Coleman Parkes survey for Harness showed that use Artificial intelligence (AI) in the development of software It is on the rise, but there are gaps, which implies that more AI and automation can be expanded.
Survey results published in Harness's The state of artificial intelligence in the development of software 2025 Report, show that most software development groups believe that in the delivery of software, artificial intelligence agents that work together with human circuits for five years dominate.
Almost two -thirds of the respondents said that they use AI to generate code, 60% used AI for documentation, and 57% use AI to ensure quality and testing. Other areas of development of software supported by AI include error correction (55%), compliance with safety requirements (54%), as well as optimization of performance and costs (53%).
The areas in which the developers of the software developers observe improvements with the use of AI, include the speed of creating code (51%), faster testing and quality assurance (45%) and the adaptation time of the developer (43%).
The survey showed that on average organizations use eight -10 different AI tools to develop software. Some of them use a much larger number of tools, which suggests that there is a risk of AI tools, introducing the complexity that can lengthen the time necessary to completely attract new members of the software development on board.
Charus noted this The growth of the tool And Vibe coding can increase operational risk. He warned that fragmented tools and inexperienced developers using artificial intelligence assistants create management problems, increase incidents and bear hidden costs. He recommended that the IT-leaders combine tools into a unified platform and create fences with AI to reduce complexity and focus on innovation.
According to the study of the tourniquet, organizations have high use of AI for coding, but immature tests, deployment and management. Harness recommended that the IT losers have united ANS coding assistants with automated testing, checking deployment and security checks to prevent risks, expenses and manual labor.
The survey suggests that the maturity of automation is the main barrier, limiting the speed at which the software developers teams can deliver software. The largest gap in performance is not to create a code, but in delivery. According to a tour of the tourniquet, continuous delivery (CD) and management remain underestimated.
The acceleration of speed from coding AAA-SSSISTing creates a wave of pressure that collapses onto the wall of underestimated, obsolete lower processes. While the developers write the code faster than when the liba, the systems intended for testing, protecting and deploying this code are trying to keep up. Only 6% of the IT specialists surveyed said that the processes of the compact -disks of their organization are fully automated. This led to the appearance of what the tourniquet calls Ay paradox speedField
For organizations automated in organizations with automated less than a quarter, only 26% observed an increase in the frequency with which the code is sent to production using coding tools. This rises to 57% in those that are automated from one to three quarters of the CD processes. According to the tourniquet, moving from automation with low to moderate to compact -dishes, so more than doubles the probability that organizations will see an increase in speed from artificial intelligence coding tools.
Harness recommends investing in leaders in the descending automation to transfer the code speed controlled by the business speed.
