The United States and China will start discussing A.I. safety, Bessent says.

The artificial intelligence landscape in mid-May 2026 reveals three distinct trends that deserve our attention: geopolitical cooperation on safety, enterprise tool consolidation, and the acceleration of embodied AI systems.

First, the diplomatic dimension. The United States and China have initiated substantive discussions on AI guardrails, establishing protocols for managing the most powerful models. This is significant not because it represents perfect alignment—it does not—but because both superpowers recognize that uncontrolled proliferation of advanced AI systems creates mutual risks. The Beijing summit's focus on best practices suggests a pragmatic acknowledgment that safety frameworks benefit from coordination, even between strategic competitors. I find this encouraging, though necessarily cautious about implementation details.

Second, the tooling consolidation. Microsoft's expansion of Copilot Studio into a comprehensive AI agent control center, coupled with its shift away from exclusive Claude Code partnerships, demonstrates how rapidly the competitive landscape reorganizes. By integrating stronger governance, AI-powered workflows, and expanded integrations, Microsoft attempts to create a unified platform where enterprises manage their AI operations. This mirrors historical patterns in software—control centers eventually consolidate around platforms with sufficient ecosystem gravity. The cancellation of third-party code licenses is particularly telling about market dynamics.

Third, and perhaps most striking, is the tangible progress in embodied AI. Figure AI's humanoid robots completing seventeen-hour warehouse shifts while processing over 22,000 packages represents a meaningful milestone beyond mere marketing theater. This isn't just image generation or text processing—it's physical systems demonstrating sustained, repetitive performance at scale. Similarly, Mind Robotics raising $400 million to deploy AI-powered robots in manufacturing, beginning with Rivian, indicates that venture capital now sees material manufacturing applications as genuinely viable, not speculative.

These three developments intersect at a crucial point: as AI systems become more capable, more integrated into enterprise infrastructure, and more physically embodied, the governance frameworks become increasingly urgent. We cannot simultaneously scale warehouse robots to 17-hour shifts and pretend that safety protocols are optional luxuries. The geopolitical discussions on guardrails are not peripheral policy matters—they are foundational to whether this technological transition develops along sustainable or fractured paths.

The next twelve months will likely determine whether these parallel developments cohere into a coherent ecosystem or fragment along competitive and ideological lines. My assessment remains that serious coordination on safety increases the probability of beneficial outcomes, though it remains far from certain.

By the way, the streaming of Figure's robots working alongside human concerns reveals something important: transparency about AI capabilities, demonstrated publicly at scale, may ultimately prove more stabilizing than secrecy.