Can I Make $250,000 in 90 Days With AI?

“Our deepest fear is not that we are inadequate. Our deepest fear is that we are powerful beyond measure. It is our light, not our darkness, that most frightens us. We ask ourselves, ‘Who am I to be brilliant, gorgeous, talented, fabulous?’ Actually, who are you not to be? You are a child of God. Your playing small does not serve the world. There is nothing enlightened about shrinking so that other people will not feel insecure around you. We are all meant to shine, as children do. We were born to make manifest the glory of God that is within us. It is not just in some of us; it is in everyone. And as we let our own light shine, we unconsciously give other people permission to do the same. As we are liberated from our own fear, our presence automatically liberates others.”

— Marianne Williamson, A Return to Love

I will not be the first person to attempt something that once would have seemed impossible.

The technology age has already produced founders and companies that have made $100 million or more in six months. If a solo founder can build a company valued at $250 million in roughly six months, perhaps a $250,000 challenge is not even large enough.

The public challenge is $250,000. Privately, I am pushing myself toward $1 million.

The point is not to make a casual prediction. It is to set a target large enough to force a real test of what these tools—and one person directing them—can now accomplish.

Perhaps we should all push ourselves harder than we think we can.

But the deeper point is that one person doing something extraordinary makes it imaginable for everyone else. It shows that the boundary has moved.

I want to shine a light on what more people can now attempt with these tools. Not because everyone will take the same path, but because anyone can learn to build, test, direct, and improve systems that once required an entire organisation.

I’m going to find out—and see how much one person can actually do now.

A few days ago, I came across a company called Polsia.

What caught my attention was not simply what it had built, but how quickly it had apparently built it.

Founder Ben Cera launched the company in late 2025. Within roughly six months, he had raised $30 million at a $250 million valuation and was publicly saying the company was approaching roughly $10 million in annualized revenue.

That is extraordinary.

But what interested me most was the product.

From the outside, Polsia did not appear to me to be based on some impossible technological breakthrough. Much of the value seemed to come from taking capabilities that increasingly exist inside systems such as Claude and making them usable for people who do not want to understand terminals, APIs, agents, software infrastructure, or how all the technical pieces fit together.

Calling it merely a Claude wrapper would be unfair.

Making complicated technology easy to use is itself valuable.

Even the name was part of the joke. Polsia is “AI slop” backwards. Cera leaned into that sensibility, building an Instagram account around deliberately absurd AI-generated videos. Some reached tens of millions of views. A link to Polsia in the bio converted that attention into traffic.

But that was exactly the lesson.

You do not necessarily need to invent the underlying technology. You need to understand what it makes possible and turn that into something useful.

That led me to a slightly uncomfortable question.

What exactly am I waiting for?

I have spent a lot of time studying AI.

Research.

Automation.

Vibe coding.

Agents.

Websites.

Databases.

Business models.

I have also been thinking about eventually creating a hands-on AI training business for experienced executives and entrepreneurs.

But before I teach anybody how to use these tools commercially, I want to see how far I can take them myself.

Not by watching demonstrations.

By building.

So I have set myself a challenge.

$250,000 in 90 days

My objective is to use AI, combined with the business experience I already have, to create $250,000 in new revenue over the next 90 days.

I do not know whether I will get there.

That is what makes the experiment interesting.

And I am going to do something else that would have sounded almost absurd a few years ago.

I am going to work on several projects simultaneously while continuing to run my existing business.

That may end up being the more important experiment.

I actually learned to program a very long time ago

There is one piece of background worth explaining.

I am not a software engineer, but I am also not completely unfamiliar with programming.

I learned BASIC when I was 13.

Later, in the 1980s, I wrote an options-pricing program in BASIC on Windows.

And then, essentially, I stopped.

For the better part of three decades, I did no serious programming.

Technology moved on.

Software development moved on.

The internet happened.

Cloud computing happened.

Modern databases happened.

Git happened.

APIs happened.

JavaScript frameworks happened.

Everything around software development became enormously more sophisticated.

I picked up a little Python here and there, but before this experiment I could not have sat down and built a serious Python project myself.

