Day two of TechEx North America offered a deeper, critical examination of artificial intelligence in the enterprise, tempered with an undercurrent of cautious optimism. The AI and Big Data program opened with a pointed reference to what speakers termed the “AI graveyard” – projects that show promise in controlled pilot environments but fail to deliver when deployed at scale in real-world operations. Despite the grim metaphor, multiple sessions and keynote addresses provided actionable roadmaps for organizations aiming to avoid that fate.
The day’s programming across the Enterprise AI Implementation, ROI, and Adoption tracks took stalled pilots as a starting point, analyzing why many AI initiatives falter before reaching meaningful deployment. Speakers emphasized that effective agentic AI should be focused on specific business areas rather than deployed broadly without clear objectives. Building agent-ready data foundations – essentially planning for success at the infrastructure level – emerged as a recurring theme, as did the financial realities of token-based AI charging models and their impact on corporate budgets.
Infrastructure Decisions and the Scaling Challenge
At the infrastructure level, panels debated whether companies should buy or build physical infrastructure to support their AI projects. The core tension, many argued, is epitomized by what one speaker called the “personal copilot” problem. A single worker, especially a C-suite executive, can achieve impressive efficiency gains from a customized AI assistant. Those individual successes generate excitement within the organization. However, scaling that success from one desk to an entire department, let alone a whole enterprise, remains the primary roadblock for most companies. The gap between promising one-user pilots and enterprise-wide transformation framed much of the day’s discussion on the show floor and stages at the San Jose McEnery Convention Center.
Security Concerns and the Velocity Gap
On the Cyber Security and Cloud Expo stage, speakers identified a “velocity gap” caused by the rapid adoption of agentic AI systems across business units. When AI deployments succeed, they gain traction quickly. But security and governance issues emerge when business units adopt generative AI tools faster than the security team can govern them. This creates what experts described as a double-edged sword: AI can improve both attack and defense in cybersecurity, but it also expands the attack surface through unbounded agents, large language models, and attackers’ use of AI scanning tools to identify potential exploits.
The familiar problem of shadow IT has resurfaced in a new form: shadow AI. Staff may place sensitive material into unsanctioned tools, or approved AI systems may be poorly bounded and managed. In either case, the attack surface expands without the cybersecurity team’s awareness. Data governance and system oversight are becoming more deeply intertwined as a result, a message repeated across both cybersecurity and cloud tracks throughout the day.
Zero Trust as a Framework for AI Adoption
For pure-play cybersecurity functions, zero trust architecture was presented as one answer to runaway AI adoption outside formal oversight. The principle of “denial by default” applies to humans and machines alike. Under this model, proof of identity and privilege levels must also apply to services and agents. Automated workflows would then be subject to the same permission models as every other element in the IT stack, providing a governance layer that can keep pace with rapid generative AI deployment.
The Rise of Physical AI and Humanoid Robots
Despite the cautionary notes, day two was not a rejection of enterprise AI ambitions. Speakers, thought leaders, and delegates accepted the role of AI and autonomous agents as established facts. Each industry representative placed concerns and enthusiasms on the table, contributing to discussions around AI implementation roadmaps heading into 2026.
The conference floor also buzzed with excitement around physical AI. Humanoid robots on display drew large crowds – everyone seems to enjoy a friendly android – but the more pragmatic draw was the new Physical AI track, which attracted some of the largest audiences of the event. Delegates showed strong interest in how embodied AI systems might move beyond research labs into practical industrial and commercial applications.
As TechEx North America continues, observers expect further clarity on how enterprises can bridge the gap between promising AI pilots and sustainable, scalable deployments. Discussions around infrastructure choices, governance frameworks, and the integration of physical AI will likely shape enterprise strategies for the remainder of the year and into 2026.
