In collaboration withSoftServe
During this century, the field of software engineering has undergone two substantial transformations. The initial shift was the emergence of the open source movement, which progressively made code available to developers and engineers across the globe. Following that, the introduction of development operations (DevOps) and agile practices transitioned software development from isolated efforts to collaborative ones, evolving from batch processing to continuous delivery. Presently, a third shift appears to be emerging with the integration of agentic AI in software engineering.
So far, engineering teams have primarily utilized AI to aid in coding, testing, and other specific tasks, within well-defined boundaries. However, with agentic capabilities, AI agents evolve into reasoning, self-managing entities that can oversee entire software projects—not just individual tasks—and often do so autonomously. Should engineering teams adopt and fully leverage agentic AI, it is set to revolutionize software process automation from start to finish, leading to agent-managed development and product lifecycle automation.
This report, based on a survey of 300 engineering and tech executives, indicates that software engineering teams are recognizing the potential of agentic AI and are starting to utilize it, albeit in a largely restricted manner. Their aspirations are high, but many acknowledge that significant time and effort will be necessary to mitigate the hurdles to its widespread implementation in software operations. Similar to the transitions seen with DevOps and agile, fully capitalizing on the advantages of agentic AI in engineering will necessitate sometimes challenging organizational and process adjustments alongside the technological adoption. Nevertheless, the potential enhancements in speed, efficiency, and quality suggest that any associated challenges will be worth the effort.
Key insights include:
Adoption momentum is increasing. While half of organizations consider agentic AI a priority investment for software engineering today, it is projected to be a primary focus for more than four-fifths in two years. This investment is propelling accelerated adoption. Currently, 51% of software teams are using agentic AI (albeit in limited capacities), and 45% plan to adopt it within the next year.
Initial gains will be gradual. It will require time for software teams’ investments in agentic AI to yield significant results. Over the next couple of years, most anticipate that the enhancements from agent use will be minimal (14%) or at most moderate (52%). However, about one-third (32%) have higher expectations, with 9% anticipating transformative improvements.
Agents will speed up time-to-market. The primary benefits derived from agentic AI utilization over the next two years will center around increased speed. Nearly all respondents (98%) expect their teams to deliver software projects from pilot stages to production more rapidly, with an average expected increase in speed of 37% across the board.
The objective for many is comprehensive agentic lifecycle management. Teams are highly ambitious regarding the scaling of agentic AI. Most aim for AI agents to oversee the entire product development and software development lifecycles (PDLC and SDLC) swiftly. At 41% of organizations, teams aspire to achieve this for the majority or all products within 18 months. That number is predicted to rise to 72% in two years if expectations are realized.
Compute costs and integration represent significant early hurdles. For all survey participants—but particularly in early-adopter sectors like media and entertainment, as well as technology hardware—integrating agents with existing applications and the expenses associated with computing resources are the foremost challenges encountered with agentic AI in software engineering. Meanwhile, the experts we consulted highlight the substantial change management challenges teams will face in altering workflows.
This content was produced by Insights, the custom content division of MIT Technology Review. It was not authored by MIT Technology Review’s editorial team. It was researched, crafted, and composed by human writers, editors, analysts, and illustrators. This includes the formulation of surveys and the collection of data for those surveys. AI tools that may have been employed were restricted to secondary production processes that underwent thorough human scrutiny.
