Every time you open ChatGPT, Claude, or any AI chat, ask yourself one simple question: what can I actually do without this? Not what you used to do. What you can do right now, today, if this window suddenly went dark and never came back.
If that question makes you pause, keep reading.
The Synthetic Euphoria Problem
Social media is flooded with screenshots of apps built in 15 minutes, platforms created over a weekend, automation workflows that would've taken teams of engineers a month to build. The dopamine is real. The results look impressive. And the promise is intoxicating: anyone can be a programmer, a designer, an analyst.
Here's the thing nobody talks about: this works differently for people who already know what they're doing versus those who don't. If you've spent years writing code, you understand what the AI is generating. You can spot the mistakes. You know when the architecture is garbage even if it runs. You're using AI as a multiplier because you have something to multiply.
But what about the generation that's skipping all that? The ones who never sat in a corporate environment getting their work torn apart by a senior developer. Never shipped a project that failed and had to figure out why. Never learned the hard way that "it works" and "it's good" are two completely different things.
They're not less intelligent or less capable. But they're building professional identities on a foundation they've never tested without the scaffolding. And that's a different kind of risk than previous generations faced.
The Economics Nobody Wants to Discuss
Let's get concrete. OpenAI lost $5 billion in 2024. Not revenue — losses. For every dollar they make, they spend $1.69. In 2025, they're projecting a $9 billion loss. By 2028, the company expects $74 billion in cumulative operating losses. These aren't my numbers — they're from financial documents shared with investors, reported by The Wall Street Journal.
The entire industry is operating at a loss. What does this mean? The cheap tokens you're using right now are subsidized. The $20/month subscription that gives you access to a system costing hundreds of millions to run? That's not a sustainable business model. It's a land grab — burn cash to capture market share, figure out profitability later.
"Later" always comes. And when it does, prices rise dramatically or services disappear.
In February 2025, Humane announced their $700 AI Pin would be discontinued — not "no longer supported," but shut off. On February 28th, all devices stopped connecting to servers. Customers who bought before November 15th, 2024 got nothing. No refund. Just instructions on how to recycle their electronic waste.
Builder.ai, a Microsoft-backed startup valued at $1.2 billion, filed for bankruptcy in May 2025. They burned through $445 million. When auditors looked at the books, they slashed revenue projections by 75%. Much of the "AI-powered" development was actually being done by offshore human developers.
In 2024, 966 startups shut down — a 25.6% increase from 2023. The category seeing the sharpest correction? AI wrappers and application-layer tools built on commoditized models.
The Productivity Paradox
Here's something that will resonate if you've actually tried to use AI for complex work: on simple tasks, AI is genuinely faster. Write a script. Draft an email. Summarize a document. Things that took 30 minutes now take 5. That's real value.
But on complex tasks — particularly when requirements are highly specific or poorly defined — something weird happens. You start spending more time explaining what you want than it would've taken to just do it yourself. You iterate. You clarify. The AI confidently gives you something that looks right but isn't. You fix it. It breaks something else.
I've watched people spend entire days trying to get AI to produce something that would've taken them a few hours to build manually — for certain types of work. The manual approach would've been more painful, more monotonous. But it would've been done. And it would've been right.
This doesn't mean AI never saves time on complex work. It means the productivity gain isn't universal or automatic. It depends heavily on what you're building, how well you can specify it, and whether you understand the domain well enough to verify the output.
The Quality Problem
There's a term gaining traction: AI slop. It describes the unmistakable aesthetic of AI-generated content — the uncanny smoothness of prose, the suspicious absence of personality, the way everything feels optimized for engagement but drained of humanity.
AI slop isn't just about text. It's in code that works but has no coherent architecture. Designs that look polished but have no design system. Automation that runs but breaks in ways impossible to debug because no human ever understood how it worked.
People are starting to notice. The pattern recognition is developing. And as AI content floods every corner of the internet, something interesting may happen: original human content — messy, imperfect, idiosyncratic — could become rare. And rare things become valuable.
"Made by human" might become the new premium signal.
The Competence Question
A new professional class has emerged: the AI-analyst, the AI-developer, the AI-creator. Thousands of people are building careers on the ability to operate AI tools. They know which models to use for which tasks. They understand prompt engineering. They can chain together services in clever ways.
Here's the uncomfortable question: what happens when you take away the tools?
An AI-developer who can't write code isn't a developer in the traditional sense — at least not yet. They're an operator. And there's nothing wrong with being an operator, as long as you understand what that means for your long-term autonomy. Operators depend on the availability and affordability of systems they don't control. That's a risk worth acknowledging, even if it's a risk you choose to accept.
Samsung discovered the stakes in 2023 when engineers uploaded confidential source code to ChatGPT. Three separate incidents in 20 days. One engineer asked ChatGPT to fix bugs in semiconductor database code. Another used it to optimize chip defect testing sequences. A third uploaded an entire internal meeting to generate minutes.
