AI's Side Effects on a Veteran Developer: Faster Answers, Atrophied Instincts
Every engineer who leans on AI for daily work is trading manual reps for speed, and the loss shows up as weaker debugging intuition and a harder time proving that experience still matters. When the barrier to producing code drops to near zero, the people who profit are tool sellers, not the people who write code.
AI coding assistants have changed the daily work of a veteran backend developer, and the side effects are showing up in the skills he used to take for granted. The old production-debugging routine of reading logs, tracing source code, and drawing call chains is now a prompt-and-wait loop, with phone-scrolling filling the gap. The intuition that once came from hunting bugs by hand is atrophying, and new vocabulary like Harness, Agent, and Skill gets a quick demo before being filed under “understood.” On the company's internal tech forum, plain knowledge-sharing posts have faded because AI can write them better; posts about AI itself are what draw traffic, even when shallow.
Learning no longer feels like an accumulating asset. Around 2018, studying JVM internals and watching tutorials on weekends built a personal technical base; now a model that has read the internet makes that base hard to price, and a newcomer with AI looks close to the same capability. The reward for blogging is gone, and the profession itself looks precarious: the human is still in the saddle, but as agents improve, the horse is heading toward not needing a groom.
The popular story that an ordinary person with a great idea and AI can get rich is naive. The output has no technical barrier, so anyone with the same tools can reproduce it. The only moat is being first to think of the idea, and that advantage expires quickly in a fast information environment. The gold-rush analogy still applies: the shovel sellers make the money.
The “troubleshooting muscle” complaint is not nostalgia: pattern recognition in debugging comes from repeated trace-and-source sessions, and a tool that removes the intermediate steps also removes the training signal that creates the intuition.
The internal-forum observation describes a market distortion: once a model can write a better explanation of any settled topic, the only content that still earns attention is content about the model itself, which rewards shallowness.
The horse-and-groom metaphor is harsher than the usual “AI augments humans” line: it casts the programmer as a supervisor whose supervision is itself being automated away.
The “everyone can code” message hurts experienced engineers by changing how the market prices output: if a junior with AI can produce something that looks like a senior's work, the price of experience falls before any quality gap becomes visible.
It's already irreversible. The next step is that computer languages won't even be needed—just straight to binary.
Indeed, it's irreversible. That's the general trend.