Artificial Intelligence
Share:
The questions in this section were answered by those who decided to continue taking the survey after the main question section ended.
As this was the first time we asked questions about AI in the Developer Ecosystem Survey, the section does not provide comparisons with previous years.
In general, developers appear to be rather optimistic about the rapid advancements in AI and are actively using its capabilities in their work.
Share:
![](/lp/devecosystem-2023/static/maria-khalusova-cbd1594a7dfe2f239af9dc318e0fd068.png)
Maria Khalusova
Member of Technical Staff, Hugging Face
It’s encouraging to see the developer community is mostly enthusiastic toward the application of AI-based tools to software development. At the same time, it’s important to acknowledge the presence of security and ethical concerns. This serves as a testament to the community's awareness of the existing limitations and potential hazards associated with these technologies. A key to addressing these concerns lies in advocating for greater transparency in the way AI systems are made available. Fully open AI models – and entire systems – enable community scrutiny that helps identify potential issues and contributes to continuous improvements in these systems. Ultimately, this makes AI-based tools both more useful and more trustworthy.
84%
of developers are familiar with generative AI tools in one way or another.
Interestingly, AI text generative tools are more familiar to developers than code generative tools, which might be attributable to the popularity and accessibility of ChatGPT.
84%
of developers are familiar with generative AI tools in one way or another.
![](/lp/devecosystem-2023/static/svetlana-zemlyanskaya-317f70ba59e7019b8bdd2a6019e33744.png)
Svetlana Zemlyanskaya
Team Lead in Machine Learning in IDE Assistance, JetBrains
In just a couple of years, AI-based code generation tools went from being an interesting research topic to being an important part of many developers' toolboxes. We’ll continue monitoring this trend closely to see how it evolves.
*This question was shown only to the developers who chose “None” in the previous question.
Only 1.6% of all respondents have never heard of generative AI tools, and slightly less than half of those who have not yet familiarized themselves with generative AI tools plan to do so in the near future.
Security concerns are the most frequently cited obstacle to the adoption of generative AI tools.
![](/lp/devecosystem-2023/static/svetlana-zemlyanskaya-317f70ba59e7019b8bdd2a6019e33744.png)
Svetlana Zemlyanskaya
Team Lead in Machine Learning in IDE Assistance, JetBrains
Most solutions still require sending data to the cloud, which makes them a security risk, but the market is already responding with local and on-premises solutions.
Agree | Neither agree nor disagree | Disagree | |
---|---|---|---|
59% | 30% | 11% | I have security concerns about using generative AI services |
53% | 30% | 16% | I am ready to use cloud-based generative AI services for work tasks |
42% | 33% | 25% | I have ethical concerns about using generative AI services |
40% | 40% | 20% | Local or offline AI tools are unlikely to reach the quality and performance of cloud-based solutions |
28% | 35% | 37% | My company’s policy limits the use of cloud-based AI tools |
19% | 37% | 43% | I am worried that AGI (artificial general intelligence) will be hostile to humans |
Our respondents appear to have a positive view of AGI (artificial general intelligence) in general. Less than one-fifth of them are worried that AGI will become hostile to humans. However, 6 out of 10 respondents have security concerns about using AI.
![](/lp/devecosystem-2023/static/svetlana-zemlyanskaya-317f70ba59e7019b8bdd2a6019e33744.png)
Svetlana Zemlyanskaya
Team Lead in Machine Learning in IDE Assistance, JetBrains
Large language models have created many ethical issues that are yet to be addressed, like the source of training datasets, fair use of open-source code, and others.
Agree | Neither agree nor disagree | Disagree | |
---|---|---|---|
60% | 30% | 10% | AI coding tools will radically change the software development job market |
51% | 33% | 16% | The adoption of AI coding will raise the demand for professional software developers |
51% | 29% | 20% | Some industries will never adopt AI coding |
49% | 37% | 15% | Employers will expect every software developer to be proficient with AI coding tools |
33% | 34% | 33% | The majority of coding will turn into prompt engineering |
13% | 27% | 61% | AI will fully write code instead of developers |
Despite the fact that 3 out of 5 respondents believe AI coding tools will radically change the software development job market, only 13% are confident that AI is going to fully write code in place of developers. Nevertheless, about one-third believe that software engineering is bound to turn into prompt engineering.
