The Serbian IT Association (SITA)<\/a> AI Working Group<\/a> recently gathered at the Endava office for an open, collaborative exchange focused on Artificial Intelligence in the Software Delivery Lifecycle (SDLC)<\/span><\/span>. <\/span>Led by Borivoje Kova\u010d<\/a>, Regional AI Evangelist from Endava<\/a> and coordinator of the working group, the session brought peer companies together to move past marketing hype and discuss how AI is actually touching day-to-day engineering practices, team dynamics, and human roles<\/span><\/span>.<\/span><\/p>\n Because the session was a closed, trusted forum meant for honest benchmarking, specific internal details remain private<\/span><\/span>. <\/span>However, the high-level insights guided by Borivoje during the micro-panel and round table discussions highlight critical shifts relevant to our entire community<\/span><\/span>.<\/span><\/p>\n A central theme from the panel was separating vendor promises from engineering reality<\/span><\/span>. <\/span>Despite intense hype around fully autonomous AI agents, the room agreed they are not ready to replace human judgment<\/span><\/span>. <\/span>Instead, AI works best as a disciplined tool under strict human control, where developers act as the essential architects who define intent, set system structure, and carefully review generated code<\/span><\/span>.<\/span><\/p>\n Furthermore, AI output quality scales directly with the precision of human guidance<\/span><\/span>. <\/span>Vague instructions introduce unnecessary risk, making crisp specifications and accountable human code reviews completely non-negotiable to prevent downstream bugs<\/span><\/span>.<\/span><\/p>\n The round table discussions offered a grounded perspective on the changing workforce<\/span><\/span>. <\/span>Rather than eliminating jobs, AI is viewed as part of software history’s long trend toward higher levels of abstraction<\/span><\/span>. <\/span>The modern engineer’s focus is shifting upward, from manual syntax to system design, orchestration, and business logic validation<\/span><\/span>.<\/span><\/p>\n Looking ahead, the group anticipated a shift toward standardized, team-wide AI workflows and smaller, highly adaptive team structures<\/span><\/span>. <\/span>However, this brings new challenges, such as managing knowledge concentration risks and protecting mentorship paths for junior developers<\/span><\/span>.<\/span><\/p>\n The session closed with a clear call to action<\/span><\/span>. <\/span>The tech sector currently lacks a mature framework to measure AI’s real value, and simplistic metrics like lines of code or raw speed were completely rejected<\/span><\/span>. <\/span>Moving forward, SITA aims to define a practical evaluation scorecard that accurately balances delivery speed against code quality, long-term costs, and human learning<\/span><\/span>.<\/span><\/p>\nEngineering Reality: An Assistant, Not a Replacement<\/h3>\n
Shifting Abstractions and Team Dynamics<\/h3>\n
Defining the Right Yardstick<\/h3>\n