, from the syllabus…

“About 50% of the class time will be devoted to a group project where you will, in small groups, build your own customized AI applications.”

My job as a teacher’s assistant wasn’t to teach, but to walk around and help each group during project time. What was cool about this job was that I got to go to each group, understand what they were building, why they were building it, and how they were building it.

Another unique part of this class is that it’s for beginners - you need very little, if any, tech knowledge before taking this class. Essentially, I got a firsthand look at what beginners build and how they do it.

After five days of watching (mostly) new techies build real AI apps, I came away with a few observations - some expected, and some that genuinely surprised me.

On The Tools

Cursor was the main coding tool of choice. While I think there are better coding products out there today (Claude Code cough cough) it was relevant for newbies to learn about IDEs (even if they aren’t around forever). Additionally, there are quicker ways to mock up applications using tools like Lovable or Base44; however, many of these students want to eventually work in tech firms, and again, it’s good to know how enterprises are shipping code today. If you want to eventually explore your own technical projects (i.e. OpenClaw right now) there is value in knowing the terminal, file stucture, etc.

Mostly all the interaction in the IDE comes from working with the agents, not the code itself

OpenAI API was our token of choice. Most people know and understand OpenAI, and their API documentation is pretty easy to read through. We also make sure to set usage limits just in case something goes wrong.

Vercel blew up this year. Seldom did people use it last year. Ease of deployment, in my mind, has made one of the biggest leaps since last year. I’ve personally used Vercel, and it’s absolutely excellent - simply connect it with a repository and it will create a shareable link for you.

Supabase was the go-to tool for databases. Supabase itself was a great, easy to use option for application databases, but databases created a bit of a hassle for some student groups. More on that below.

If I were to create an application from scratch, as a beginner, I would generally follow this flow, which is what most students did.

Code in cursor. Push to Github. Deploy to Vercel. You have a fully functioning, shareable application

On Getting Stuck

Troubleshooting is a skill. For beginners, most troubleshooting (primarily in Cursor) consisted of the following prompt - “plz fix.” Not really…but closer to that than an actual, effective troubleshooting message. Truthfully, I believe good troubleshooting comes down to a bit of experience. The best troubleshooting comes from knowing where the error is and what the expected behavior should be, not from immediately knowing what the error is. Most of my time as a TA was troubleshooting with teams during group work, and I didn’t know 80% of the issues upon initial review; however, all were solved because I knew how to troubleshoot. Learn that skill.

Reduce any need for unnecessary connections. For example, one team ran into issues connecting to their supabase database. Other teams had trouble connecting to multiple Google Calendars and Drives. This is a five day class where you showcase your product at the end - in my mind, there is no need to overcomplicate things, especially for demos. This is even true for a production grade application I’m trying to demo to investors - the first focus isn’t auth or connecting different data sources - it is making something functional and that sizzles. Keep it as simple as possible, even if that means using mock data locally.

Github, specifically branching, is really difficult for beginners to wrap their heads around. People worked in teams of 4-5, and for some of the teams that tried to do branches they ran into merge issues pretty much instantly. For this environment, it was better to have one person focus on the code at a time.

Some AI’s went rogue with output, and it takes someone with a bit of experience to see when it goes down weird little rabbit holes. For example, one group’s Cursor agent made a .md file for every single prompt they asked the machine. Know what you want, and if something seems to be taking too long or seems overcomplicated, it probably is. It’s good to prompt Cursor at the beginner discussing what you are trying to build, and request for simplicity over complication.

On The Class

In some of the theory parts of class, we used Google Colab. This worked really well since we didn’t run into any dependency issues that you might face with something like Jupyter Notebook. Even with the non-group work, we were getting hands on computer.

Last year, we discussed HTML. This year, we substituted that lesson for teaching deployment. That was a good swap of information, and with technology moving so quickly, I expect more changes to be made next year.

Virtual environments are slightly difficult to explain, especially as to why they are a best practice. That said, it was well worth it to help people quickly set one up for each project to minimize dependency issues, and basically give a one-minute explanation of why they are important, even if people don’t understand them 100%. It’s better for them simply to know that they are a best practice and less on why…at least for this class.

The five-day, intensive curriculum works amazingly well for this type of class. Lack of momentum is the biggest killer of technology projects - keeping everyone together for five days, working consistently toward a common goal created the best outputs. Dedication & Momentum > Time

To me, this is what AI education should look like - there is theory, but more than that, there is real world, project based work that pushes people to create. Creation is the top-level cognitive work, and more classes should focus their curriculum on analyzing, evaluating, and creating rather than remembering, understanding, and applying.

Bloom’s Taxonomy is a six-level hierarchy used to classify learning objectives by their cognitive complexity, moving from foundational memorization to higher-order critical thinking. Creation is the highest cognitive level because it requires students to synthesize all previous levels to produce original work

On the Bigger Picture

The reason this class is so beneficial is that everyone knows AI can help, but few know how to make it work for them. People need a little push to get from 0 to 1. This class teaches you, through both theory and project based work, how AI can help you build. Many students have taken their group projects and continued to work on them after the class or even created new products on their own time. Continue to learn more and more, just by doing.

The apps built this year were technically better than last year, and next year will be dramatically better than this year’s. That said, I still remember some stand out applications from last year’s class that I would use over some of today’s more recent applications.

Building, deploying, even databasing - that is more accessible than ever. But troubleshooting, knowing what you’re prompting toward, knowing what to build and why - that is still on us, and it makes all the difference.

The tools and execution got easier. The thinking did not. And that’s what we should continue to work on every single day.

