If you read the press releases, AI coding adoption is a straight line upward. Every company is “going all-in.” Every developer is “vibe coding.” The future is decided.
Then you open Hacker News.
Two massive threads this week — Ask HN: Anyone still coding like 2021? (90 comments) and I’m going back to coding by hand (46 comments) — tell a completely different story. Together, they paint a picture of a developer workforce fractured by AI adoption: people being fired for their stance, others faking compliance, entire sectors refusing to budge, and a growing movement of developers deliberately returning to hand-coding not because they don’t understand AI, but because they’ve used it and found it wanting.
This isn’t a technical debate. It’s a culture war.
The Myth of Universal Adoption
The first thing the threads demolish is the idea that everyone has already switched. A significant number of respondents described workplaces where AI coding tools simply don’t exist — and not because of ignorance.
A defense contractor explained that frontier models are “verboten anywhere in the building,” and even running an open-weight model on your laptop requires a strict approval process. A medical device developer described processes that change “very slowly” because of heavy regulation and legal reviews that take months. A cybersecurity lead said their company only just started integrating Copilot into code review pipelines, with no agentic coding on the horizon.
One developer at a “significant computer security software vendor with 5-digits of employees” described their reality: AI tooling is provided and encouraged, but “it’s not like ‘use this or we will fire you’.” They use it “a couple times a week” for code review and boring boilerplate.
Meanwhile, a government developer in California noted that the public sector has “a major reluctance to use techniques and tools that haven’t been proven out for like 5+ years.” Their coding practices aren’t even at 2021 best-practices level yet.
The pattern: the industries with the most to lose — defense, medical, government, security — are the slowest to adopt. Not because they can’t see the potential, but because they understand the risk in ways Silicon Valley doesn’t.
The Workplace Politics Are Brutal
The darkest thread in these discussions isn’t about code quality. It’s about what happens to developers who push back.
One developer described being fired — explicitly for refusing to use LLMs. “I was fired for this stance, not for performance reasons (officially stated that there was no concern with my output), only for political reasons (insubordination, refusing to use LLMs to do my work for me).”
The responses to this were split, and the split itself is revealing. Some said: “You don’t HAVE TO do anything your employer tells you to do. And they don’t HAVE TO keep you employed.” Others pushed back harder: “The job is to make widgets, not to use specific tools to make widgets. When has that ever been the case before? None of my bosses ever gave a shit what editors I used.”
But the fired developer made a deeper point: after three years of their company diving into AI coding, they were “one of the only people in my team able to critically analyse changes and plan complex work.” The employer, however, didn’t value that. What they valued was compliance.
Other developers described subtler versions of the same dynamic:
- One said they were “pretending to drink the koolaid, faking interest in agentic coding while I figure out something else to do.”
- Another described chaos: “One day it’s allowed, the next day IT security guys have blocked it. Local models are also not allowed. I’m carrying on coding everything myself, because I can’t keep changing my workflow.”
- A contractor situation revealed a strange asymmetry: “Our contractors can [use AI]. They wrote it into their contracts. It’s not even fair, really.”
The thread about a CTO who went “full AI psychosis” was particularly telling. “Directive from the top was to ‘use AI for everything’ with no real regard for if it helped productivity, service delivery, or quality of output. That was very quickly reigned back in when they saw the bill. Multiple people using tens of thousands of dollars of tokens in a week, with results that were trash.”
“Claude Said” Is the New “Stack Overflow Says”
One comment captured a cultural shift that’s arguably more damaging than any technical limitation: the death of the engineering argument.
A developer described the new reality at their company: “The transition of winning an argument changed to ‘claude said…’ which is fucking annoying.” The codebase was “becoming a blackbox where god and AI know what it’s like — every function has 4-line paragraphs historically describing changes from past to the future.”
Another respondent connected this to a pattern of brain decay that goes beyond coding skill: “Sycophancy makes me overestimate myself. When my stress meets external pressure, quality goes down. The little voice of ‘surely the AI got it right’ can lead to defects and code quality going down.”
A manager running quarterly AI trials described the cycle: “Every four months or so we try getting hyped on agents for about a week. Up to this point the answer has always been a pretty solid ‘nope, this sucks, back to manual code.’” But they acknowledged the current tools are getting better — the cycle is getting shorter, even if the endpoint hasn’t changed yet.
