No Permanent Underclass in the AI Era

I have been building SaaS products, tinkering with AI, and watching this industry evolve for years. Lately I have been angry about something specific, and it is not a sci-fi doomsday scenario.

Anthropic dropped another “safety-first” decision. It reads like the same playbook gatekeepers have used forever: scream about risk in public, then control access in private.

They are trying to sell us a permanent underclass. They are calling it safety.

We have seen this before. “Think of the children.” “National security.” “Public safety.” Every time a new technology threatens the status quo, the incumbents panic in public and build walls in private. This time the technology is AI. The wall is access.

If I had full access to the best model available, I would not lose sleep over someone with “higher access” replacing me.

The fear of being replaced is not really about what the model can do. It is about who is allowed to use that capability.

When the strongest models sit behind whitelists, enterprise deals, or “responsible use” committees, companies like Anthropic are not protecting society. They are building a two-tier world: people who get the full model and people who get the watered-down version.

That is not safety. That is market positioning dressed up as morality.

AI coding is not magic. It is another abstraction layer, the same story as assembly to C to Python to today’s agentic tools. We had compilers decades ago.

The unsolved problems are not “AI will kill us all.” They are the boring ones: better data pipelines, integration friction, edge-case reliability, shipping products people actually want. Restricting access does not fix any of that. It just slows everyone down except the people already at the top.

The marketing is part of the game too. Headlines about rogue AI and existential risk get clicks, funding, and regulatory moats. The quieter story, that open and powerful AI could be a real equalizer, gets buried under safety theater.

We should push back. Not with pitchforks. With clear demands.

Blacklist truly dangerous use cases. Most people can agree on that. Stop expanding vague whitelists that only big players can navigate. Give developers, indie hackers, researchers, and curious people the same raw capability the labs keep for themselves. Let the market decide what gets built, not a handful of safety bureaucrats.

History is pretty clear on this. The printing press, the internet, open-source software. The technologies that actually changed things for the better won when they left the hands of gatekeepers. When fear wins, you get stagnation and inequality.

I am not saying there are zero risks. I am saying the biggest risk right now is a future where a tiny elite controls cognitive firepower while everyone else is told to stay in their lane for their own good.

That future is not inevitable.

We have the tools. We have the talent. What we need is the will to reject this underclass narrative before it hardens into policy.

Anthropic, if you are reading this: the people actually building things see the play. We see the power grab. We are not buying it.

The rest of us should stop waiting for permission. Build anyway. Ship anyway. Demand full access anyway.