The Privilege of Refusing AI

AI skeptics often come from privileged positions. This explores why that matters, its costs to others, and how founders should engage outside the gatekeeper economy.

I’m not naive about AI.

I see the risks. The ethical gaps. The very real concerns about labor, bias, power, and control.

Those conversations matter.

But what I struggle with is how often the loudest “AI is making you dumb” narratives often come from places of deep privilege.

Maria-Ines | Founder

The equalising power of AI

For the first time in history, access to leverage is not reserved for multi-million dollar companies with massive teams and budgets. For the first time, a single person can build, create, test, iterate, and ship without asking permission, without gatekeepers, without capital they do not have.

That matters.

It matters especially if you come from a country where systems have failed you. If you have never had institutional backing. If you have had to be resourceful by necessity, not by choice.

It matters if your accent has cost you opportunities. If your name made it harder to get the email opened. If you grew up knowing that the playing field was already tilted before you stepped onto it.

For these people, AI is not a “shortcut.”

When systems fail in front of you, you learn that the systems were not protecting you in the first place.

When the lights go out.

I am Venezuelan. I learned English as an adult.

I grew up watching infrastructure collapse in real time. Not in the abstract way people use that phrase in podcasts. Literally. The lights would go out. The water would not come on. The currency would lose value while you were standing in line to spend it. Hospitals ran out of medicine. Universities lost professors who could no longer afford to teach.

When systems fail in front of you, you learn something most people in stable economies never learn. You learn that the systems were not protecting you in the first place. You learn that “the rules” were always conditional. You learn to build the small structures around your own life that the larger structures were never going to hold.

You learn that resourcefulness is not a personality trait. It is a survival skill.

So when I hear well-resourced people in stable economies simplify AI as dangerous and we should all slow down, I hear something familiar underneath it. I have heard this voice before. It is the voice of people who already have access. People who already have leverage. People who have never had to invent their way through a closed door.

And maybe they are right that they personally do not need this tool. But the woman in Manila building her first digital product does. The founder in Lagos who cannot afford a marketing team does. The single mother in São Paulo competing against companies with ten times her resources does.

AI is not a toy for people who are already winning. It is infrastructure for people who were never supposed to get a seat at the table.

The typo that became the argument

I learned English as an adult. I have built my life and my business in it. I can think in it now. I dream in it sometimes. But the small mistakes never fully go away. The wrong preposition. The mixed-up words (your instead of you’re). The sentence that reads slightly off.

For most of my professional life, those small mistakes have been treated as evidence.

Evidence of what, exactly? People are rarely explicit. But the underlying message is consistent. Your English is the proxy for your competence. The proxy for your seriousness. The proxy for whether anyone is going to take what you actually said seriously.

I have been corrected publicly in meetings. I have had strangers reply to my emails not to engage with the idea but to point out a typo. I have been marked down in corporate performance reviews for my English while watching colleagues working on Latin American development projects butcher Spanish without consequence and call it “international experience.”

I am not telling you this for sympathy. I am telling you because I recognise the energy.

When someone says “AI is making your writing worse” or “just write it yourself,” I hear something I have heard many times before in different costumes:

The way I learned is the right way. The advantages I had are invisible to me. Your shortcut is cheating.

For some of us, AI is not a shortcut. It is the first tool that does not come with humiliation attached.

They always had AI. They just called it staff.

Here is what people forget about the “do it yourself” argument.

The executives writing think pieces about the evils of AI have always had support. They have always had research assistants, junior associates, ghostwriters, editors, communications staff, and friends from school willing to “give your draft a quick read.” They have always outsourced the work of polish to other humans, and that outsourcing was treated as a sign of seniority, not weakness.

Nobody ever told a Fortune 500 CEO that working with a ghostwriter was “rotting their brain.”

But when the solopreneur, the immigrant, the non-native speaker, the single parent, or the person without a Stanford MBA gets access to similar leverage through software, suddenly we need warnings.

Suddenly the language changes. It is not “support.” It is “dependence.” It is not “delegation.” It is “laziness.” It is not “leverage.” It is “cheating.”

The argument that AI robs you of agency, robs you of agency.

It assumes you cannot tell the difference between using a tool and being used by one. It strips judgment from the very people who have always had to make harder, faster judgments than the people who get to write rules about it.

I trust people to think for themselves. I trust them to use tools consciously. I trust them to know the difference between support and surrender. The people I work with are sophisticated adults running real businesses. They do not need to be protected from a technology by the people who already had every advantage before it existed.

The real issues, and the fictions that hide them

This is where I want to be careful.

I am not arguing that there are no real concerns about AI. There are.

