AI Coding Assistants in 2026: Worth It? What 1,262 HN Commenters Really Think
The short version: The "everyone is a 10x developer" era is over and the measured truth is boring: AI coding assistants are worth it for experienced devs in familiar codebases, marginal for juniors, and actively dangerous without review discipline. The landmark posts of late 2025–2026 — "AI coding assistants are getting worse?" (451 pts) and "After two years of vibecoding, I'm back to writing by hand" (865 pts) — aren't anti-AI manifestos; they're the pendulum settling after peak hype. The 10%-productivity survey is closer to the real number than the 10x claims.
The context: the pendulum swung twice
2024–2025 was the "vibe coding" era: agents generating whole apps, everyone claiming 10x. The 2026 correction was brutal and public. Two threads define it: the question of whether assistants are getting worse, and a very liked post about a two-year vet going back to writing code by hand. We read both threads plus the surrounding ecosystem — 1,262 substantive comments — and the middle ground is surprisingly consistent.
The bull case (smaller than it was, still real)
Even in the correction threads, the win conditions are concrete. Autocomplete and routine code still carry their weight:
"Reviewing code can be hard but it's not as hard as writing the code. Even with the best autocomplete and ergonomic editors, it still takes quite a bit of time to write code compared to the actual concepts being implemented."
And the 10x crowd's defense, in the "getting worse" thread, was mostly the "secret workflow" argument — which the skeptics immediately turned into a burdden-of-proof point:
"Anything you say can be countered by their 'secret workflow'. If you're not seeing a huge speedup, well, you're just using it wrong! The burden of proof is 100% on anyone claiming the productivity gains."
The bear case (the part vendors won't quote)
1. The measured gains are modest. A Feb 2026 survey thread reported productivity gains "haven't budged past 10%". One anecdote in the thread: a dev watched someone struggle for an hour with a simple extension his assistant should have handled — and realized doing it by hand would've taken the same hour.
"He didn't manage to finish in the hour what he was planning to. When I thought about how long it would have cost me by hand, I found it would have taken me just as long."
2. Agents now delete things. The March 2026 "Claude Code wiped our production database with a Terraform command" thread (145 pts) is the genre-defining cautionary tale. This is the risk that didn't exist with autocomplete — you're no longer just accepting suggestions; you're approving autonomous actions.
3. The 10x claims are unfalsifiable. A recurring commenter frustration: "I'd like to see a live coding session from one of these 10x AI devs, and how it compares to their coding without AI." (godelski, HN)
The honest verdict
Worth it if:
You're an experienced developer working in a codebase you understand, you use assistants as an autocomplete/boilerplate/rubber-duck layer, and you review everything. The "getting worse" complaints correlate heavily with people letting agents drive unfamiliar territory.
Not worth it if:
You're learning, or working in a domain where you can't judge the output. Assistants give juniors confident mistakes faster. If a wrong fix takes you an hour to detect, the assistant is costing you money.
The practical rule:
Treat agent autonomy like production access: scope it, gate it, review it. The productivity is real at the margins; the disasters are real at the extremes. Expect the ~10-20% range, not 10x — and budget accordingly.
Method: 1,262 comments read across the two landmark 2026 threads (451p "getting worse", 865p "back to writing by hand") plus ecosystem threads (survey, production incidents), categorized 2026-08-27. Quotes verbatim with links. We don't take payments from vendors — this site survives on being useful.