Special thanks to Professor Johar and my trusty TA partners in crime Chitipat and Xilin. Wouldn’t have been as informative (and fun) without you

One of the great things about living in New York is that the Times is technically your local paper. It’s pretty cool to have one of the most prestigious newspapers in the world to provide you all the neighborhood hot spots, gossip, and restaurants. That said, opening the NYT can sometimes be slightly distracting…national politics, international news, and more flood the homepage when all I’m trying to see is where to get the best bagel in UWS.

I wanted to personalize my reading experience just a bit more - this way, I could still read through the great work of the NYT while focusing on my interests: local NYC food, culture, and news mixed with just a bit of business and tech coverage I need for work.

I decided to build something that did just that. Using the NYT API, pythonanywhere, Gmail App Passwords, and trial-and-error debugging, I built a personalized daily newsletter that sends me one NYT article each morning—automatically curated based on the day of the week.

Here is how it all works together:

And here is the type of article I receive each morning:

• Monday: Top national business story
• Tuesday: Top national tech story
• Wednesday: NYC Food and Dining article
• Thursday: NYC Arts and Culture article
• Friday: NYC Metro News Article
• Saturday: Top Op-Ed story
• Sunday: Most viewed article of the week

Could I just open the Times and scroll? Sure—and I still do. But I find it beneficial to start the day with one focused article rather than scanning endless headlines - One article, one topic, delivered to my inbox each morning, and maybe I explore the Times further from there. Plus, I wanted to build something I thought I would actually use — and a personalized newsletter seemed like a good excuse to get hands-on.

That is my WHY…I now had to think through the HOW. I could identify three primary pieces that would need to fit together in order to build this out:

1. I would need to use the NYT APIs to get the articles I wanted to read
2. I would need to work with Gmail to receive those articles in my inbox
3. I would need some level of cloud compute to automate this process for me

The NYT APIs

Getting started with the NYT APIs was pretty easy. You sign up for a nyt developer account and create an app. In my case, the app was the NYT daily newsletter. Your app will provide you your API key, which you can use for any of the numerous APIs the NYT offers.

A few of these APIs are deprecated, and some aren’t super relevant - while every use case is different, the most relevant APIs for most projects are likely the following (which I used for my project):

• Article Search API: By far the most well-documented and robust NYT API. Used for NYC-focused days (Wed/Thu/Fri) since it supports location filtering
• Top Stories API: A very basic API that returns an array of articles currently on a specified section. Used for national news to get current business/tech/opinion pieces
• Most Popular API: Pulls the most popular article on NYT for a specified time period. You can define ‘most popular’ based on most emailed, most shared, or most viewed.

Article Search

The Article Search API was my choice for NYC-specific content. It’s the most robust of the three, with extensive filtering options for location, desk, and date ranges.

For my Wednesday, Thursday, and Friday articles, the config looks like this:

# 7-Day Rotation Schedule

WEEKLY_SCHEDULE = {
    2: {  # Wednesday
        "name": "NYC Food & Dining",
        "api": "article_search",
        "filter": 'desk:"Dining" AND timesTag.location:"New York City"',
        "description": "NYC restaurant reviews and food scene"
    },
    3: {  # Thursday
        "name": "NYC Arts & Culture",
        "api": "article_search",
        "filter": 'desk:"Culture" AND timesTag.location:"New York City"',
        "description": "Theater, museums, and cultural events"
    },
    4: {  # Friday
        "name": "NYC Metro News",
        "api": "article_search",
        "filter": 'desk:"Metro"',
        "description": "Local NYC news and community stories"
    },
}

The filter parameter is crucial—it determines what articles get returned. But first, I had to get it working…

My initial attempt with Claude-generated code returned zero articles. Every single query came back empty. I knew this was wrong — there had to be NYC dining and culture articles. So I went back through my conversation with Claude and noticed something I’d initially overlooked:

Claude couldn’t access the actual NYT developer documentation. The site blocked it (likely to prevent AI scraping of their content). Instead, Claude relied on older, incomplete sources. I pulled up the documentation myself and compared it line-by-line with the generated code. There it was:

# What Claude generated (wrong):
'fq': 'news_desk:"Metro"' # returns 0 articles

# What actually works
'fq': 'desk:"Metro"' # Returns 10,000+ articles

The field name was desk, not news_desk. One word. Completely broke the query.

The lesson learned here: AI can generate code quickly and often accurately, but it still hits walls — especially with documentation behind paywalls or anti-scraping protections. You need to understand enough to debug when things don’t work as expected. I only caught this because I knew to check the actual API response structure.

Once that was fixed, the next challenge: filtering for NYC content specifically.

I tried several approaches:

• section_name:"New York" → 0 results
• glocations:"New York City" → 0 results
• geo_facet:"New York" → 0 results

Finally, buried in the filter query examples (not the main field documentation), I found:

timesTag.location:"New York City"

That unlocked everything:

• Dining + NYC: 1,251 articles ✓
• Culture + NYC: 977 articles ✓
• Metro desk: 10,000+ articles ✓

Once those were resolved, it was off to the races. The article search API lets you filter by so much more…I just tailored it to my needs. It would be easy for others to tinker with those parameters to find articles they were interested in.

Top Stories

For all my non-local articles, I just wanted to know what the top stories were for industries I am interested in - in this case, that is business and tech (there are many more options than just that). What I didn’t originally understand, however, was what ‘top story’ meant. The documentation defines it as this…

“The Top Stories API returns an array of articles currently on the specified section (arts, business, …).”

