The more I watch AI develop, the more I think it may amplify a familiar economic pattern: a widening gap between people whose capabilities compound and people who struggle to keep up.
Economists sometimes describe an uneven recovery as K-shaped: one part of society moves upward while another moves downward.
AI could create a new version of that divergence.
Not simply a gap in income, but a gap in access to cognitive tools, automation, and the ability to shape one’s working environment.
The emergence of personal infrastructure
Powerful models, agents, and modular workflows are making it possible for individuals to assemble tools that once required dedicated software companies.
A person with the relevant skills can create systems for meeting transcription, document analysis, research, task coordination, and recurring operational work.
Some of these systems can run locally. Others combine open-source software, hosted APIs, and small automation layers.
Take meeting analysis. Instead of depending entirely on a commercial product, someone may choose to assemble a pipeline using a transcription model such as Whisper, a language model, and a structured archive of decisions and follow-ups.
That does not mean the homemade version will always be better or cheaper. Maintenance, reliability, security, and time have real costs.
The point is that the option to build one’s own infrastructure is becoming more accessible — but not equally accessible to everyone.
Possibility is not the same as practical access
On paper, many of these technologies are available to nearly anyone.
In practice, building a system and using it reliably every day are entirely different things.
You need computational resources, technical understanding, time, and the patience to maintain a solution as models, APIs, and dependencies change.
You need to understand where data lives, what breaks when an integration changes, and how to check whether an output is useful or simply convincing.
Someone who already has these resources can use AI to become more productive — and then use the additional time and income to develop still better systems.
Someone without them may remain dependent on standardized tools, rising subscription costs, or whatever capabilities an employer happens to provide.
This is a compounding difference.
Dependency has a political dimension
I sometimes think about this in terms borrowed from dystopian literature.
One direction is concentrated public authority: governments with increasingly capable systems of monitoring, classification, and control.
Another is concentrated private authority: a small number of firms controlling the essential platforms through which people work, communicate, and access intelligence.
Neither scenario is inevitable, and public institutions and private companies are not simply interchangeable villains. Both can provide important services, protections, and opportunities.
But the concentration of digital infrastructure is a real question of power.
If your tools, data, memory, and working processes all live inside systems you cannot meaningfully inspect, modify, or leave, how much autonomy do you actually have?
Autonomy does not require building everything yourself
The answer cannot be that everyone must become a systems engineer.
Most people do not want to maintain models or run servers, and they should not have to.
A more realistic goal is meaningful choice: portability, interoperability, comprehensible privacy rules, strong public safeguards, and access to education that makes the technology less mysterious.
Open tools and smaller providers may help. So can well-designed commercial products that let people retain control over their information and move elsewhere when they choose.
I think the emerging divide will be shaped by a deceptively simple question:
Do you have the ability to change the system you depend on, or can you only live inside it?
That question reaches well beyond AI. It is becoming part of the architecture of autonomy itself.
Filed underHuman Systems
