Expect AI to Spark a New Way of Thinking About System Design

Artificial intelligence has significantly influenced how organizations approach IT modernization, while AI itself is rapidly evolving. Stuart Brown, Partner and Technology Leader at Guidehouse, highlighted during Federal News Network’s AI and Data Exchange that 2025 will see substantial improvements in efficiency, reduced hallucinations, and increased accuracy for large language models. He introduced the concept of multithreaded thinking, which differs from standard queries and conventional prompts by employing a reasoning model that searches across domains to find answers.

To enable AI models to engage in multithreaded thinking, organizations must expose them to a wide range of data, necessitating a system architecture that ensures data availability. Brown emphasized that much of the data within organizations is underutilized unless its location is known. By designing architectures that expose this data, AI models can perform more like human reasoning.

Guidehouse has explored advancements such as DeepSeek, OpenAI Deep Research, and Grok, which demonstrate deep reasoning and the ability to ask clarifying questions. Brown predicted that U.S. LLM providers will focus on developing similar capabilities in 2025, impacting customer service and various other areas.

Brown advised IT staffs to carefully consider their architectures for wider data exposure, addressing challenges such as hallucinations, security, user permissions, and bias. A responsible AI architecture will bring predictability, ensuring that the model’s answers are trustworthy and within a range of trust for users to make decisions.

The shift in architecture involves recognizing data as the core of the organization, rather than an output of systems. This requires ensuring data accuracy, reliability, security, and proper tagging. Brown stressed the importance of rethinking best practices, policies, and processes to avoid compromising organizational competencies or violating regulatory requirements.

Finally, Brown predicted that multithreaded thinking will enable more natural interactions with AI models, reducing the need for elaborate prompt engineering training. AI models will become more proactive in offering decision options, enhancing human productivity through intelligent information and options provided via emails, pings, and meeting invites.