The worst report came from a developer whose team’s code is now mostly AI-written: “Every change, even a few lines long, will trigger 25–50 additional file changes just because everything is so horribly written. We have so many useless tests, so many repetitive code.” They estimated the real productivity gain was “nearly 3x, and that’s assuming every new MR matters” — far short of the 10x executives were promised.
The Deliberate Return Movement
The second thread — “I’m going back to coding by hand” — is about something different. These aren’t people who never adopted AI. They’re people who did, and chose to come back.
The original poster was clear: “The performance was never in question — it’s the control I missed.” They described variable name changes that required “chasing down all the places they were referenced” — something that should be trivial in a well-understood codebase but becomes opaque when you can’t remember what the AI actually changed.
The thread revealed several distinct motivations for the return:
Identity and craft. A graphics programmer put it bluntly: “I would not disrespect the years I’ve put into this craft by outsourcing my thinking. Nor do I want to wake up in ten years and realize I have willingly made myself stupid through overuse of these tools.” Another: “Coding used to be fun when I used to write things by hand, the complete project mental model was always there in mind. With LLMs, output has increased but that end-to-end mental model is lost.”
Control over fundamentals. Multiple developers described the same pattern: they build the base manually, then let AI iterate. “I build the base manually and then I can iterate with AI faster once the fundamentals are solid. Sometimes the AI tends to forget those fundamentals or tries to change those and add unnecessary stuff on top.”
The practical compromise. The most nuanced take came from a developer who described the three modes: “By hand, hands off, and AI assistance while you code by hand. I’ll do any of the three depending on what I’m building or how I feel. Why does it have to be a hard line?”
The mental gym. One developer started doing LeetCode every morning to stay sharp: “Even though agents are building all of my code now, I’m starting every morning with a LeetCode problem. There is no practical use, but it makes me feel good and sharp for the rest of the day.” Others pushed back — LeetCode doesn’t exercise debugging, architecture, testing, or any of the skills that atrophy. A 30-60 minute daily hand-coding session was suggested as a better alternative that “helps me be a bit more critical and questioning of decisions made in the code.”
The Real Insight: It’s Not About the Tools
The unifying theme across both threads is this: the debate was never about whether AI coding tools work. It’s about who gets to decide, what gets lost, and whether developers are treated as craftspeople or as interchangeable token-processing units.
The developer who quit a 10-year contract put it in the starkest terms: “I would rather stare at a wall than ‘manage’ an agentic team.”
The developer who got fired said: “The job is to solve a problem, implement a solution. As a developer I should have some leeway on how I accomplish that.”
The developer faking compliance said: “I’m pretending to drink the koolaid while I figure out something else to do.”
These aren’t luddites. They’re not afraid of new tools. They’re developers who watched their codebases become unrecognizable, their engineering debates reduced to “the AI said,” and their value proposition shift from “I can build things” to “I can tell the AI what to build.” That’s a different job — and many of them didn’t sign up for it.
What It Means for You
If you’re a developer navigating this landscape, the threads suggest a few practical takeaways:
-
You’re not alone if you’re resisting. A huge chunk of the industry hasn’t fully adopted AI coding tools, and many of those holdouts are in serious, consequential industries — not just laggards.
-
The political dimension is real. “Just use what your employer tells you” is clean advice, but it ignores the fact that mandated AI adoption often comes with contradictory policies, no training, and no understanding of the tradeoffs.
-
Hand-coding is not regression. The developers going back to manual coding are doing it deliberately, with clear reasoning about control, understanding, and quality. The best approach seems to be hybrid: build the foundation yourself, let AI handle iteration and boilerplate.
-
Watch for the “claude said” problem. If your team has replaced engineering debates with AI authority, that’s a red flag — not for AI adoption, but for engineering culture.
-
The middle ground exists. The most sustainable developers described a pragmatic mix: AI for exploration and boilerplate, hand-coding for architecture and critical paths, and daily practice to keep skills sharp. No ideology — just what works.
This article synthesizes developer opinions from Hacker News threads 49626431 (90 comments) and 49622554 (46 comments), posted September 9–10, 2026. All quotes are paraphrased from real comments. Original threads contain 136+ total comments.