Skip to content

State of AI 2026

Usage

How developers use AI.

Use Cases

While code generation still comes first, it's interesting to note that code review (a new option this year) has jumped straight to third place. After all, those 30k lines of codes that got vibe-coded over the week-end will need to be double-checked at some point…
What do you use AI tools for?
Multiple
0%
20%
40%
60%
80%
100%
01
Code generation
5,771
02
Code review & assistance
+8
4,359
03
Learning & research
-1
4,356
04
Debugging
4,277
05
Summarization
-1
3,623
06
Text generation
-3
3,502
07
Translation
-2
2,761
08
Image generation
-2
2,388
09
Computer vision
-2
949
10
Speech recognition
-2
831
0%
20%
40%
60%
80%
100%
6,421
88%
5.5
10
8%
Percentage of question respondents who stated the entirety of their coding output is now AI-generated.

AI Code Generation

The difference with last year is striking here. Back then the average respondent was using AI sporadically to generate a small percentage of their code, but today that same developer is much more likely to rely on AI for the bulk of their coding.

While AI usage is up across the board, it's interesting to note that higher usage brackets are correlated with lower median years of programming experience. And as you'd expect, usage of a coding agent such as Claude Code is correlated with a higher proportion of AI-generated code.

What proportion of the code you produce is AI-generated?
0%
20%
40%
60%
80%
100%
1
0% AI
532
2
12.5%
688
3
25%
798
4
37.5%
404
5
50%
629
6
62.5%
523
7
75%
1,145
8
87.5%
1,203
9
100% AI
498
0%
20%
40%
60%
80%
100%
6,420
88%
54%
63%
10

AI Code Refactoring

Similarly, using AI-generated code without having to refactor it first is no longer the exception, with many respondents reporting only needing to refactor around 25% of AI output.
When using AI to generate code, what proportion do you rewrite or refactor before use?
0%
20%
40%
60%
80%
100%
1
0% refactored
312
2
|
825
3
|
1,184
4
|
565
5
50%
990
6
|
521
7
|
956
8
|
486
9
100% refactored
288
0%
20%
40%
60%
80%
100%
6,127
84%
47%
50%
10

Reasons for Refactoring

When respondents do need to refactor AI-generated code, the culprits tend to be poor code style and hallucinations. Interestingly, variable renaming does not seem to be as much of a factor anymore, pointing to the fact that LLMs are getting better at following instructions and learning from existing codebases.
What are the main reasons that require you to refactor AI-generated code?
Multiple
0%
20%
40%
60%
80%
100%
01
Poor code style
+9
3,197
02
Hallucination & inaccuracies
+6
3,121
03
Poor readability
-2
2,816
04
Excessive repetition
-1
2,543
05
Faulty code
+2
2,336
06
Variable renaming
-4
1,862
07
Performance issues
-2
1,412
08
Outdated imports
-4
1,179
09
Security issues
-3
1,154
10
🚫 None
237
0%
20%
40%
60%
80%
100%
5,457
75%
3.5
20
10
61%
Percentage of question respondents who use AI to generate code on a daily basis.

Code Generation Frequency

It seems clear that AI has become embedded in our work, with a majority of respondents using it multiple times per day, and a growing segment using it constantly.

And the more frequently you use AI, the lower your median number of years of programming experience is likely to be.

How frequently do you use AI tools to generate or refactor code?
0%
20%
40%
60%
80%
100%
1
Never
503
2
A few times per year
266
3
A few times per month
525
4
A few times per week
1,191
5
A few times per day
1,776
6
A few times per hour
774
7
Constantly
1,366
0%
20%
40%
60%
80%
100%
6,401
88%
3.8
4
10

Other Tasks Frequency

While the main focus of this survey is coding, other AI uses are also seeing increased used compared to last edition. After all if you trust AI with the main part of your job, it does seems logical to trust it with everything else, too.
How frequently do you use AI tools for tasks other than code generation?
0%
20%
40%
60%
80%
100%
1
Never
483
2
A few times per year
237
3
A few times per month
645
4
A few times per week
1,873
5
A few times per day
2,200
6
A few times per hour
378
7
Constantly
622
0%
20%
40%
60%
80%
100%
6,438
89%
3.4
3
10

Generated Code

Almost every type of generated code saw an increase compared to last survey, with tests in particular becoming a key reason to use AI.
What kind of code do you generate using AI tools?
Multiple
0%
20%
40%
60%
80%
100%
01
Helper functions
4,715
02
Frontend components
4,434
03
Tests
+1
4,416
04
Documentation & comments
-1
4,066
05
Core app logic
3,730
06
API integration code
3,701
07
Scripting
+1
3,500
08
CSS code
-1
3,460
09
Configuration code
2,807
10
Database queries
+1
2,774
11
🚫 None
422
12
All of the above
13
Other Answers
141
0%
20%
40%
60%
80%
100%
6,281
87%
6.4
10

Personal Expenses

The proportion of respondents enjoying the benefits of AI without personally having to spend their hard-earned cash has dramatically shrunken year-over-year; while the $100-$500 monthly spend demographic is seeing significant growth. This may be due to the rise of coding agents, which often carry heavier financial costs compared to simple chatbots.
How much do you spend on AI tools per month (in USD)?
0%
20%
40%
60%
80%
100%
1
$0
2,532
2
$1-$20
1,428
3
$20-$50
1,232
4
$50-$100
447
5
$100-$500
669
6
$500-$1000
41
7
$1000-$5000
23
8
Cutoff Answers
-2
6
0%
20%
40%
60%
80%
100%
While the 2026 survey also included a question about company AI spend, its data wasn't usable because it didn't clearly state whether the question was about total company-wide spend, or per-employee spend.
6,378
88%
10

Industry Sector

Don't hesitate to use our built-in Query Builder on any other chart to filter these survey results according to any specific industry sector.
Which industry sector(s) do you work in?
Multiple
0%
20%
40%
60%
80%
100%
01
Programming & Technical Tools
2,900
02
Consulting & Services
1,525
03
E-commerce & Retail
749
04
Finance
662
05
Marketing/Sales/Analytics Tools
+1
538
06
Education
-1
516
07
Healthcare
347
08
Entertainment
338
09
Government
+1
267
10
News, Media, & Blogging
-1
228
0%
20%
40%
60%
80%
100%
5,686
78%
1.5
20
10

Local AI

More and more respondents have tried local AI models, and while they aren't as powerful as their cloud-based cousins right now, there is strong potential for disruption as they provide a cheaper alternative to traditional frontier labs models.
Have you tried running AI models locally on your own machine?
0%
20%
40%
60%
80%
100%
1
No, not interested
1,291
2
No, but interested
2,012
3
Yes
3,118
4
Other Answers
35
0%
20%
40%
60%
80%
100%
6,421
88%
10

Recommended Resources

Sentry MCP

Sentry MCP

Sentry MCP plugs Sentry's API directly into your LLM, letting you ask questions about your data in natural language.
Sponsored content from our partners. Learn more.