What does that mean? Currently on the specified section of what? And where on that section?

In order to find this out , I tested two things:

1. Run the API and see what it returns
2. Look at that particular section on the nytimes.com and see where those API articles are.

To do that, I run the top stories for business:

curl "https://api.nytimes.com/svc/topstories/v2/arts.json?api-key=mykey"

which returns the following:

While there are many more articles listed, the api returned two that we can see without scrolling down:

• ‘Zootopia 2’ Goes Wild at the Worldwide Box Office
• Two Retail Chiefs Take Stock of a Make-or-Break Holiday Shopping Season

So now it was time to find these articles on nytimes.com, which was relatively easy. I went to the business section and found the most recent articles.

So, by the documentation’s description, it is exactly this - “articles currently on the specified section.” I expected “Top Stories” to mean something more…maybe front-page articles or the most prominent story on the page (like that “Silicon Valley’s Man in the White House” article with the big photo). Instead, it’s just the most recent articles published under “Latest” in that section.

Since I don’t want ALL the articles and just want one, I will simply pull the top story listed for that particular section at 8am, when the task is executed (in the example above, the email article sent to my inbox would be the one about zootopia 2). Does Zootopia 2 really qualify as a ‘top business story’? Not really — it’s no more important than the retail chiefs article below it. But they’re both recent, relevant business coverage. And that’s the key point: this API gives me current, editorially-vetted articles in my chosen topics. It’s not perfect, but it works.

WEEKLY_SCHEDULE = {
    0: {  # Monday
        "name": "Business News",
        "api": "top_stories",
        "section": "business",
        "description": "Latest breaking business news"
    },
    1: {  # Tuesday
        "name": "Technology",
        "api": "top_stories",
        "section": "technology",
        "description": "Latest tech industry news"
    },
    5: {  # Saturday
        "name": "Opinion",
        "api": "top_stories",
        "section": "opinion",
        "description": "Top opinion and editorial pieces"
    }
}

This API is designed to return an array of stories, not just one. Ideally, I’d use the Most Popular API filtered by section — but that API doesn’t support section filtering. The Top Stories API was my best option for getting current business and tech articles. Good enough beats perfect when you’re trying to ship something that works.

Most Popular

I wanted to add the most popular story in a given week to my reading repetoire just because I want to see what everyone else is reading. This API is, by far, the most straightforward of the bunch - what is the most popular news story within the last 7 days. Since ‘most popular’ can be defined as most shared, most emailed, or most viewed, I had to decide what I wanted for my newsletter. While most shared and most emailed might be better for understanding engagement, I was simply trying to read what everyone else was reading. For that reason, I used the parameter of most viewed for this API call.

In this example call, you can notice how it says ‘viewed’ in the parameters. That could’ve been swapped with ‘shared’ or ‘emailed.’

curl "https://api.nytimes.com/svc/mostpopular/v2/viewed/1.json?api-key=yourkey"

Connecting the APIs to my Gmail inbox

While knowing how to call the NYT APIs and get the articles I want is cool, to make this application actually useful, I want to connect that data to something. I don’t want to just run this code and get the article in my IDE output. In this instance, I want to connect the articles called from the NYT APIs to my Gmail inbox.

In order to allow our API code to connect to my inbox, we need to use the Gmail SMTP, which allows emails to travel through Google’s servers. SMTP stands for standard mail transfer protocol - it’s how all emails travel across the internet. In our case, we want the emails we create (the email with NYT articles) to be sent to our inbox.

We connect our code to Gmail through Gmail App Password. You create a password just for this particular code. You can then plug in that information to the code below, that sets up the SMTP. Think of this like google creates a password just for this project , and then you plug that password into the code.

# Connect to Gmail's SMTP server
server = smtplib.SMTP_SSL('smtp.gmail.com', 465)

# Log in with your email and app password
server.login(EMAIL_FROM, EMAIL_PASSWORD)

# Send the email
server.send_message(msg)

Since SMTP is how emails travel, Gmail provides an SMTP server you can use, and App Passwords let you connect securely without exposing your real Google password.

This is the same flow shown in the architecture diagram at the beginning of this article — the Python script bridges the NYT APIs and your inbox, formatting article data into readable HTML emails.

Deployment

So we have now written code that calls the NYT articles I want, formats it as an email, and set up the connection that allows those emails to be sent to inbox. But we’re not quite done.

If I let this as is, I would need to get on my computer every day, run this script, and then the article will be sent to my email. While that is cool, I don’t want to do this manually every day. I want an automated email sent to my inbox, similar to email newsletters you might sign up for elsewhere.

We are going to do this by deploying our script to the internet. Our code currently runs on our local computer, the one we are literally using. But, if we instead put this script on another computer, perhaps one that was ALWAYS on and one that could run certain things at certain times, that could help us run this script automatically.

This is what deployment does - you upload your code somewhere and another computer runs it. That is how applications are always running - we might code them on our computers, but that code is uploaded to computers (often time hundreds, if not thousands of computers) somewhere else in the world. When you hear of cloud companies like Microsoft , Amazon, and Google, this is a core part of their business - they host your applications in the ‘cloud’ - which really just means they host your stuff on their servers somewhere.

For this application, pythonanywhere worked really well - pythonanywhere’s free version gets me 512mb of storage, one daily cron job, and enough compute to run my script, which is all I need. I’d recommend pythonanywhere for anyone that has a lightweight application that doesn’t need more than one automated task a day.

