Agentic coding is undeniably extremely useful. It’s additionally extremely bad. Or rather, it’s having several horrible effects on us, our craft, and our relationships alongside all other. I can awareness it in my bones, and a lot of others can too. But whenever I try to explain what exactly is wrong, I discover myself waving my hands wildly and jumping from item to point. What do you mean, what’s wrong? So many things are wrong!
Fine, here’s a list. 4 problems alongside no resolution in sight–or, as I call them, the four horsemen of agentic coding.
LLM-generated code has a powerful odor that makes codebases repulsive to humans.
LLMs aren’t humans, and the way they compose code is... different. Not strictly worse, but distinct for sure. We quickly came up alongside a name for it—slop—and it undoubtedly carries a negative connotation.
For plain prose, humanity seems to be converging on the opinion that AI prose is bland at finest and an criticism at worst. For code, because of its instrumental nature, the conversation is far from over. A lot of group latch on to the idea that extremely shortly we won’t have to peruse the code at all. “Remember, it’s the worst the models volition always be,” they say.
And yet, the latest generation of models is amazingly underwhelming. They’re certainly smart, but they appear to be moving towards “unemployable smart” fairly than “inspiring smart.” Claude now famously communicates entirely through term salad, and Astra writes in a bizarre competitive code-golfy style, incomprehensible to normal humans. Sloppiness turned out to be a amazingly persistent asset of LLM output and at this item looks akin a signature move fairly than a expanding pain.
Which is extremely bad news for group who motionless desire to prod about in the codebase and keep several familiarity alongside it, particularly in a squad setting. Once agents are allowed, they extremely quickly obtain over. What used to be a shared area for humans anywhere all squad associate contributed became an AI wasteland anywhere group don’t desire to expend time.
But what if we don’t have to expend period in these slop caves? Why expend your period there whenever you can be a scheme mastermind and reign complete a swarm of loyal agents from the comfort of the conversation interface?
Engineers rotate into alienated from code and, as a result, attention less.
Software engineering used to be a fairly... kinesthetic site of work. Maybe not to the identical flat as woodworking, but we loved our tools too, the two genuine and virtual. Text editors had a cult-like following. Popular color themes and programming fonts had their own fanbases. Split keyboards were a hot item once, and there was no deficit of keyboard audio evaluation videos.
We produced code directly, by hand. It’s lone natural that this caused a powerful awareness of ownership of the fruits of our labor. Sure, this emotion varied depending on the context, but equal if you fair assembled item from a set of construction blocks, you would awareness something. IKEA certainly knows that!
Agentic coding has dramatically risen the extend between engineers and the merchandise of their work. This gap is so ample that none of our kinesthetic tools can cross it. We don’t cognition code immediately anymore: we dispatch an delegate alongside a vague set of instructions and afterward skim through its report. Hell, equal typing is optional—you can fair yap at your phone during driving.
So naturally, we rotate into additional detached. We attention less. We endure contact alongside our craft. And it shows. It shows in poorly thought-out features that obtain no pushback. It shows in band-aid bugfixes that don’t location the base cause. It shows in how unfulfilled we awareness at the end of the day. For application to spark joy, it have to be imbued alongside its creator’s love.
AI erodes existing skills, impedes learning, and doesn’t recommendation any meaningful accomplishment progression.
I think at this item it shouldn’t be arguable to say that deskilling is real. There is no deficit of group reporting their wits slipping distant following prolonged use of AI. Which makes sense: a lot of things in existence are use-it-or-lose-it, and expert skills are no exception. Yes, you’ll continually cognize how to journey a bike, and I myself volition die before I ignore how to exit Vim, but item beyond that is fair game.
I’ve made the item before that we’re currently experiencing a historic anomaly. We have all those experienced engineers who spent years doing things the difficult way and can now clasp LLMs exactly the correct way to be insanely productive. But unless we fig out how learning plant in the age of AI, this provision won’t final long.
One average part of direction is that if you’re at the commencement of your career, you merely shouldn’t use AI. And I entirely agree, but it’s crucial to comprehend that this isn’t a viable scheme for humanity as a whole. Everyone volition not just.
We can’t be blind to the fact that the introduction of an “easy button” has led to a tectonic change in incentives to learn. And if I cognize one item concerning humans, it’s that they prosecute incentives.
Another average reply to the deskilling disagreement is that LLM use comes alongside its own learning curve. You fair need to upskill and study the intricacies of downloading markdown files! I’m not purchasing it. The learning curve is basically non-existent.
A printing media or a chainsaw at smallest requires several accomplishment to operate, but the fancy new magic box? You don’t have to cognize how to peruse or write. A caveman could build a B2B SaaS and be on period for his weekly podcast appearance. And if you think your tradition delegate orchestration workflow is an crucial feat of engineering, fair you wait.
Agents weaken the social cloth of engineering teams by reducing interpersonal communication, expanding self-reliance, and hindering accomplishment expression.
Slack channels used to be crowded alongside people. Anything that your rubber duck couldn’t answer would go into a squad chat. Now the conversation is a faint shade of its erstwhile self. Everyone is too occupied talking to their agents. Each of us has received a magic acquainted that follows us about akin we’re several benevolent of Disney protagonists. It’s extremely smart, accessible 24/7, and volition never study you to HR. The rubber duck can eventually conversation back!
In a certain sense, being additional autonomous is virtuous. Why distract colleagues alongside your silly nonsense? The issue is, assisting another group alongside silly nonsense is our thing. It’s how we build and keep relationships. You need to stroll the path; otherwise, it volition be overtaken by grass.
I’m convinced that AI has dramatically reduced interpersonal communication. The average squad conversation correct now consists of delegate threads and an occasional LLM-generated manuscript followed by a humble “what do you guys think?”
Agentic coding additionally changes the way we see another people. Teams used to be kind of akin parties of adventurers: a Git wizard, a Rust witch, a bard who is really into mechanical keyboards. There was plentifulness of area for accomplishment expression, and the assortment of ability was celebrated.
Now, an average squad consists of a Claude operator, a Codex whisperer, and a guy alongside AI psychosis who insists on committing .md files. “This is John—you’ll desire to see how he prompts,” stated nobody ever.
And that’s true: nobody wants to see how John prompts. There is item repulsive and basically unsexy concerning talking to an agent. It feels fine whenever you are doing it, but others... ugh, get a chatroom. Their prompts are cringe; their agents are weird. Like looking at person else’s TikTok feed.
So current we are, motionless peers in the org chart, but expanding additional and additional distant complete time. I used to praise your Vim skills and Advent of Code solutions. These days you immediate Claude all day, and I do too. Are we equal a squad anymore?
So that’s my list. And I’ll be honest alongside you, I have no idea what to do alongside any of this. I can lone anticipation that eventually we’ll move onward in this cursed discussion. However beneficial AI is, we pay dearly for it alongside our curiosity, craftsmanship, and social connection.