So when I say I have a technology background, that is what I mean.

I understand logic.

I understand systems.

I understand what software can do.

But modern software development had passed me by.

And then AI did something unexpected.

It brought me back in.

Except I am not really becoming a programmer

This is one of the strangest things I have discovered.

I am learning more and more about software development while allowing AI to write almost all of the code.

That sounds contradictory.

It is not.

Because increasingly, my job is not to type the code.

My job is to understand what the system should do.

How the components should connect.

Where the data should live.

How the application should behave.

What the user should experience.

What happens when something fails.

How security should work.

How the database should be structured.

How the application should deploy.

How different services should communicate.

And how to recognize when the AI has built the wrong thing.

In other words, I am increasingly becoming something closer to a software architect.

And I think that may be one of the major changes AI brings to software development.

The valuable skill may increasingly become not:

Can you personally write every line of code?

But:

Can you design a good system and direct AI well enough to build it?

That requires a different kind of competence.

You still need to understand software.

Possibly more than people realize.

But you may no longer need to personally perform every mechanical part of creating it.

I will write much more about that later because it has been one of the most important lessons from this entire exercise.

The first experiment: DarinOliver.com

One of the first things I did was build the site you are reading.

DarinOliver.com was partly an exercise in teaching myself modern AI-assisted software development.

I wanted a multilingual publishing platform where I could write about business, decision-making, technology, and what I was learning.

So I built it.

I used Astro.

I added multiple languages.

I added voice narration.

I built the article architecture.

I connected the infrastructure.

I deployed it.

The first version took roughly 24 hours.

And here is the part that still surprises me.

It worked extremely well immediately.

This was not some broken prototype that barely loaded.

The site deployed.

The architecture worked.

The multilingual system worked.

The publishing system worked.

And I had built something in roughly a day that, not very long ago, I would have expected to involve a developer, weeks of back and forth, specifications, revisions, testing, and probably a meaningful budget.

Of course I have continued improving it.

But the astonishing thing was how good the first result already was.

That changed my perception of what had become possible.

AI has become unbelievably good at coding

There is something else people who are not using these systems every day may not appreciate.

The rate of improvement is extraordinary.

Even a year ago, AI coding tools could be very useful, but they also made a lot of mistakes.

They would misunderstand architecture.

Break things.

Invent libraries.

Lose track of context.

Create one problem while solving another.

Today, the experience is dramatically different.

The current systems are astonishingly capable programmers.

They still make mistakes.

Of course they do.

But give a strong model a well-structured problem, enough context, and clear instructions, and it can produce an enormous amount of high-quality software remarkably quickly.

And its ability to debug may be even more impressive.

Something breaks.

You show it the error.

It traces the problem.

Inspects the surrounding code.

Finds the dependency or architectural conflict.

Fixes it.

Runs the tests.

Finds something else.

Fixes that too.

Sometimes I simply watch it work and think:

This is a machine.

An absolute machine.

And the important thing is that this improvement is happening incredibly quickly.

Even the tools Ben Cera had available when he started Polsia late last year were not as capable as what we have today.

That means the opportunity is moving underneath us while we are trying to understand it.

People look at what was possible six or twelve months ago and assume that defines what AI can do now.

It does not.

That is one of the reasons I feel some urgency around this experiment.

Then I built TallyPop

After DarinOliver.com, I wanted something more technically demanding.

A publishing site is one thing.

I wanted to build an actual application with persistent data.

I found the problem at home.

My son was struggling with multiplication tables.

So I decided to build something for him.

That became TallyPop.app.

Initially, it was simply an application for one child.

Then I thought:

Other parents must have this problem.

Then:

Teachers might find this useful.

Then:

This might actually be a commercial product.

Over roughly 48 hours, I built and deployed the application.

It had a database.

User-state tracking.

Feedback loops designed to respond to how a child was performing.

Multiple languages.

A functioning web application.

And I had to learn technologies I had never personally used before.

Supabase.

Vercel.

GitHub.

Cloudflare.

Claude.

Codex running through Terminal on my MacBook.

I had never properly used GitHub in my life.

Now I was deploying a database-backed application through it.

