AI
AI Outpaces Organizations: Study Reveals Tool Integration, Not Capability, as Key Bottleneck in Requirements Engineering
A new study of real-world AI adoption in software requirements engineering finds that tool integration challenges and organizational lag are slowing benefits, even as AI itself moves faster than teams can absorb.
A detailed examination of generative AI adoption at a mid-sized software engineering firm reveals a widening gap between what AI tools can do and what organizations are equipped to leverage from them. Rather than capability limits, the real constraints are integration challenges and misalignment between rapid technology deployment and slower organizational readiness.
How the Study Unfolded#
Researchers conducted a comprehensive investigation at XITASO, a medium-sized enterprise specializing in high-tech software development. The work combined a company-wide survey of use cases in 2024 with two rounds of semi-structured interviews involving eight product owners (POs) across late 2025 and spring 2026. The inquiry covered both an in-house AI chatbot and seven commercial AI tools in active use.
What AI is Being Used For#
The team catalogued 15 distinct use cases for generative AI, organized into four functional areas:
- Product backlog management — organizing and prioritizing work
- Tender management — responding to client proposals and bids
- Requirements and domain understanding — clarifying technical specifications and business context
- Document and artifact creation — generating specifications, diagrams, and other deliverables
Three Core Findings#
The research surfaced three unexpected patterns that contradict common assumptions about AI in professional practice.
First, AI changes—but doesn't always improve—team dynamics. The dominant way product owners use AI today is a single-user interaction: the PO chats with a tool rather than collaborating directly with developers. This can replace collaborative dialogue rather than enhance it, and developers frequently resist accepting AI-generated artifacts without substantial rework. The social and collaborative dimensions of requirements engineering, in other words, are not automatically solved by faster tooling.
Second, integration is the real constraint. Where AI tools connect seamlessly with existing systems, time savings are dramatic. Where that integration is absent, product owners revert to manual workarounds—copying and pasting between systems, re-entering data, losing efficiency gains. The sophistication of the AI model matters far less than whether it fits into the workflow.
