Industry Backs Open AI Standards as 95% of Pilot Projects Fail
Updated
Updated · O'Reilly Media · May 19
Industry Backs Open AI Standards as 95% of Pilot Projects Fail
2 articles · Updated · O'Reilly Media · May 19
Open standards such as Agent Skills, MCP and plugins are emerging as the common layer companies use to turn one-off AI wins into reusable internal catalogs.
Those catalogs package domain-specific workflows, tool connections and guidance so employees and coding agents can reuse hard-won knowledge instead of rebuilding it team by team.
95% of AI pilot projects are failing, the report says, pushing companies to avoid proprietary vendor lock-in and cut switching costs as coding tools shift from Copilot to Cursor to Claude Code and Codex.
Adoption is strongest for skills and MCP, with plugins also widely supported; less standardized artifacts such as CLI tools, hooks, rules and roots remain useful but more fragmented.
The broader bet is that storing these artifacts in shared repositories can turn individual productivity gains into durable organizational capability even as AI tooling keeps changing.
As AI automates coding, what new skills will define an engineer's value and prevent their obsolescence?
With 95% of AI pilots failing, are open standards a cure for low ROI or just more complexity?
From 95% Failure to Real ROI: How Open Standards and Policy Are Reshaping Enterprise AI in 2026
Overview
As AI agents become more sophisticated and widely used, the industry is uniting in early 2026 to establish open standards that ensure these systems are interoperable, secure, and trustworthy. This effort, led by NIST through its AI Agent Standards Initiative, aims to prevent market fragmentation and build a robust AI ecosystem. By releasing research and guidelines, NIST is guiding the development and adoption of advanced AI systems across industries. These open standards are crucial for supporting secure integration, fostering innovation, and ensuring that AI agents can work together effectively in a rapidly evolving landscape.