Again, AI wrote the overwhelming majority of the code.

But I was learning what everything did.

I had to understand the architecture.

I had to understand why the database connected the way it did.

I had to understand authentication.

Deployment.

Environment variables.

Branches.

Errors.

Permissions.

And when something broke, I had to learn enough to tell whether the proposed solution made sense.

That is what I mean when I say I am becoming a software architect.

TallyPop is now live.

I intend to add several more features and then start putting it in front of teachers here in Belgium.

And that introduces the next problem.

Because AI has made building dramatically easier.

It has not eliminated the need to get customers.

Meanwhile, Simply Alpha keeps running

I am doing all of this while continuing to run Simply Alpha Capital, LLC, my iGaming consulting business.

My team is already using AI extensively across the company.

We are also working on a separate project in the iGaming sector that is currently in stealth.

I think that project could eventually become very valuable.

But it does not count toward this challenge.

Neither does ordinary Simply Alpha revenue.

I want the rules to be clean.

The $250,000 must come from the new projects I am publicly developing as part of this experiment.

Simply Alpha continues.

Our clients continue.

Our existing work continues.

The stealth project continues.

And simultaneously, I am building these new systems.

That is part of what I am trying to test.

Then I turned AI onto research

A significant part of what we do at Simply Alpha requires keeping up with developments across iGaming, sweepstakes gaming, crypto, prediction markets, game development, regulation, payments, marketing, and related sectors.

There is an enormous amount of information.

Legislation.

Litigation.

Regulatory activity.

Companies entering and leaving markets.

Payment developments.

Product launches.

Marketing activity.

Transactions.

Competitors.

So I started experimenting with AI research.

The results were extraordinary.

I built increasingly sophisticated research prompts and had the system gather, organize, and summarize an enormous amount of useful information.

Eventually, I had Claude turn the output into a searchable HTML research document.

It was extremely useful.

But I soon realized I had simply created a very sophisticated manual process.

I still had to initiate it.

I still had to run it.

And if I did not run it, nothing happened.

That led me to agents.

This weekend: SweepsMonitor.com

My next experiment is SweepsMonitor.com.

I plan to build the first version this weekend and, if everything goes as expected, have it running by Monday or Tuesday.

It will focus specifically on developments in the U.S. sweepstakes gaming market.

But the site itself is almost secondary.

What I really want to learn is whether I can deploy an agent onto a virtual private server and make the entire system operate continuously.

A VPS is essentially a computer running somewhere else 24 hours a day.

The agent should be able to wake up automatically.

Research.

Identify important developments.

Analyze them.

Produce publishable material.

And feed that material into the site.

That means I will be testing whether I can create an autonomous publishing system.

Not theoretically.

A real one.

Can it run continuously?

Can it distinguish news from noise?

Can it generate useful content?

Can it publish?

Can it keep working when I am doing something else?

That is what I want SweepsMonitor to prove.

Then I want to build the marketing agent

Once I have proved the research-and-publishing architecture, I want to tackle distribution.

I have been watching people use AI-generated short-form video to create extraordinary amounts of traffic on platforms such as Instagram.

What interests me is not copying any particular content.

It is the model.

Generate creative variations.

Test them.

Observe what gets attention.

Learn.

Produce more.

Use that attention to drive people somewhere useful.

Once SweepsMonitor proves that I can deploy and manage a persistent agent, I intend to build a marketing agent.

The objective will be to create a system capable of identifying ideas worth promoting, developing short-form video concepts, producing scripts and creative assets, preparing posts, and driving traffic from platforms such as Instagram into one of the businesses I am developing.

I expect that system to become particularly useful for Agentic Radar, which I will explain in a later article.

And that brings the architecture closer to a complete commercial machine:

Research → analysis → product → content → publishing → distribution → traffic → customer

If I can make those pieces work together, one person starts to have capabilities that once required an entire team.

This is the part I find profound

Think about what I am doing at the same time.

I am running Simply Alpha.

My team is developing a separate stealth iGaming project.

I am continuing DarinOliver.com.

I am improving TallyPop.

I am building AI research infrastructure.