What I Learned

Could this be built with no-code tools (Loveable, Replit) or a different tech stack? Absolutely. But building it myself taught me things I wouldn’t have learned otherwise:

1. AI has blind spots — you still need to know fundamentals to catch them

Claude couldn’t access the NYT documentation and confidently gave me broken code (news_desk instead of desk). I only caught it because I understood how to read API documentation and debug when things didn’t work.

Even as AI gets better, this won’t change: you need to understand how systems work. AI can generate code, but if you can’t read documentation or debug when things break, you’re stuck.

2. Architecting solutions is still your job

AI can implement once you tell it what to build. But I still had to decide:

• Which of three APIs to use for each day
• That timesTag.location works when other location fields don’t
• How to connect Gmail SMTP to my script
• Where to deploy and how to automate it

Knowing how pieces fit together and making those architectural decisions — that’s what makes you effective with AI. You need to have the strategic vision and decision-making skills, but AI executes. (My old coworker Dan Giannone has a really good post about this thought you can find here)

3. Hands-on keeps you sharp and up-to-date

Being hands-on keeps you on top of the latest tech stack, which helps with the point above. I learned PythonAnywhere is perfect for lightweight automation. I learned about API rate limits the hard way. I learned the quirks of each NYT API. Each project teaches you new tools and patterns that transfer to the next one.

The best way to learn is to build. Even if it’s “just” a daily email newsletter, you’ll learn something valuable in the process.

“I understand that ChatGPT is good for coding but I don’t actually get how people do anything with it” That stuck with me - it was a good reminder that even among those involved in business and tech, the real power of ChatGPT, Claude, Gemini, and others isn’t always obvious. Besides making ghiblis or doing your homework, there’s one benefit I find extraordinary - lowering the barrier to code creation. In its simplest form, vibecoding is just a way that AI makes coding accessible to everyone. Vibing opens the door for everyone to build stuff. And that’s the beauty of it. It lets anyone with an idea simply vibe with AI and end up with something real.

“I’m a real SAAS business application!”

Vibing Origins

Back in early 2025, Andrej Karpathy coined the term “vibecoding” in this tweet. I think this is pretty much correct definition, although people have their unique spin on it. The way I see it, vibecoding is simply using AI to help you code. No Computer Science degree required. No bootcamp. No months of learning syntax. Just you, your idea, and an AI tool.

AI tools like ChatGPT, Claude, and Gemini have gotten so good at writing code that they can build entire things for you. A game. A website. A tool that solves your specific problem. You describe what you want in plain English, and it codes it. That’s the whole magic trick.

DALL-E generated image of “Two businesss school students discussing AI on a sunny day in New York City’s Upper West Side, using a cartoon art style”

A general walkthrough of vibecoding

I am not a programmer by nature myself, but vibecoding has helped me build more than I ever imagined I could have. It helped me build this game, or even this very website. It helped my non-technical friend build his website, and helped countless others build to their hearts’ desires.

It’s not only the simple stuff either. I’ve seen my friend, a former educator, build an application that helps teachers build out entire lesson plans and associated slides. I’ve seen another friend use vibecoding to help him build smart home applications. None of these are earth-shattering, but that doesn’t matter. What matters is that people ideas and they made them real.

If you haven’t tried vibecoding yet, here’s the basic flow:

1. You pick an idea for something to build. It can truly be anything, but the recommendation is to always start small (build a tic tac toe game if you don’t have any better ideas)
2. You open an AI tool like ChatGPT or Claude and describe what you want - “build me a tic tac toe game” or “create a tool that converts temperatures.”
3. The AI gives you code. I typically like to ask the tool to explain what it’s building so you can learn as you go.
4. You paste that code into files on your computer using a tool called an IDE (like VS Code or Cursor – these are just apps where you write and organize code…for example, you’re not pasting code into Microsoft Word, you make or paste it in an IDE).
5. Then you run it and see what you made. If something doesn’t work or you want to change it, you just tell the AI and it fixes it.
6. If you want others to use what you built, you can deploy it online with tools like Netlify or Vercel or any cloud service like Azure or AWS – this can be complicated sometimes, but the AI walks you through this.

That’s it. The AI handles the technical complexity and you handle the strategy and vision. Vibecoding empowers those with drive and vision to create.

DALL-E generated image of “someone vibecoding and creating a game that they are sharing with the world, using a fun and vibrant art style

To get started, I’d literally recommend opening ChatGPT or Claude and typing:

“I’ve never coded before and want to build something simple to get started - maybe a tic tac toe game or [your idea here]. Walk me through the entire process step by step: what tools I need, how to write and organize the code, and how to actually run it on my computer so I can see it work. I want to learn as we go, not just copy and paste. And let me know that I can ask you to change things once it’s working.”

Then just follow along. It seems like a lot the first time, but once you do it once, you’ll realize how straightforward it actually is. And like anything else, the more you do, the more you learn.

The broader point

Does this mean we should vibecode everything? Definitely not. AI still is not as good as our best programmers at producing code. It may produce a lot of it, but it doesn’t mean it’s always right. AI often overcomplicates really simple tasks and that can lead to a plethora of issues. I absolutely do not want AI (at least right now) coding critical software that people’s lives might depend on. AI models are nothing more than a prediction tool and sometimes those predictions can go horribly wrong. For now, and possibly forever, there is always a place for programmers.

However, as you might’ve been able to decipher from this post, I think vibecoding is incredible. It helped me create so much more than I ever could have imagined. It can help you do the same. And I believe creation - building things - is a net positive for society as a whole.