The engineers weren't trying to leak secrets. They were trying to be productive. The AI had become such a natural extension of their workflow that the security implications didn't register. Samsung banned ChatGPT, but the information was already on OpenAI's servers.
The Automation Risk Equation
Automation has costs beyond token pricing. When you automate something, you're handing control to a system that doesn't understand context the way humans do. You're giving AI access to social media accounts, APIs, financial systems, customer data.
When a human makes a mistake, it's usually caught before it cascades. When an automated system makes a mistake, it can execute that mistake across thousands of operations before anyone notices.
Here's the irony: sometimes it's actually cheaper to hire a junior employee to do routine work than to automate it. Not because automation is expensive in terms of tokens, but because the junior employee understands context. They won't accidentally post sensitive information publicly. They won't delete the wrong database. And when they make mistakes, they're small, human-scale mistakes that get caught and corrected.
As token costs rise and AI slop becomes more recognizable, we may see selective returns to human labor — not because humans are faster, but because humans are safer in contexts where the cost of errors is high.
The Dependency Architecture
Let's zoom out. The major AI players — OpenAI, Google, Anthropic, Microsoft — own the infrastructure. They own the models. They control the APIs. Everyone else is building on their land, paying their rent, following their rules.
Your entire business, your entire workflow, your entire professional capability can be affected by a configuration change in someone else's data center. When they raise prices, you pay. When they change their acceptable use policy, you comply. When they decide your use case isn't aligned with their values anymore, you adapt or you're gone.
This dependency extends to physical infrastructure too. When Cloudflare goes down, half the internet becomes unreachable. When AWS has an outage, companies lose millions per hour. We've accepted this fragility as the cost of doing business.
Now add the AI layer. Imagine professionals who can't do their jobs without AI tools. Imagine businesses whose core operations depend on API availability. One major outage, one policy change, one geopolitical event that cuts access — and what happens?
We saw rockets hitting Dubai skyscrapers in 2025. Nobody thought that was possible either. "The infrastructure is too important" isn't the protection it used to be.
The Privacy Dimension
The new generation of AI tools — Claude Code, Cursor, various "computer use" features — don't just process text. They access your computer. With your permission, they can see files, run commands, interact with applications.
Think about what you're granting access to. Every password manager. Every document. Every browser session. Every piece of code you've ever written. All of it accessible to systems controlled by companies you've never visited, governed by laws you've never read.
Maybe you're not handling sensitive data. But millions of people are using these tools, and some of them are processing trade secrets, legal documents, medical records, financial information. All flowing through systems we're asked to trust on faith.
Companies serve shareholders. They operate within legal frameworks that can compel disclosure. The convenient AI assistant exists within power structures you don't control.
This isn't paranoia. It's just reading the terms of service.
The Internet Comparison
You might be thinking: "This sounds like the panic people had about the internet. We adapted then. We'll adapt now."
That's partially right. We do adapt. The internet turned out to be transformative in mostly positive ways.
But there's a difference worth noting. The internet gave you access. It lowered barriers. It put information at your fingertips. But it didn't do the work for you. You still had to read the articles, learn the skills, practice the craft.
AI gives you results. It does the work. And in doing so, it can remove the process that creates deep understanding — if you let it.
I want to be careful here, because this isn't an either/or. AI can absolutely expand capabilities when used consciously — as a learning accelerator, a research assistant, a way to prototype ideas before committing to learning full implementation. The danger isn't in the tool. The danger is in using the tool as a replacement for capability development rather than a complement to it.
The question isn't whether AI is good or bad. The question is whether you're using it to amplify genuine skills or as a substitute for developing them.
What This Actually Means
I'm not telling you to stop using AI. I use it. It's valuable. It makes certain things genuinely better.
But be honest with yourself about what you're trading. Every time you let AI do something you could do yourself, you're choosing convenience over capability. Sometimes that's the right choice. Often it is. But make it consciously, understanding the implications.
Ask yourself regularly: what would I do if this tool disappeared tomorrow? If your honest answer is "I'd be lost," that's a signal. Not to stop using the tool, but to maintain the underlying capability that makes the tool optional rather than mandatory.
Here's a practical starting point: take one task you currently rely on AI for, and do it manually once a month. Just to keep the muscle memory. You don't have to go back to the stone age — just maintain the option.
Build on top of AI, but don't hollow out the foundation.
The Actual Question
We started with: what can you do without AI?
But maybe the better frame is: what are you becoming?
Are you using AI to amplify genuine skills? Or as a substitute for developing them? Are you building something that would survive a change in the technological landscape? Or constructing an elaborate dependency that looks impressive but has no foundation?
These aren't comfortable questions. The technology itself won't help you answer them — it only knows how to respond to the next prompt.
The AI revolution is real. The capabilities are genuine. The transformation is happening.
The question is whether we'll navigate it wisely, or wake up one day to discover we've traded capabilities for conveniences that can be revoked with the flip of a switch.
That's not a question AI can answer. That's on us.
