AI
How much of the code shipped in Korea is written by AI, which tools people reach for, and how they feel about it.
70%
median share of shipped code that is AI-generated
74% use AI to generate code at least daily
Starts with 3 cross-tabs, or jump to the 7 individual answers ↓
Cross-tab
AI writes most of everyone's code; only the enthusiasm differs
AI-written code and belief in AI, by years of experience
The median share of shipped code written by AI is 70% under four years, 80% at four to six, 70% at seven to nine, and 65% at ten or more. Agreement that AI makes them more productive slides from 82% to 68% across the same buckets. Both spreads sit inside their margins of error, so read this as no clear relationship rather than as a trend.
Median share of shipped code written by AI
0–3 years
4–6 years
7–9 years
10+ years
Agree AI makes them more productive
0–3 years
4–6 years
7–9 years
10+ years
The share is self-reported with no shared definition of what counts as AI-written, so compare the buckets against each other rather than treating the level as exact.
Groups differ in more than the split shown, so read these as descriptions, not causes.
Cross-tab
Korean respondents let AI write far more of their code
How much AI writes, by respondent group
The Korean developers here report a median of 90% of shipped code written by AI, against 60% for international residents, and 72% of them use AI to write code constantly against 29%.
Median share of shipped code written by AI
Korean, in Korea
International, in Korea
Use AI to write code constantly
Korean, in Korea
International, in Korea
25 Korean respondents, self-reported, with no shared definition of what counts as AI-written. Both groups arrived through the same channels, so this compares two small self-selected samples, not two countries.
1 group with fewer than 8 answers are not shown (35 respondents).
Groups differ in more than the split shown, so read these as descriptions, not causes.
Cross-tab
Claude Code everywhere, Copilot mostly among internationals
Which AI tools each group reaches for
88% of Korean respondents use Claude Code against 69% of international residents, while GitHub Copilot runs the other way, 12% against 29%.
Use Claude Code
Korean, in Korea
International, in Korea
Use GitHub Copilot
Korean, in Korea
International, in Korea
Dev Korea's own channels shape this tool mix. Read it as what this community reaches for, not as market share.
1 group with fewer than 8 answers are not shown (35 respondents).
Groups differ in more than the split shown, so read these as descriptions, not causes.
Every question in this chapter
AI tools
131 answered · multiple answers allowed
Claude Code
Claude
ChatGPT
Gemini
Codex
Cursor
GitHub Copilot
DeepSeek
Grok
Perplexity
Qwen
Other
DeepL
Stable Diffusion
Midjourney
v0
Devin
What AI is used for
131 answered · multiple answers allowed
Code generation
Learning/research
Summarization
Translation
Text generation
Image generation
Computer vision
Speech recognition
Speech generation
Other
Share of shipped code that is AI-generated
130 answered
0%
1–24%
25–49%
50–74%
75–99%
100%
How often AI generates their code
131 answered
Never
A few times a year
Monthly
Weekly
Daily
Multiple times a day
Constantly
Biggest AI pain points
131 answered · multiple answers allowed
Hallucinations/inaccuracies
Poor generated code
Cost
Lack of context
Misunderstood prompt
Privacy concerns
Environmental impact
Integration difficulties
Training-data concerns
Ethical concerns
Other
"AI makes me more productive"
131 answered
Strongly disagree
Disagree
Neutral
Agree
Strongly agree
"AI is a threat to my job security"
131 answered
Strongly disagree
Disagree
Neutral
Agree
Strongly agree