Concentration of power in a handful of companies is real. Environmental costs of training large models are real. Labor displacement is happening, right now, in industries that already had thin margins and few protections. Bias baked into training data is shaping outcomes for people who have no idea they are being shaped.

Believe me, I know.

Type the word Latina into any image generator and watch the model strip her down to the cliché it has decided that word means.

I am not asking anyone to look away from those things. I am one of the people working inside the technology, and I want it built better than it currently is.

But I want to name a quiet fiction that often shows up next to the alarms. The fiction is that pre-AI systems were somehow fair, humane, or equitable. That if we just slow down or refuse to engage, the playing field stays level. That the alternative to AI is some neutral baseline.

That fiction is convenient if you were doing well in the pre-AI baseline. It is not true if you were not.

The gatekeepers were already in place. The barriers were already high. The credentials were already designed to favour people who looked like the people writing the credentials. The capital was already concentrated. The hiring pipelines were already broken. The publishing industry, the venture industry, the consulting industry, the credentialing industry, all of them were already filtering out the kinds of voices that AI, despite its own biases, is now amplifying in new ways.

AI is not creating inequality. It inherited it.

And used intentionally, used by people who refuse to be polite about exclusion, AI can also disrupt it.

The room is filling up and you are not in it.

If you are reading this from somewhere comfortable, from a country whose electricity does not flicker and whose internet does not lag, from a name and an accent and a set of credentials that the rooms you walk into already respect, you may not see what I am about to describe. The data is going to read as abstract. Read it anyway.

For the rest of us, the global majority, the Global South, the women whose adoption gap is widening rather than closing, the Black and Hispanic and Indigenous founders, the immigrants, the non-native speakers, the people whose names the algorithms have never seen, the people whose languages do not get the best models trained on them first, this is not abstract. This is the room we are watching fill up without us in it.

This is the part of the AI conversation that does not get loud think pieces. The adoption gap is not closing on its own. It is widening.

According to the Microsoft AI Diffusion Report released in January 2026, the gap in AI adoption between the Global North and the Global South grew over 2025. Adoption in wealthy markets reached 24.7% of working-age adults. The Global South reached only 14.1%. The gap widened from 9.8 to 10.6 percentage points in a single year, because Global North adoption grew nearly twice as fast.

Globally, women have 22% lower odds of using generative AI than men, according to a Harvard Business School working paper drawing on 18 studies and 143,000 individuals across 25 countries. The gap holds across nearly every region, sector, and occupation studied. In the US specifically, Deloitte projects the gender gap will close by the end of 2025. Outside the US, it has not.

The racial gap in the US alone runs along familiar lines. Pew Research found that 27% of Black Americans and 29% of Hispanic Americans interact with AI daily, compared to 31% of white Americans and 39% of Asian Americans. The pattern flips among teens, where Black and Hispanic teens are actually adopting faster than their white peers. But income remains the gate. Households earning over $75K are far more likely to use ChatGPT than those earning less.

Income is the cleanest predictor of access. US households earning over $100K are more than twice as likely to use generative AI as households earning under $50K. Of the 8.1 billion people on this planet, only 1.2 billion have ever used an AI tool. The bottleneck is not interest. It is electricity, internet, digital skills, and money for paid tiers.

Anthropic’s own published data confirms the concentration. The top five US states account for roughly half of all Claude usage, despite holding only 38% of the working-age population. Country-level usage correlates strongly with income.

This is what the data is telling us. The people who could be using AI to compete against the gatekeepers are not in the room because the room is filling up with people who already had every other advantage. Women report lower confidence with the tools, higher ethical concern, and fear of being judged for using them. The people most likely to be told their use of AI is “lazy” or “unethical” are statistically the same people most likely to walk away from the tools entirely.

Every “AI is dangerous” think piece pushes that needle further. Not because the writer intends it. Because the people who hear “AI is dangerous” most viscerally are the ones who never had the institutional cushion to absorb being wrong about a tool. The wealthy male engineer in San Francisco who reads the warning will keep building anyway. The single mother in São Paulo who reads it will close the tab.

That is the actual cost of letting this conversation be dominated by the people who can afford to opt out.

What it actually costs to be in the room

I want to be honest about a tension in what I just told you.

The data on the access gap is real. Electricity, internet, digital skills, money for paid tiers. These are not excuses. They are constraints. There are people for whom $20 a month for a paid AI subscription is not feasible, and those people deserve to be part of the policy conversation about AI access.

But if you are reading this essay, you are probably not that person.