We should build more things. Vibecoding is democratizing that ability like never before. Before, if you had an idea for a simple game or a productivity tool, you had two options: spend months or years learning to code or pay someone else to do it for you. Now, you can sit down on a Saturday afternoon, vibe along with an AI tool, and walk away with something real you created.

This, to me, is more than just individual empowerment, although that is incredible as well. It’s about unleashing human creativity at scale. When more people can build things, we get more solutions to problems we didn’t even know we had before. We get more tools, more games, more experiments, more applications. We get someone who’s sat on a business idea for years finally bringing it to life. We get someone making their first app and having the same incredible feeling of seeing a finished built product as they did when they used to build legos (oh wait that’s me).

Not everyone needs to vibecode nor does everyone need to build anything. But the fact that we can is pretty remarkable. It’s turned this exclusive club into an open sandbox, and everyone is invited to start building sandcastles.

Look at this little man building this awesome sandcastle…needs a moat, though

Every year, my buddy Drew K. and I run through our division, record, and playoff predictions for the NFL year, as well as a bold prediction for each divison. Here we go!

AFC East

This is, by FAR, the easiest division winner to guess. The Buffalo Bills will, for the 6th year in a row, roll through the AFC East. The Bills return a strong core, have signiicant offensive weapons, and have the leagues reigning MVP Josh Allen. I think the Patriots are much improved , but the line play around second-year play caller Drake Maye still isn’t good enough. Additionally, the receiver core is still bad even with the addition of an aging Stefon Diggs. That said, watch for TreVeyon Henderson to become a household name by mid-year, especially for Fantasy owners. I think he will be the first offensive player with real pop (besides Maye) that the Pats have had in quite awhile. I think the Jets will improve as well, but Aaron Glenn’s first goal should be to change the culture of that team. I don’t expect more than 6 wins there. Lastly, the Dolphins are a wreck - coach on the hot seat, Tyreek disgruntled, Tua one concussion away from retiring, and it seems like half their players are always on IR. At least they are fast.

1. Buffalo Bills: 14-3 (*Z - First Round Bye)
2. New England Patriots: 9-8
3. New York Jets: 6-11
4. Miami Dolphins: 3-14

Bold Prediction: Is it even a bold prediction to predict Tyreek Hill is not on this team next year? What about halfway through the year? He hates being there. It was clear from end of last year and it’ll be clear halfway through this year. Too bad the market for his has dramatically shrunk.

miss you babe

AFC North

One of the more interesting and entertaining divisions in football, not just this year, but for many years in a row now. You know what isn’t interesting? Aaron Rodgers to the Steelers. I don’t care if it gets the headlines. Do I trust a grumpy 41 year old to mesh well with a buttoned up organization and a coach who runs the show in Pittsburgh? I do not! The Steelers have an AWESOME defense, but I expect some kind of implosion for this team. The Browns are dog sh*t and that’s all I will say there. The Ravens I think are one of the best built rosters in football, even slightly better than the Bills because I think the Ravens defense is better, and I think Harbaugh is just a bit better than McDermott. Also, probably-should-have-been-MVP-last-year Lamar Jackson is their QB. Ravens will be your division winner. The Bengals are the most fascinating since I think their offense will be absolutely insane, half in part because of the three headed monster in Burrow, Higgins, and Chase, and the other half in part due to their defense will be subpar yet again.

1. Baltimore Ravens: 13-4 (X - Division Winner)
2. Cincinnati Bengals: 10-7 (Y - Wild Card)
3. Pittsburgh Steelers: 8-9
4. Cleveland Browns: 2-15

Bold Prediction: As you can see from the record predictions, this will be the first year in X years that Tomlin does not have a winning record. “But Joe…he’s done it every year sinc-“ I DON’T CARE!!!! I find this to be somewhat of the hot hand fallacy - just because he did it last year doesn’t mean he will do it again. This isn’t like people picking the Patriots of old or Chiefs to win all the time, because they have / had proven infrastructure (top QB and HC) to do it. People just pick the Steelers to go above .500 because of Tomlin. Nah…I don’t buy it.

AFC West

Anyone that is choosing someone as divison winner that is not the Chiefs is out of their mind or trying to garner attention to themselves. They are proven winners and will be my pick for division winner every year until they simply aren’t, as long as they have the best / near-best QB and best HC in the game. That alone is the top recipe for success in the NFL. I like what Harbaugh is doing with the Chargers culture but they still don’t have enough weapons offensively to make any major breakthroughs. I’m not gunna dunk on Herbert like a bunch of other people, especially because he’s still young, but I do think he needs to step up soon here. The Broncos are everyone’s sneaky dark-horse team but I think they will probably get as far as last year, which is lose in the Wild Card. I think they will be BETTER, especially with a killer defense and more mature QB, but I still haven’t seen enough out of Bo Nix to make me think they will be great. Also, besides Sutton, it’s not like they have much to play with on offense. Lastly, the Raiders are fun with an always energetic Pete Carroll, improved QB in Geno, and exciting RB rookie in Jeanty, but let’s not pretend they will be good. Another team with more of a culture shift goal more than anything else. Do your magic Raiders-minority-owner-also-somehow-lead-FOX-commentator Tom Brady!!

1. Kansas City Chiefs: 12-5 (X)
2. Denver Broncos: 10-7 (Y)
3. Los Angeles Chargers: 9-8
4. Las Vegas Raiders: 6-11

Bold Prediction: Mahomes throws under 3500 yards with less than 30 TDs. I don’t really think it’s because Mahomes will take a step back..I actually think he could be a bit better under an improved (hopefully?) O-line. I just don’t think the scheme will ask that much of him. The defense is nasty and KC has a way to win low these days.