I am launching SweepsMonitor.

I am deploying an autonomous research and publishing agent.

I intend to build a marketing agent.

And I intend to develop Agentic Radar.

Traditionally, you would look at that list and say:

That is too much. Pick one thing.

And historically, that was probably excellent advice.

But AI is changing the unit economics of human effort.

I do not have to personally write every line of code.

I do not have to personally research every story.

I do not have to manually translate every page.

I do not have to personally create every first draft.

I do not have to personally debug every software problem.

I do not have to create every piece of marketing content.

And increasingly, I do not even have to personally initiate every task.

The machines can work while I am doing something else.

That does not make my attention unlimited.

My judgment remains finite.

My ability to make decisions remains finite.

And agents can make spectacular mistakes if poorly directed.

But the constraint is changing.

The old question was:

How much work can one person do?

The emerging question is:

How many productive systems can one person intelligently direct?

I find that profound.

Almost superhuman productivity

People describe AI as a productivity tool.

I increasingly think that description is too small.

Productivity means:

I used to take four hours to do something and now I take one.

That is useful.

But what I am experiencing feels different.

AI allows me to attempt things I previously would not have attempted at all.

TallyPop would once have meant hiring developers before I knew whether anyone wanted it.

SweepsMonitor might have required researchers, writers, developers, editors, and somebody monitoring the market all day.

A serious social-media acquisition operation might require researchers, scriptwriters, designers, video editors, and marketing people.

Now one person can at least attempt to orchestrate all of those capabilities.

That starts to feel almost superhuman.

Not because the person has become superhuman.

Because the person suddenly has machines.

And the more capable those machines become, the more important the human role shifts toward architecture, judgment, taste, strategy, and direction.

That may be the real skill I am learning.

Not how to code.

How to direct intelligent systems that code, research, publish, analyze, and eventually market for me.

Why I am doing this before teaching it

One obvious way for me to monetize what I am learning would be to teach it.

There is a very large population of experienced executives and entrepreneurs who understand their businesses extremely well but have barely moved beyond using ChatGPT to write emails.

I have thought seriously about creating hands-on AI education and advisory services for that group.

I probably will.

But first I want to see how far I can take these tools myself.

I want practical competence in AI-assisted software development, research, databases, automation, agents, infrastructure, content production, marketing, customer acquisition, and monetization.

Because getting an AI system to do something impressive is not the same as creating economic value.

The test is whether someone pays you.

That is why I like the 90-day challenge.

It forces the issue.

What counts

The rules remain simple.

Revenue from Simply Alpha does not count.

Revenue from the separate stealth iGaming project does not count.

The $250,000 must come from the new projects built or developed as part of this experiment.

At the moment, that means:

• TallyPop.app
• SweepsMonitor.com
• Agentic Radar
• Anything new directly created through DarinOliver.com
• Anything else I invent during the next 90 days

The point is not to move existing revenue into another bucket and declare victory.

It is to see what I can create from here.

The experiment

So for the next 90 days, I am going to build.

Deploy.

Research.

Publish.

Market.

Sell.

Automate.

Break things.

Fix them.

And show what happens.

Maybe I make $250,000.

Maybe I do not.

But increasingly, the money is not even the most interesting part of the challenge.

The bigger question is:

How much can one person actually do now?

How many businesses can one person test?

How many products can one person build?

How much software can one person direct?

How much research can one person supervise?

How much content can one person publish?

How much distribution can one person create?

How many autonomous systems can one person operate while still running an existing company?

I do not know.

But I suspect the answer is dramatically more than most people currently imagine.

So far:

SimplyAlpha.com — operating.

DarinOliver.com — built and continuing.

TallyPop.app — built, live, and continuing.

AI research infrastructure — underway.

SweepsMonitor.com — this weekend.

Autonomous publishing agent — next.

Marketing agent — after that.

Agentic Radar — coming.

And I intend to run them in parallel.

That may ultimately be the real story.

Not that AI makes one person faster.

That AI allows one person to become the architect of an amount of productive activity that once required an organization.

$250,000.

90 days.

While running the business I already have.

Let’s see what one person can do now.