In general, developers are rather optimistic and believe that AI will become a new additional tool to help them write code as opposed to something that will replace them entirely.
Cloud-based services
Local or offline solutions
Cloud-based solutions managed by you or your organization
Other
I don’t know
The majority of developers use the potent generative AI cloud-based solutions – cloud-based services or solutions managed by the respondent’s organization.
![](/lp/devecosystem-2023/static/svetlana-zemlyanskaya-317f70ba59e7019b8bdd2a6019e33744.png)
Svetlana Zemlyanskaya
Team Lead in Machine Learning in IDE Assistance, JetBrains
Local and on-premises solutions are already emerging, but the final quality is often worse than that of cloud-based solutions. While the quality is improving across the board, the gap between local and cloud-based solutions will probably remain unbridged for the next couple of years.
I use it | I tried it, but don’t use it now | I don’t use it | |
---|---|---|---|
77% | 20% | 3% | ChatGPT |
46% | 33% | 21% | GitHub Copilot |
26% | 50% | 25% | Midjourney |
26% | 15% | 59% | Visual Studio IntelliCode |
21% | 42% | 37% | OpenAI DALL-E |
17% | 21% | 62% | Dream Studio (Stable Diffusion) |
9% | 20% | 71% | Tabnine |
7% | 11% | 81% | Pictory |
7% | 17% | 76% | Synthesia |
5% | 9% | 85% | Amazon CodeWhisperer |
4% | 7% | 89% | CopyAI |
4% | 8% | 88% | AIVA |
4% | 10% | 86% | Soundraw |
4% | 7% | 88% | Boomy |
3% | 5% | 91% | Codeium |
3% | 11% | 85% | Kite |
2% | 7% | 90% | Jasper |
2% | 6% | 92% | Replit Ghostwriter |
1% | 3% | 96% | Sourcegraph Cody |
1% | 3% | 96% | Atlassian Intelligence |
Developers are using general AI text generative tools more often than they do specialized AI code generative tools. However, since the latest AI text generative tools, such as ChatGPT, are also capable of writing code, developers may be using them for this purpose. The specialized tools mentioned above seem to be drawing interest, but whether it's because of imperfect integration with the workflows or generic approaches, few developers are sticking with them at this time.
Quite often | From time to time | Rarely | Never | |
---|---|---|---|---|
26% | 33% | 17% | 24% | Asking general questions about software development in natural languages |
24% | 37% | 24% | 15% | Generating code |
19% | 26% | 22% | 33% | Generating code comments or code documentation |
18% | 26% | 21% | 36% | Explaining bugs and offering fixes for them |
14% | 27% | 22% | 37% | Explaining the code |
12% | 21% | 24% | 42% | Generating tests |
11% | 21% | 19% | 48% | Search in natural language queries for code fragments |
9% | 17% | 21% | 53% | Performing code review |
9% | 16% | 19% | 55% | Summarizing recent code changes to understand what happened more quickly |
9% | 20% | 23% | 47% | Refactoring code |
9% | 17% | 20% | 54% | Generating CLI commands by natural language description |
6% | 12% | 20% | 62% | Generating commit messages |
The most common way for developers to use an AI assistant is to ask general questions about software development using natural language.
Regularly | From time to time | Never | |
---|---|---|---|
35% | 47% | 18% | Learning new things |
26% | 44% | 29% | Brainstorming and evaluating ideas |
24% | 42% | 34% | Summarizing content |
23% | 35% | 42% | Proofreading content (e.g. fixing spelling and grammar mistakes) |
21% | 34% | 45% | Rewriting content in a desired style (e.g. more friendly or more consistent with your company’s brand voice) |
21% | 36% | 44% | Translating texts |
20% | 39% | 40% | Generating content not directly related to code (websites, release notes, tweets, etc.) |
11% | 21% | 69% | Preventing the use of inappropriate language |
AI tools are popular learning assistance and brainstorming buddies. They are also useful for summarizing and proofreading content. As far as generating non-code content, only 20% of our respondents regularly use AI tools for that purpose.
