With roughly 27 standardized events, a frozen 1.0 specification, and MCP and other projects moving under neutral stewardship at the Linux Foundation, AI coding competition is shifting from model benchmarks to the protocol layer. The change matters because it will also alter how enterprises evaluate and procure AI tools.
What this is
MCP, a common interface that lets Agents call external tools and data sources; AG-UI, an event protocol that streams Agent execution to front ends in real time; and Agent Skills, reusable capability packages that carry operating specifications together form the connectivity layer of AI coding. CopilotKit launched AG-UI in May 2025. Its ordered event stream represents text, tool calls, status changes, and human interruptions, sparing each front end from rewriting state mappings for every Agent.
This is not one protocol replacing all the others. A more complete stack combines MCP for connecting tools such as browsers and databases, A2A, the Agent-to-Agent protocol, for coordinating work among multiple Agents, AG-UI for displaying execution, and Agent Skills for preserving how a team works. The Linux Foundation has established the Agentic AI Foundation and placed Anthropic's MCP, OpenAI's AGENTS.md, and Block's goose under its stewardship, signaling that major participants are beginning to maintain standards as shared infrastructure.
Industry view
We believe the primary value of protocol standardization is lower switching costs. Developers no longer need to bind workflows to a single model or Agent framework. When models improve, tool connections and interface logic do not have to be rebuilt from scratch. Installable components such as Chrome DevTools MCP and stitch-skills are also turning specifications into ready-to-use engineering capabilities.
But neutral stewardship does not mean the ecosystem is already mature. Different protocols may still overlap in functionality, while SDK compatibility, security permissions, and data formats may continue to diverge. AG-UI has only just frozen its 1.0 specification, and neither Linux Foundation stewardship nor its membership list guarantees that vendors will implement the standards identically. More protocols could instead create new selection and adaptation costs for developers.
Impact on regular people
For enterprise IT: procurement will increasingly focus less on how intelligent a model is and more on whether it can connect to existing systems, whether its execution can be audited, and whether it can preserve organizational standards. Portability will gradually become an important metric for enterprise Agent projects.
For individual professionals: prompt engineering remains useful, but professionals should pay more attention to encoding experience as reusable rules and judging whether an AI's tool calls are reliable. Standardization will make expertise easier to share across teams, while reducing the value of repetitive operational work.
For the consumer market: once the protocols mature, AI applications will connect more easily to browsers, productivity software, and business systems. The scarce resource may then no longer be the chat interface, but whether users are willing to hand over critical permissions and data.