As enterprise agent adoption scales, infrastructure is formalizing into a modular four-pillar architecture that’s replacing monolithic platforms with specialized, best-of-breed components.
SEP-2133 introduces reverse-DNS identifiers, official and experimental extension tracks, and capability negotiation, transforming the agent protocol into a modular platform with a governed extensibility layer.
Version 2.4.0 introduces Workflow as Tool, ManagedAgent, mTLS, and Daytona integration — four layers that make ADK the deepest vertical integration play in the agent SDK space, and the deepest lock-in risk.
An IETF draft from Cisco Systems defines a single ‘_agent._tcp’ service type over mDNS/DNS-SD that lets agents find each other on local networks with zero configuration. A2A is the first protocol instantiation.
With RemoteA2aAgent and A2A Protocol v1.0 now backed by 150+ organizations, agents written in different languages can discover, coordinate, and fail over without sharing a framework.
Yellow Network proposes a Layer 3 state channel architecture to resolve the blockchain trilemma, enabling peer-to-peer trading without cross-chain bridges through a collateralized security model.
The company that generates 95%+ of its revenue renting GPUs to Google and Anthropic just became their model competitor. No safety documentation included.
Information agents that reason 24/7 without prompts are now live for AI Ultra subscribers. With 1 billion monthly Search users as distribution, Google has moved the agent war from model benchmarks to background processes.
The safety-focused lab currently runs on Google TPUs. Early talks with Samsung Foundry signal a push toward infrastructure independence — and into a foundry pipeline that already includes Tesla, Meta, and AMD.
With the public release of Terra and Luna, the AI market enters a new era where the most powerful capabilities remain permanently government-gated.