Adoption with caution: AI’s enterprise journey thus far

Several institutions have moved decisively beyond pilots, embedding generative artificial intelligence into development workflows, client-facing tools and data programs. But the industry remains split between enthusiasm for rapid experimentation and caution about governance, data foundations and long-term architectural readiness.

Key takeaways: 

• AI adoption has moved from experimentation to enterprise integration, with coding assistants, platform modernization tools and internal LLMs now embedded in daily workflows.
• Agentic AI is emerging cautiously, with the consensus being that fully autonomous behavior is “nowhere near production-ready” due to risk, governance and security concerns.
• Data quality and architecture remain the biggest obstacles to scaling AI, with leaders warning that poor lineage, inconsistent definitions and domain silos “will not be fixed by AI”.
• Banks continue to struggle with describing and mapping their data, slowing the development of semantic layers, knowledge graphs and AI-driven analytics.
• AI governance and risk controls are lagging, especially around prompt injection, data poisoning and internal-agent access privileges.
• Modernization efforts are accelerating through AI-assisted refactoring, but remain being seen as a “tax” compared with strategic initiatives.
• Cloud strategy and regulatory expectations remain inconsistent, limiting the ability to deploy complete AI and data stacks across public cloud environments.