October 2026 marks a definitive inflection point for agent infrastructure. While industry standards bodies have spent months debating the architecture of trust, the market has bypassed them entirely. Three distinct developments – the release of PACT, the emergence of Salesforce’s A2A-compliant labeling, and the expansion of the Linux Foundation’s Agentic AI Identity Framework (AAIF) – demonstrate that the industry is no longer waiting for top-down ratification. Instead, trust infrastructure is being built bottom-up through shipping code, creating a functional, if fragmented, consensus.
The most significant shift is the arrival of PACT, an open-source mechanism for signed requests and direct permission granting. Co-developed by Decagon and Instinct and released under an Apache-2.0 license, PACT provides a complete consent-layer stack. It includes a normative specification, a reference provider implementation, and a suite of conformance tests. By building on A2A 1.0 and OAuth, PACT allows personal agents to identify themselves via signed requests while enabling customers to grant permissions directly to businesses without exposing their login credentials. It is a pragmatic, working solution that functions independently of a central registry.
This bottom-up momentum stands in stark contrast to the stalled efforts of the A2A project. The formal A2A Registry specification, which was the subject of intense debate from June 2025 to August 2026, never materialized. Discussion #741, which spanned 102 comments and 79 replies, was officially closed on August 20, 2026. The maintainers redirected the focus toward AI Catalog and ARD specifications, effectively abandoning the registry model that many had hoped would serve as the industry’s backbone. By the time the discussion was shuttered, five incompatible registry implementations were already in the wild, illustrating the futility of waiting for a single, ratified standard.
Enterprise incumbents are filling this vacuum by co-opting the language of standards to drive their own ecosystems. In the Salesforce Winter ’27 release, the company introduced an A2A-compliant label for its Third-Party Agent Orchestration. This label allows Agentforce to orchestrate agents across AWS, Azure, and Google, yet it operates without a formal, industry-wide conformance specification. For enterprise architects, this label functions as a marketing signal of ecosystem participation rather than a guarantee of technical interoperability, highlighting how vendors are defining compliance on their own terms.
Simultaneously, the Linux Foundation is positioning identity-first frameworks as a foundational pillar for the agent economy. The AAIF project has gained significant traction, adding 43 new members in May 2026, including F5, GoDaddy, Stripe, and TRON. By focusing on portable identity, dynamic trust, and cross-domain delegation, AAIF is building a framework that operates independently of the stalled A2A registry efforts. As GoDaddy noted, the goal is to extend open standards for verifiable agent identity, providing a structural alternative to the fragmented registry landscape.
This structural pattern reveals a clear reality: trust infrastructure is being built by those who ship, not those who write specs. For builders, the implication is to prioritize integration with shipping implementations like PACT and AAIF rather than waiting for formal standards that may never arrive. However, this approach carries significant risks. The lack of formal conformance specifications creates high technical debt and potential security vulnerabilities in cross-ecosystem interactions. Furthermore, the proliferation of competing bottom-up standards risks long-term fragmentation, while reliance on vendor-led implementations introduces the danger of vendor lock-in under the guise of open standards.
The industry is moving toward a reality where different agent ecosystems operate on internally consistent but externally incompatible permission models. This trajectory is evident in the draft specification landscape, the ongoing bifurcation of tool discovery, and the Gemini governance model. The absence of ratified standards does not mean the industry is stalled; it means the industry has decided that utility is more valuable than consensus. The next phase of agent infrastructure will be defined by these competing, shipping implementations, leaving the dream of a universal, top-down standard behind.