*Shares of respondents that selected each activity as one of their three most time-consuming activities.
Writing code takes the top spot as the most time-consuming activity, though it is also the most enjoyable one for our respondents (see the chart below). Spending most of your time doing something you enjoy – isn’t that the recipe for happiness?
Enjoyable | Neither enjoyable nor unpleasant | Unpleasant | |
---|---|---|---|
82% | 14% | 3% | Writing code |
51% | 36% | 14% | Understanding the code |
45% | 36% | 18% | Refactoring |
40% | 49% | 11% | Internet searches |
35% | 50% | 15% | Writing code comments or code documentation |
34% | 43% | 24% | Debugging |
33% | 49% | 18% | Performing code reviews |
31% | 55% | 14% | Performing actions in CLI |
29% | 55% | 16% | Understanding recent code changes |
27% | 45% | 28% | Writing tests |
27% | 59% | 14% | Writing commit messages |
22% | 56% | 23% | Searching for code fragments inside the codebase |
If writing code is a developer's most enjoyable activity, it’s natural that they’re not ready to delegate it to the AI tools!
Simple | Neither simple nor difficult | Difficult | |
---|---|---|---|
54% | 38% | 8% | Writing commit messages |
54% | 39% | 8% | Internet searches |
46% | 43% | 11% | Writing code comments or code documentation |
40% | 48% | 11% | Writing code |
36% | 49% | 14% | Searching for code fragments inside the codebase |
36% | 53% | 11% | Performing actions in CLI |
26% | 58% | 16% | Understanding recent code changes |
25% | 51% | 25% | Writing tests |
25% | 56% | 19% | Performing code reviews |
24% | 54% | 22% | Understanding the code |
23% | 51% | 26% | Refactoring |
22% | 51% | 27% | Debugging |
How likely is it that you would delegate the following activities to an AI assistant (in an ideal world where the performance of an AI assistant is humanlike)?
I would delegate it | I am not sure yet | I would still do it myself | |
---|---|---|---|
56% | 23% | 21% | Writing code comments or code documentation |
56% | 26% | 18% | Writing tests |
55% | 26% | 19% | Searching for code fragments inside the codebase |
50% | 23% | 27% | Writing commit messages |
46% | 23% | 31% | Internet searches |
35% | 34% | 31% | Performing actions in CLI |
34% | 31% | 35% | Performing code reviews |
34% | 31% | 35% | Refactoring |
31% | 28% | 41% | Understanding recent code changes |
30% | 30% | 40% | Debugging |
23% | 26% | 51% | Understanding the code |
17% | 28% | 54% | Writing code |
AI assistants are most commonly used to help developers perform routine tasks, like writing documentation, code comments, and commit messages, as well as searching. However, developers prefer to do their own coding, including understanding the code and recent code changes, debugging, and of course, writing code, even though 79% of the respondents mentioned that writing code is their most time-consuming activity.
Writing quality code is a difficult task, and generative AI coding tools are showing some evidence of making it less time intensive. While approximately one-third of developers surveyed remain unsure about delegating tasks to these tools, this doesn’t mean they aren’t using the tools to complete tasks collaboratively rather than completely outsourcing the work.
JetBrains AI Assistant
AI Assistant provides AI-powered features for software development. The JetBrains AI service transparently connects IDE users to different large language models (LLMs). AI Assistant is context-aware and helps developers complete their tasks faster, boosting productivity.
Thank you for your time!
We hope you found our report useful. Share this report with your friends and colleagues.
If you have any questions or suggestions, please contact us at surveys@jetbrains.com.