AFC South

Not going to cover this division too much because I don’t find it particularly exciting. The Texans will probably win the division because of a nasty defense and likely-improved CJ Stroud over last year’s sophomore slump. I do find Travis Hunter very exciting , but I don’t think he will revolutionize the Jags, especially in year. He will probably be a decent-good WR, and possibly a decent 3rd down CB, but nothing outstanding on either side of the ball. I think this is due to the fact I never played football, but I just think dominating players on either side of the ball will be more entertaining than Hunter who could play both sides. Maybe I’m a hater, and I promise I do think it’s really cool, but I just don’t get the impact he will make if no one thinks he will dominate on offense or defense, compared to someone in the MLB like Shoehei who can do both but also dominates batting. That all said, look for the Jags to be a sneaky wild card contender this season since I think Lawrence improves with new coaching. Indy won’t be any good until they figure out their QB situation. The Titans are all about Cam Ward and helping their first year quarterback. Their offensive line is improved and I think Ward might go on a “I’ll show you that I’m still the most important player of this draft” tour. Gimme the Titans as the surprise AFC playoff team. You know what? I actually do think this division is exciting, just not any good.

1. Houston Texans: 10-7 (X)
2. Tennessee Titoooons: 9-8 (Y)
3. Jacksonville Jaguars: 8-9
4. Indianapolis Colts: 4-13

Bold Prediction: Cam Ward wins OROY. Why is this a bold prediction if he’s the first overall pick from last draft? Because literally no one is talking about this guy headed into the season. I swear I hear more Trevyon Henderson OROY picks than Ward. Gimme the guy that is built like a QB, throws like a QB, and has a few weapons (ridley, pollard, O-line) that will help him out this season.

NFC East

Fly Eagles, fly. It’s all about the Eagles, your eventual division champs. I feel like people say this every year but I think this team has a legit chance to go back to back. They bring back much of their coaching core, playing core, and have probably the best built roster in the NFL. Hurts (as we saw last night against the Cowboys) can get it done when it matters. The Commanders were a great story in 2024, and I have a lot of faith in Jayden Daniels to be a great NFL QB, but I think they take a step back this year while still making the playoffs. Their offense still rocks but I worry about the typical sophomore slump for the QB, a bad defense, as well as a simple regresssion to the mean. I must admit, I wasn’t super high on the cowboys coming into the year, but I thought they looked pretty good last night even thought they didn’t win. Dak looked sharp (better than Jalen IMO), the defense tightened up in the second half, and his weapons got open even if CeeDee dropped, like, 18 passes. Finally, the Giants won’t be good, but I do think their Defensive Line with X, Y, Z has a chance to be the best unit in ALL OF FOOTBALL! Look for them to wreck a game or two.

1. Philadelphia Eagles 14-3 (Z)
2. Washington Commanders: 10-7 (Y)
3. Dallas Cowboys: 10-7
4. New York Giants: 5-12

Bold Prediction: Dak will be in the MVP conversation in the final weeks of the season. I just think their defense isn’t good enough to win on its own and he will be asked to do a lot. Having CeeDee and George Pickens will help him air it out and expect some monster stats.

NFC North

I think this is a loaded division and is my choice for best in football, edging out the AFC West. There are three legitimate contenders in the Lions, Packers, and Minnesota (although I believe in them less). Chicago should be an upgrade from last year with the addition of Ben Johnson and hopefully he can help Caleb Williams to a better season two. Even though there is some turnover in Detroit, with both coordinators leaving, they still leave in place an awesome core of players and a coach that has completely turned around their franchise. The Lions won the division last year with a slew of injuries, and with everyone returning, I think Detroit is your division winner. The Vikings 52 man roster, sans their QB, might be the best in football, but you’re gunna have show me JJ is any good before I make them true SB contenders. I think the Packers could have a legit shot at taking home the Lombardi this year. I think Jordan Love is a good QB, you have a great coach, and an AWESOME new game-changing player in Micah Parsons. I have other worries about their defense, but if the D can round into form, be on the lookout for the Packers. Lastly, the Bears will be fun, and that’s what should matter for the lovely people of Chicago. They still have too many holes in their roster and I doubt they make the playoffs, but expect to see a much more competent team this time around.

1. Detroit Lions: 11-6 (X)
2. Green Bay Packers: 10-7 (Y)
3. Minnesota Vikings: 9-8
4. Chicago Bears: 7-10

Bold Prediction: Caleb Williams has 30+ total TD’s (running and passing) and goes for over 4000 yards. I think Ben Johnson unlocks Caleb with a better run game, unlocking the play action pass or option for Williams, and better play calling.

NFC West

I think this is a sneaky good division this year. My guess is a re-tooled and uninjured (cmon CMC you can make it) 49ers will win the division this year, even if they lean on the youngsters moreso in the 2025 campaign. The big reason for the improvement will come from new DC Robert Salah, who led the defensive rebuild for SF in 2017 and will lead a ‘mini-rebuild’ again in 2025. The Rams will be a great team IF Stafford doesn’t have major injury concerns throughout the year. Their defense should actually be pretty good this year. Even though Kupp is out, Davante is in, and the offense should be strong enough to hang with the big boys (although where’s the run game?). The Seahawks I think will be a tough out for anyone - they have an awesome defense and an offense that can be good enough if Darnold plays anything like his regular season form from last year. Just depends if the O-line isn’t ass. Lastly, the Cardinals could be exciting on offense and will have a better defense, and have steadily improved every year under head coach Gannon, but I just can’t get behind Kyler Murray. He hasn’t been consistent enough to make me believe.

