4 min Devops

Three-quarters of developers haven’t warmed to AI code assistants (yet)

Three-quarters of developers haven’t warmed to AI code assistants (yet)

AI complements the programmer in a quarter of organizations, while two-thirds will implement the technology in the future. According to GitLab, only one conclusion can be drawn: “AI in software development is here to stay.”

The latest annual survey by DevSecOps platform GitLab probed companies’ attitudes and implementation toward developing using an AI assistant. That’s an obvious topic, considering there are an increasing number of generative AI solutions aimed at programmers. The field has expanded considerably in the last year from GitHub Copilot and Google Project IDX, Meta Code Llama and StableCode LLM were all made.

In fact, it is also possible to get started with “general” generative AI tools as a developer, but the accuracy of those outcomes is often questionable. However, the technologies offered on coding platforms are also far from perfect, and plagiarism and generating faulty code are still common.

Also read: ChatGPT writes incorrect (but convincing) code half the time

Improvements are slow

Companies will have closely watched the whole emergence of these tools, and a quarter decided to use these tools to supplement the programmer’s work right away. This way, the developer’s control remains at least.

In turn, two-thirds of the companies say they plan to implement the tools sometime in the next three years. Perhaps they are waiting until the biggest early development mistakes are filtered out. Small steps have already been taken in this area. For example, GitHub has already made its AI assistant Copilot smarter by having the assistant send an alert to the developer when it plagiarizes code. However, the tool has been available since June 2022, and it took almost a year for this improvement to come. So improvements are coming, but tremendously fast are things not going.

AI offers help with various tasks

The developers who do already use generative AI in their work also appear, in many cases, to be staying away from the more risky uses in which the assistant generates code itself. 36 percent of respondents said they use generative AI for this purpose.

More common was the use of the AI assistant for checking code (62 percent). Although only 36 percent indicated that the AI performs this task before the developer himself has already looked at the code.

Developers are just not spending whole days writing code themselves either. In the survey, respondents say they spend about a quarter of their time writing code. The range of tasks is varied and the AI assistant may take over tasks from this varied range of tasks.

AI remains (to the dismay of developers)

Even though many companies are still somewhat reluctant, sooner or later the AI coding assistant will enter many companies. Consequently, the report sees a future for generative AI in development platforms: “If there is one inescapable conclusion from the survey data, it is that AI in software development is here to stay.”

Are developers however waiting for this evolution? Around the use of generative AI in software development, respondents still appear to have a few concerns. The biggest concern was that AI will make developers redundant or cause fewer jobs for developers will remain. 29 percent of respondents agreed.

Nearly half of respondents further expressed concern that AI-generated code will not be copyrighted. Nearly 40 percent fear that bringing an AI coding assistant to the company is equivalent to bringing security risks to the company. The same reasoning also caused three-quarters of companies to ban ChatGPT from the company.

Developers demand training

What could potentially already allay fears about these tools is training around using an AI coding assistant in a business environment. Developers are also asking for these training opportunities themselves, with 81 percent saying that more training is needed on how to use AI in the workplace.

That could possibly temper the anxiety surrounding the tools, but human control will always remain according to GitLab: “AI may be able to generate code faster than a human developer, but a human team member must verify that the AI-generated code is free of errors, security issues or copyright issues before it goes into production.”

GitLab concludes these findings from a survey that successfully reached approximately 5,000 individuals worldwide. A clear majority of respondents were male (71 percent) and between the ages of 18 and 34 (51 percent). The profiles were more varied, but most of the participants were software developers or engineers.

So the embrace of generative AI by developers is cautiously on the rise. Developers are still concerned about this evolution, though, and fear for their jobs and security risks. Yet the AI tool appears to be a solution that can support developers in many areas of their job. To ensure a smooth introduction and address developers’ concerns, developers are asking companies to invest in training.