Vibe Coding

All You Need?

“Learn to sell. Learn to build. If you can do both, you’ll be unstoppable.”

—Naval Ravikant, Entrepreneur and Investor

Its a tale as old as time. Jobs and Woznick. Bill Gates and Paul Allen. Reed Hastings and Marc Randolph. The list goes on.

In many of these cases, you had a outward facing CEO, who was concerned with vision, taste, and fund raising.

And you had a technical founder who excelled at creating a product roadmap, managing the technical aspects, and executing on the vision.

Long has there been a raging debate in tech about which is more important.

0% distribution x 10x product experience = 0% ARR

And yet, no amount of marketing will save a substandard product, once it gets into the hands of users.

Ideally you’d have both an exceptional product and exceptional distribution.

And now you can.

Vibe Coding

“There's a new kind of coding I call "vibe coding", where you fully give in to the vibes, embrace exponentials, and forget that the code even exists. It's possible because the LLMs (e.g. Cursor Composer w Sonnet) are getting too good. Also I just talk to Composer with SuperWhisper so I barely even touch the keyboard. I ask for the dumbest things like "decrease the padding on the sidebar by half" because I'm too lazy to find it. I "Accept All" always, I don't read the diffs anymore. When I get error messages I just copy paste them in with no comment, usually that fixes it. The code grows beyond my usual comprehension, I'd have to really read through it for a while. Sometimes the LLMs can't fix a bug so I just work around it or ask for random changes until it goes away. It's not too bad for throwaway weekend projects, but still quite amusing. I'm building a project or webapp, but it's not really coding - I just see stuff, say stuff, run stuff, and copy paste stuff, and it mostly works.”

—Andrej Karparthy, previous director of AI at Tesla, co-founder of OpenAI

Karparthy is one of the godfathers of the new AI era, right up there with Ilya Sutskever and Noam Shazeer.

And while what he describes may seem trivial to your average SDE, its actually a huge unlock for people who consider themselves “non-technical.”

Take me for example. I know in theory how an LLM works. I’ve seen people use Python to leverage solutions to business problems, so I know what’s possible when you apply a bit of code to something. But I do not consider myself technical at all.

And yet, I’ve started using SQL and Python every day in my role. How? Vibe coding.

It’s really quite simple. If I have several excel sheets that I want to combine into one sheet, or merge together to create a separate output file, I just ask Claude to write some Python for me.

If I have to query our database, as long as I know what table to query and what the column names are, Claude writes the SQL query for me. And I can ask for any parameters that I want.

An LLM is an API between humanity and machines. It translates natural language, the language we speak, into code, the language computers speak.

I tell Claude what the current state is, I attach the file I am trying to manipulate, and I tell it the input and output file paths.

It will write a full script for me in 30 seconds~, and if VS Code throws an error, I just put that error message back into Claude, and Claude will de-bug the script for me.

This process has already saved me hundreds of hours of manual effort and reduced my need on annoying excel formulas by at least 85% ( I think excel is a terrible product experience).

Before, when problems were painful, but not painful enough to warrant engineering effort, you just had to grit your teeth and work through the process.

Now you are your own engineer, and can automate your own processes.

The Lever the Moves the World

“Give me a lever long enough and a fulcrum on which to place it, and I shall move the world.”

—Archimedes, ancient Greek mathematician, physicist, engineer, astronomer, and inventor

I think its important to stipulate that the more complex the project, the less you should rely on vibe coding.

The danger is that Software Design Engineers will start to be overly reliant on vibe coding, build a product from the ground up without having any understanding of the underlying code, and if there is a dependency built in that wasn’t accounted for, it could create massive amounts of tech debt that may be impossible to undo.

But vibe coding is going to move mountains for those of us who are not SDEs, but who can leverage code to make our own lives easier.

In the same way that Chat GPT 4o is not better than a 10x writer, Cursor, Co-pilot and Claude are not a substitute for a 10x SDE.

But what it does do is allow the bottom quartile to accelerate their skills at a pace that just wasn’t possible five years ago.

Someone who has no technical ability, suddenly has the same technical ability as a mid level junior dev. And with the help of an LLM, anyone can code and execute basic projects to make their lives easier.

Code is the 21st century lever that moves the world, and now, everyone has access to it. Which had broad implications for the future.

A World without SaaS

Software as a service emerged in the early-to-mid 2000s to the solve the problem statement of verticalized solutions for non-technical people.

Need a database for your sales leads? Salesforce.

Want to build a website? Wix or Squarespace.

Want a program management software? Asana.

Need a centralized place for your business operations? Hubspot.

All these companies have market caps in the billions. Why? Because they coded from scratch verticalized solutions that solved business problems, and because of this convenience, they can charge a per seat monthly subscription.

But in an age of LLM assisted coding, companies may start to look at these numbers a bit more closely.

Here is a breakdown of what these services would cost for a modest department of 15 people per year:

So a small business, startup, or even a department in a bigger company, could spend $4,500 a year for Salesforce, or you could spend $216 a year for Claude Pro, and give your engineering team a quarterly goal to build your own internal CRM.

Then its yours. You built it. You own it. And maybe you can even license it. That’s how Slack was created.

Slack’s origin is one of the most iconic startup pivots in tech history. It began as an internal communication tool built by a company called Tiny Speck, founded by Stewart Butterfield, the same guy who co-founded Flickr.

Tiny Speck was developing an ambitious multiplayer online game called Glitch, which launched in 2011, but failed commercially despite a niche following. While building the game, the remote team struggled with coordination, so they created an internal tool to streamline communication, what would later become Slack.

After shutting down Glitch in 2012, Butterfield realized that the internal tool had real potential, and spun it out as a standalone product in 2013.

Slack officially launched in 2014 and quickly went viral thanks to its intuitive interface, channel-based communication, integrations with other tools, and searchable message logs.

It grew bottom-up, teams adopted it organically without top-down mandates, and within six months had over 15,000 daily active users.

By 2015, it was valued at $2.8 billion. In 2021, Salesforce acquired Slack for $27.7 billion, its largest acquisition at the time, positioning it as the centerpiece of its “digital HQ” strategy to rival Microsoft Teams.

Ironically, this wasn’t Butterfield’s first accidental success, Flickr was also born from a failed game project, making him a rare founder who turned two failed games into two billion-dollar software companies.

Now that we are in an age of code based augmentation, I would expect this phenomenon to become more common, as companies begin to build their own software stacks, rather than relying on off the shelf solutions from languishing SaaS companies.