A survey of 555 software executives published this week finds that 75% have seen up to a 50% reduction in development time by implementing various artificial intelligence (AI) and automation technologies. The survey, conducted by the CIO Dive arm of studioID on behalf of Outsystems and KPMG, finds more than half (56%) said they experienced or expected to experience a higher quality of applications, with fewer bugs and improved performance. One-third of respondents (33%) said they had a backlog of between 150 and 800 use cases for generative AI specifically.
Those use cases for generative AI include DevOps optimization (59%), code generation (58%), documentation (56%), and user-interface design (50%). Rodrigo Coutinho, AI Project Manager at Outsystems, said most organizations are using generative AI today for writing code and testing, and that it may be a while yet before it is applied across the entire software development lifecycle (SDLC). In fact, 38% of executives cited difficulties integrating generative AI into existing workflows as their primary barrier to adoption.
A total of 39% also noted there is still a lack of AI expertise within their software development teams. In addition, other challenges such as security concerns (56%) and regulatory and compliance challenges (42%) remain. Nevertheless, nearly all respondents (93%) are planning to increase their investment in AI-augmented tools over the next two years, with 71% planning to incorporate AI into application development and SDLC management workflows.
Most of the benefits of AI have been seen by professional developers but in time AI should expand the ranks of citizen developers, said Coutinho.
Ai-augmented development efficiencies and challenges
Nearly half of the survey respondents (47%) said they expect a new type of application developer to emerge with specialized AI skills, such as prompt engineering, while 43% said they expect the responsibility of developers to expand.
In general, organizations should embrace generative AI with care. The recommendations surfaced are probabilistic, so DevOps teams need to understand that some of the answers being surfaced are very creative, noted Coutinho. Over time, however, as more domain-specific large language models (LLMs) are trained, the overall quality of the suggestions being made will become more accurate, he added.
The one certain thing is the pace at which software is being built and deployed is only going to increase in the age of generative AI. The quality of the code used to create that software in the short term may not substantially improve over what humans write today, but in time, the quality of applications should improve as, for example, fewer vulnerabilities will be created. In the meantime, organizations rather than simply diving into AI should put plans in place for operationalizing it thoughtfully, said Coutinho.
The winning combination going forward will always include some mix of humans and machines that ensure high-quality applications are deployed, he added. The challenge, of course, is that enthusiasm for AI far exceeds our collective ability to govern existing SDLC workflows. As such, AI despite best intentions might otherwise soon prove to be too much of a good thing.