1. San Francisco 49ers: 10-7 (X)
2. Los Angeles Rams: 10-7 (Y)
3. Seattle Seahawks: 9-8
4. Arizona Cardinals: 6-11

Bold Prediction: While I don’t foresee the Rams winning the superbowl , if they do (I do believe they have an outside chance here), McVay retires. He has mentioned retirement in the past, and unlike some other coaches who lead teams that have a shot at winning it all, I don’t think his team as set up as well as others to continue their success (i.e. Andy Reid). It might be a good time to step aside if he is serious about hanging it up…at least for now.

NFC South

Probably the least exciting division for like…the 500th year in a row…so I won’t spend much time here. I think the Bucs will win the division and can be a legit good team that can make a deep playoff run if Baker continues his elevated level of play. Also, I think it’s time we think of Baker as a legit top 10 QB. They are just a sound team - sound QB, sound offense and defense, sound coaching. The Falcons have IU LEGEND MICHAEL PENIX JR. who will be sure to air out the ball often this year. They have an exciting offense with London and Robinson but their defense isn’t good enough to compete with the best of the best. The Panthers can be better if Bryce Young continues his positive play from the end of last year, but like the Falcons, they have a defense that just isn’t good enough for anything more than helping the team hover around .500. Lastly, the Saints will just be bad, which is too bad because they have great fans. The QB situation is a MESS and hopefully HC Kellen Moore can bring some juice on the offensive side of the ball. Unfortunately, expect the Saints to compete for the 2026 number one draft pick.

1. Tampa Bay Buccaneers: 10-7 (X)
2. Atlanta Falcons: 8-9
3. Carolina Panthers: 7-10
4. New Orleans Saints: 3-14

Bold Prediction: Bucs to the NFC Championship game. Baker has balls, they have great weapons in Godwin and Evans, and I think Bowles is a good enough coach and can definitely coach a great defense. They will prey on easy wins in the division, but I don’t think the team is built to “get soft” from a soft schedule and shit their pants in the playoffs against good competition. You only need to get hot in the playoffs…I think the Bucs can do it.

Playoff Teams

AFC

1. Bills
2. Ravens
3. Chiefs
4. Texans
5. Bengals
6. Broncos
7. Titans

NFC

1. Eagles
2. Lions
3. 49ers
4. Bucs
5. Commanders
6. Packers
7. Rams

Awards

These awards kind of dumb IMO so I’m going to keep this to a minimum:

• MVP: Joe Burrow - He will have to air it out this year and will put up monster numbers. I think they can be a bit better defensively and will be a good team that makes the playoffs. I’m banking on a bit of “voter fatigue” here with Mahomes, Jackson, and Allen having all already won.
• Comeback player of the year: I’m actually just going to use this section to complain about how I’m still baffled Damar Hamlin didn’t win this award a few years ago. These awards are all narrative driven and the man literally DIED on the field and came back to play. You know how much mental and physical fortitude that takes? Congrats on voting Joe Flacco for that award who got shit on in the playoffs. Just remember we gave MVP to a kicker in the 80s so don’t tell me its purely on pure football player acumen. Christ that was so dumb.

Predictions that could be laughably incorrect

Many, but specifically…

1. Eagles not making the NFC championship game when I could easily see them repeating as SB champs. All for the clicks baby!!
2. Titans making the playoffs and then they turn out to be the worst team in the NFL

SB Prediction

Bills over Packers. I have reservations about both of these teams - I think the Ravens are generally better than the Bills and I didn’t even choose the Packers to win their division. That said, I just think the Bills are finally ready for this moment. I think it’s important that they get the number one seed this year, and when they play the Ravens in Buffalo in January, I’ll take Allen and company over Lamar in that situation. For the Packers, sometimes you just gotta roll on vibes. This is a young team I can see getting hot in the playoffs and playing without fear. If Love can pick up where he left off at the end of the season TWO seasons ago, they will be dangerous.

Between the Bills and Packers, Bills are just the better team. Their defense will finally show up in the playoffs and it’s their time to shine. Let the confetti rain down for a fanbase starving for a championship. Circle the wagons, baby.

Gemini: *Unfortunately, I’m unable to create images that depict real people [Josh Allen and the Bills] in a way that could be seen as misrepresenting actual events [me asking Gemini to make an image of the bills winning the SB] or endorsing outcomes of real-world competitions. I can generate an image of a generic football team in blue and red, celebrating a championship win with a trophy, and a player who resembles a quarterback celebrating. How does that sound?

I’m excited to launch this new space where I’ll be sharing my thoughts on topics I’m passionate about. You’ll find blogs on tech, economics, and more.

What You Can Expect

I’ve had the opportunity to work across different industries - from startups to consulting to big tech - and I try and use this experience to simplify complicated topics and bridge the gap between technical and non-technical people.

Additionally, I’ve spent some time as a group fitness instructor and a LOT of time watching sports, so every now I’ll write about those topics as well. Go Hoosiers.

With another debt ceiling ‘crisis’ looming, the time is ripe for a plethora of economists, journalists, and twitter-goers to opine on the issue. I say ‘crisis’ because the debt ceiling is more of a politically-charged gimmick than anything else, but I don’t want to focus on that right now. Instead, I want to spend some time outlining an unconventional solution to the debt ceiling that has gained traction over the past few years: minting a trillion dollar coin. On Twitter, you’ll see this commonly referred to as #MintTheCoin.

