In my previous essay, I argued that agentic coding tools like Claude Code represent an emergent genre -- Terminal Tycoon -- where the interaction model resembles MUDs and building sims more than traditional programming. You design systems, manage throughput, and observe outcomes through text, rather than writing code directly.
But there's a parallel that kept nagging at me while writing that piece, one that felt too heavy for what was already a long essay: modern warfare.
The Drone Operator Problem
Over the past decade, drone operators have become some of the most effective combatants in contemporary conflicts. And the skills that make a good drone pilot turn out to be... not what you'd expect.
Not the physical conditioning of special forces. Not years of flight school. Not the combat instincts honed through live-fire exercises. Instead: pattern recognition, situational awareness from multiple screens, comfort with controller interfaces, the ability to make split-second decisions from abstracted representations of reality.
The Air Force and Space Force now actively recruit from gaming communities. This isn't a gimmick. The skills of someone who's spent thousands of hours in competitive FPS games -- tracking multiple targets, processing information from minimap and HUD simultaneously, staying calm under time pressure while operating through an interface -- transfer better to drone warfare than traditional combat training.
The drone operator doesn't need to be physically present. They don't need to feel the G-forces or smell the cordite. Their entire perception of the battlefield comes through screens and data feeds. And crucially, they need to be comfortable acting decisively based on that mediated perception, without the direct sensory confirmation that humans evolved to rely on.
Is Software Having Its Drone Moment?
Something similar might be happening to software development.
The skills which made someone a great traditional programmer -- meticulous attention to syntax, deep knowledge of language internals, the ability to hold complex state in your head and manipulate it precisely -- might matter less in agentic development than entirely different capabilities. Systems thinking over code manipulation. Throughput intuition over correctness obsession. Comfort acting on agent reports without inspecting every line. Pattern recognition across dashboards and activity feeds, the way a drone operator reads sensor data.
Traditional programming rewards people who can hold a complex codebase in their head and manipulate it precisely. Agentic orchestration might reward people who can manage systems they can't directly see -- who are comfortable treating the codebase as a black box they interact with through textual commands and reports.
Reading the Meta
Another gaming skill might matter even more: reading the meta.
In competitive games, the "meta" is the current landscape of optimal strategies -- what works, what counters what, what's fallen out of favour. It shifts constantly as patches land, new techniques emerge, and counter-strategies develop. Top players in games like League of Legends or StarCraft don't just master one approach. They develop a sense for when the ground is shifting, and they adapt before everyone else catches on.
The AI tooling landscape has this quality. What's optimal today -- the specific agents, workflows, prompt patterns, model versions -- might be obsolete in six months. Gas Town itself might not outlive the year. The skill isn't mastering this tool; it's developing intuition for how the whole space is evolving and repositioning accordingly.
What does this look like in practice? A meta-reader notices when a new model drops and immediately tests whether their existing prompts still work, or whether the new model responds better to different phrasing. They track which workflow patterns are spreading through the community and which are quietly being abandoned. They sense when an approach that worked last month is hitting diminishing returns -- maybe context windows got bigger and the elaborate chunking strategy is now overkill, or maybe a new tool handles a whole class of tasks they were doing manually. They're willing to throw away working setups because they've pattern-matched that something better has emerged.
If you've ever mass-deleted a workflow you spent days perfecting because you sensed the ground shifting underneath it, you know this instinct. It feels wasteful to engineers. It feels obvious to competitive gamers.
This is a competitive gaming skill, not a traditional engineering one. Engineers are trained to go deep -- to master their tools, understand the internals, build expertise over years. Meta-readers are trained to go wide -- to sense patterns across the landscape, notice when something new is working, and switch before their current approach becomes a liability.
Both skills are valuable. But they're different skills, and they tend to live in different people.
The Discontinuity Question
If this is true, we might be watching something more dramatic than gradual evolution. Not a smooth transition where existing experts add new tools to their toolkit, but a genuine discontinuity -- where the new masters come from unexpected places.
The shift from cavalry to mechanized warfare didn't just change tactics. It changed who was good at war. Horsemanship, sabre technique, the instincts developed over centuries of mounted combat -- these didn't transfer. The skills that mattered were different skills, and the people who had them were different people.
The shift from dogfighting to drone warfare is similar. The fighter ace's reflexes and spatial intuition, developed through years of pulling Gs in a cockpit, don't help you operate a drone from a climate-controlled trailer in Nevada. The ace learned to feel the aircraft shudder at the edge of a stall; the drone operator watches a latency indicator tick up and makes the same call from a swivel chair, Mountain Dew on the desk. What helps is what gamers have: screen-mediated situational awareness, comfort with abstracted interfaces, pattern recognition across multiple data feeds.
I'm not claiming vibe coding is definitely this kind of shift. Traditional programming skills surely still matter somewhere in the stack. Someone has to build the agents, train the models, debug the infrastructure when it breaks. The question is whether that "somewhere" shrinks to a smaller and smaller portion of what we call software development -- while the larger portion becomes something closer to systems orchestration.
The Dark Mirror
If the interface is a game, it inherits gaming's darker patterns too.
The same building sim instincts that help you think in throughput also make you susceptible to "one more zone" compulsion. The same competitive drive that helps you read the meta keeps you grinding when you should step away. The same comfort with abstracted interfaces that lets you act decisively without seeing the code also distances you from the consequences of what you're building.
Mediated interfaces change our relationship to what we're doing. Drone operators report a specific kind of strain -- not from the violence itself, but from the dissonance between the screen and the reality it represents. The interface that makes them effective also creates distance. This isn't a moral equivalence between software and warfare; it's an observation about what happens when we act through abstractions. When you can't see the code, when your entire perception of the system comes through agent reports and dashboards, do you understand what you're actually building? Do you feel responsible for it in the same way?
The people most naturally suited to this paradigm -- those who thrive on systems thinking, who are comfortable with abstraction, who can act decisively on incomplete information -- might also be the ones most vulnerable to its disconnection. Not just the compulsive loops of the activity feed, but a deeper detachment from the reality of what they're creating.
An Honest Uncertainty
The parallels in this essay might be too neat. Gaming comparisons are seductive, and seductive comparisons often mislead. The skills that make a good Factorio player might not actually transfer to agentic development -- it could be a satisfying metaphor that doesn't survive contact with reality.
Every technological shift brings predictions of dramatic discontinuity, and most turn out more gradual than the prophets expected. Traditional programming skills might adapt and remain relevant in ways I can't foresee.
But the question feels worth asking: In five years, will the best "developers" be people who never learned to program in the traditional sense -- but who know how to read a meta, think in throughput, and manage systems they can't directly see?
The factory must grow. But so does the distance between you and what it's making.
Comments
Comments