I started programming over 40 years ago on a TRS-80, where programs and data were stored on cassette tapes.
In those days, every byte mattered.
Memory was limited. Storage was limited. Processing power was limited. Even screen space mattered. Developers learned very quickly to conserve resources because the hardware demanded it. Variable names were short. Logic was compressed. Readability was often secondary to simply making software work within the constraints of the machine.
Over time, that changed.
Hardware became cheaper. Memory became abundant. Storage exploded. Teams became larger. Software systems became more complex and long-lived. The industry shifted from optimizing for machine efficiency to optimizing for human efficiency.
We moved toward long descriptive variable names, layered architectures, self-documenting code, and readability-first development practices because maintenance and collaboration became more expensive than hardware.
For years, that made perfect sense.
But recently, I have been wondering if AI may eventually shift the optimization target once again.
Not because hardware is becoming limited, but because tokens are becoming a metered resource.
As developers, we already understand utility-based thinking in other areas of life. We turn off lights when we leave a room because electricity costs money. We avoid wasting water because utilities are metered. In cloud computing, we learned to think about storage, bandwidth, and compute as measurable operational costs.
AI introduces a new kind of utility: tokens.
Right now, most discussions around AI-assisted development focus on productivity, automation, or whether AI will replace developers. I think the more interesting question may be how AI changes the economic pressures that shape software engineering practices themselves.
Historically, coding styles evolved around the dominant constraint of the era.
Early development optimized for hardware limitations.
Modern enterprise development optimized for human readability and collaboration.
But what happens when AI increasingly becomes the primary reader, writer, refactorer, debugger, and maintainer of implementation-level code?
If AI systems can understand compressed context naturally, will verbose naming conventions continue to matter in the same way they do today?
Could we eventually see development practices shift back toward brevity, not because humans forgot the value of readability, but because the economics of token consumption reward more compact forms of communication?
That does not mean humans disappear from software engineering.
Humans will still define intent, validate business outcomes, establish architecture, enforce governance, ensure security, and remain accountable for what software does. But the relationship between humans and implementation-level code may evolve significantly.
It is possible that future developers spend less time reviewing individual lines of code and more time validating system behavior, orchestration logic, and AI-generated outcomes.
If that happens, readability itself may stop being the dominant optimization target it has been for the last several decades.
And maybe that would not be unprecedented at all. After all, software engineering has always adapted to the scarcity of its era. I started in an era of cassette tapes and severe hardware constraints. Today, we are entering an era of AI systems, orchestration, and token economics.
The constraints are different, but the pattern feels familiar. Maybe code readability was never a permanent law of software engineering.
Maybe it was simply the correct optimization for the constraints of its time.

Eldon Elledge

A reflective piece tracing how software engineering practices have always adapted to the dominant constraint of each era — from hardware limitations in the TRS-80 days to human readability in the enterprise era. The author speculates that AI may shift the optimization target once again, this time toward token economics. If AI becomes the primary reader and writer of implementation-level code, verbose naming conventions and readability-first practices may give way to more compact forms, not because readability lost value, but because the economic pressures changed. Humans would still own intent, architecture, and governance, but their relationship with low-level code may fundamentally evolve.

From Cassette Tapes to AI Tokens