What is the debt ceiling and the issue it creates?

The government of the United States is like a big family. Just like how a family has to pay bills and taxes, the government also has to pay its bills and taxes. But sometimes, the government doesn’t have enough money to pay all its bills, so it has to borrow money from other countries. That borrowing limit is called the “debt ceiling.”

To avoid the debt ceiling, or the need to borrow more money, there’s an idea of creating a new coin that’s worth a lot of money. That coin is called the “trillion dollar coin.” The US Treasury would create new money out of thin air by depositing a platinum (yes, the platinum part is important) coin with a face value of $1 trillion at the Federal Reserve. This would increase the amount of money the government has, and thus, it would be able to pay its bills without borrowing more money.

DALL-E generated image of “Digital Art of the United States Government making a trillion dollar coin.” This isn’t close to what the coin would look like. It has to be platinum, for starters.

How minting the physical coin would work:

1. The order would be called by the Secretary of Treasury before the debt ceiling is hit.
2. A platinum coin would be minted after it is struck at West Point.
3. After being minted, it would theoretically* be accepted by the Fed.

*we will get into this a bit more below.

1. At this point, you can melt the coin, put it in a museum, or put it on a rocket ship to Mars. Depends how creative the Fed is feeling that day.

A DALL-E generated photo of “Jerome Powell sending a coin to another planet”

What’s the problem?

So we’ve established that a) we can actually do it and b) it would avoid our debt ceiling issue. So what’s the problem?

There are a few criticisms of this solution. The ones I most commonly see thrown around are that it would cause inflation and that it’s too gimmicky.

I find both of these criticisms not terribly concerning.

Regarding inflation

The coin would just be used to pay off debts as if the ceiling was raised, which would ultimately happen anyway. It’s not directly injecting one trillion dollars into the economy. Furthermore, as Josh Barro points out in his newsletter…

…the coin would not go into circulation - Treasury would deposit it with the Fed. Then, the idea is that, as Treasury used its new cash stockpile to pay the government’s bills, the Fed would sell Treasury bonds into the open market, as a substitute for the bonds Treasury itself isn’t allowed to sell right now…In theory, this should all wash, with no effect on the money supply compared to a normal scenario where Treasury borrows and spend…

All of this states that we’re not really adjusting the money supply in any drastic way.

Concerns over inflation are always present. In 2008, when the Fed embarked in Quantitative Easing, a massive increase in the balance sheet to levels previously untested, there were fears that the increase in the money supply would cause rampant inflation and cripple the economy. But alas, we got a decade of record-low inflation because this money stayed in the financial system and avoided the real economy.

Regarding the ‘gimmick’ criticism

Well, so is the debt ceiling and the political infighting that surrounds it. So are some other solutions proposed to fix this issue, like premium bonds. The way we move around money in the government to temporarily avoid the debt ceiling is a gimmick. The whole thing is gimmicky, and I find this solution no less so.

Famed economist Paul Krugman agrees with my logic regarding these two points.

More serious concerns

That all said, I do think are other issues that are less often discussed but more serious:

It would politicize the Federal Reserve

If they even accepted the coin in the first place: As I mentioned earlier, in the situation where we do mint the coin, the Fed would accept said coin from the Treasury. The problem here is that the Fed, as it’s constructed, said it might not accept a trillion dollar coin. It doesn’t have to. And, even if they did end up accepting it, this would essentially be the Fed stating that they are participating in an administration’s effort to circumvent the House of Representatives in creating fiscal policy. They would help one party in power bypass what is supposed to be a bipartisan compromise. This would set a difficult precedent of using the Fed as a political tool moving forward, regardless of the party in power.

Economic expectations are too unpredictable

As described earlier, while inflation wouldn’t be an issue on paper, inflation is still a byproduct of expectations. If people believe there will be inflation, that can cause inflation, just or not. So if the market doesn’t believe debt would end up being issued, it would cause inflation expectations to rise, which would likely in turn cause inflation to go up. Furthermore, if this solution is publicly advertised, then the public would surely catch wind of ‘the government creating a trillion dollars out of thin air,’ which could also cause inflation expectations to rise. (FWIW, I think there are ways to prevent public expectations from rising, but no guarantees)

Conclusion

Ultimately, there are some very valid reasons to not move forward with minting the coin, and it’s probably best to not seriously consider this idea; however, as the United States’ political scene gets more and more polarized, the financial issues we face will get more and more challenging to resolve. So, if we’re going to need to get unconventional, don’t write off #minthecoin.

A Quick 2025 Update

Some time has passed since the publishing of this article, and this idea continues to resurface every time a debt ceiling crisis rolls around. From a policy perspective, not much has changed - this idea continues to be nothing more than a thought experiment.

Back in 2023, although I wouldn’t have endorsed this idea, I would say I was moderately intrigued and could at least entertain the conversation. Today, however, I figure it’s best we leave this idea behind. Minting the coin is too politically toxic in an already politicized country. We don’t need one party or another to try and take this non-traditional measure that bypasses Congress’s oversight of fiscal policy. Additionally, while actual liquidity may not be drastically affected, I’m far less confident that this idea would not cause inflation. I just think the optics of “creating $1T out of thin air” still carries too much potential to unsettle investors and stoke inflation fears. Inflation is all about expectations and credibility…the more non-traditional debt-avoidance routes we take, the less credibility we will have, which will likely increase inflation anyway.

Until we are really up against the wall (which I worry about more and more these days) I’d prefer if we never see this coin made. Save it for debate with your econ buddies.