If you are running a business, even a small one, even an early one, the honest math has changed. The free tiers of these tools are now stronger than the paid tiers were eighteen months ago. The paid tiers are stronger than what entire teams of consultants used to deliver. A single $20 subscription, used intentionally, can replace a stack of other tools, contractors, and stalled projects.

I have personally consolidated multiple software subscriptions into one AI tool in the past month. The net monthly savings are substantial. And that is before counting the work that simply gets done now that used to sit in a queue waiting for someone to find the time.

And if even $20 is not in the budget, the open-source frontier has matured. Llama, Mistral, DeepSeek, Qwen, all free, all downloadable, all powerful enough for serious work. Apps like PocketPal and MLC Chat let you run capable AI models directly on your phone, offline, with no account, no subscription, and no data sent to anyone. The output is not as polished as the paid frontier tiers, but it is far beyond what $20 a month bought two years ago. “I cannot afford a tool” is a smaller and smaller fraction of actual cases.

Early generative AI was a toy. You played with it, and the outputs were almost useful. That phase is over. The output of the current models, used by someone who knows what they are asking for, is genuinely effective. It is doing work that used to require a junior associate, a research assistant, an editor, or three meetings with a contractor. The threshold has been crossed.

Which means the “I cannot afford to learn AI” position has gotten quietly more expensive than it looks.

The people who decide to sit out this round are not staying neutral. They are losing ground. Not to the AI itself, but to the founders who learned to use it while they were still discussing whether they should. The same dynamic that played out with the internet, social media, mobile, and cloud is playing out faster this time. The window in which “I am not on this platform yet” was a defensible position closed years ago for those tools. It is closing now for AI.

AI is not going to replace you. The founder down the street who learned to use it intentionally is.

---

If you want in, here are your options

So here is what to do if you have decided you want to be in the room.

Pick one tool. Not five. Not a stack you read about on a podcast. Pick a single frontier model, Claude or ChatGPT or Gemini, whichever fits how you think, and commit to it for thirty days. Tool fatigue is the fastest way to fail at AI. The leverage compounds when you go deep, not wide.

Start by extracting, not generating. The most underrated use of AI is not making new things. It is pulling what you already know out of your head. The methodology you have been delivering for ten years. The way you explain your offer on every sales call. The thinking you do reflexively that you have never written down. Have the AI interview you about it. Then have it organize what you said back to you. That codified intelligence is worth more than a year of social media posts.

Bring context. Generic prompts produce generic outputs. The AI is only as useful as what you tell it about you, your business, your client, your voice, your specific situation. Build a working document of context you can paste at the start of every conversation. A page of “here is who I am, here is what I do, here is who I serve, here is what I sound like.” That alone takes outputs from a 4 to a 9.

Iterate. The first answer is the worst answer. The skill is not “write a perfect prompt.” The skill is the back and forth. Push back. Say “no, that is not the angle.” Tell it what is wrong. Ask for three more versions. The output that lands at iteration five is going to outclass anything you would have produced alone in the same hour.

Measure it against the right baseline. Stop comparing AI output to a Pulitzer-winning essay or a million-dollar consultancy report. Compare it to what you would have done without it. Compare it to the email you would never have written, the methodology you would never have documented, the strategy you would have kept procrastinating. The right question is not “is this output perfect?” The right question is “did this make me three times more effective today?”

That is how you get in the room. Not by becoming an AI expert. By treating one tool, intentionally, like the leverage it is.

---

Where I land

AI is not “the future.” It is here. It is in your pocket. It is in your inbox. It is in the supply chain of the next product you buy.

The real question is not whether you like it or fear it. The real question is whether you are willing to engage with it consciously, or whether you are going to leave its shape entirely in the hands of the few.

Refusal does not stop the shift. It just removes your voice from it.

If you opt out, the future of AI is decided without you. By the people who already had the loudest voices in every previous round.

If you engage, you get to be one of the voices in the room. You get to use the tool against the gatekeepers who built it. You get to extract your own intelligence into systems that hold and amplify it instead of locking it inside your own head.

That is the part nobody talks about, because it is the part that scares the people who benefit from the current arrangement.

The question underneath this question

I am going to plant something here that I am not opening in this essay, because it deserves its own.

Once the alarm bells are gone, and the cost is solved, and all the obstacles I have spent this essay dismantling are gone, what is actually left is not “access to AI.”

What is left is the question of who you are when the thing you deemed your value, the thing you spent twenty years getting good at, the work that defines what you charge for, gets done by software in eight seconds.

That is the harder conversation underneath all of this. It is the question every founder I work with eventually has to sit with, and most of them do not realise it is coming until they are already inside it.

I am writing that one next. Subscribe so you do not miss it.

“Refusal to AI does not stop the shift. It just removes your voice from it.”