By Anthropic opened Agent Skills as a public standard on December 18, 2025, becoming the first major AI company to open-source its agent framework. The move comes nine days after donating the Model Context Protocol to the Linux Foundation’s Agentic AI Foundation, establishing a pattern: Anthropic is positioning itself as the “open alternative” to OpenAI’s closed ecosystem. The strategic bet is that developers and enterprises will choose openness over vendor lock-in—and history suggests they’re right.
Skills created in Claude can now work in ChatGPT, Cursor, or any platform adopting the standard. Anthropic published the specification and SDK at agentskills.io for any AI platform to use. It’s a direct challenge to OpenAI’s proprietary approach, mirroring the Android vs iOS playbook that reshaped mobile computing.
The Android Parallel
When Google opened Android in 2008, Steve Jobs declared he would go “thermonuclear war” on what he called a “stolen product.” Apple’s closed ecosystem seemed destined to win. But open won market share: Android captured 76% of the global market while iOS maintained 23%. Open platforms trade premium positioning for volume, and that volume compounds into ecosystem dominance.
Anthropic is running the same playbook. By opening Agent Skills, they’re betting developers and enterprises will choose freedom over lock-in, just as they did with Android, Linux, and web standards.
Market Share Reversal
Anthropic’s enterprise LLM market share jumped from 12% in 2023 to 32% in 2025. OpenAI went the opposite direction: from 50% to 25% in the same period. That’s a complete reversal in two years. Enterprises prefer Anthropic for safety (99.4% prevention rate with safeguards versus OpenAI’s ~20% vulnerability to injection attacks), trust (153-page system card versus 60-page), and most importantly, no vendor lock-in.
When enterprises evaluate AI platforms, they ask: how difficult would it be to exit this solution? If the answer is “very difficult,” they lean toward open options. OpenAI’s closed ecosystem creates dependency. Anthropic’s open standards provide an exit ramp.
What Agent Skills Are
Agent Skills are organized folders of instructions, scripts, and resources that AI agents discover and load dynamically. At their core, they’re markdown files with progressive disclosure—only loading what’s needed. No prompt engineering required for repeated tasks. Enterprises use them for coding workflows, legal processing, finance automation, and data science pipelines.
Enterprise management lets admins provision skills centrally while letting employees customize their experience. Partner integrations include Atlassian, Figma, Stripe, Notion, and Zapier. One Hacker News developer described it as clicking immediately: “take a general-purpose agent and turn it into a specialized one with procedural knowledge that no model can fully memorize.”
The Standards War Context
December 2025 has been busy for AI standards. On December 9, the Linux Foundation formed the Agentic AI Foundation, co-founded by Anthropic, OpenAI, and Block, with support from Google, Microsoft, AWS, and Bloomberg. Anthropic donated the Model Context Protocol, which has over 10,000 active servers and 97 million monthly SDK downloads. Nine days later, Anthropic opened Agent Skills. The pattern is clear: building infrastructure for the agentic economy, not just products for their own ecosystem.
Gartner predicts 40% of enterprise apps will feature AI agents by 2026, up from less than 5% in 2025. The AI infrastructure market is projected to hit $758 billion by 2029. Who controls the standards will control the ecosystem.
Why Vendor Lock-In Matters
Enterprises worry about proprietary AI platforms for predictable reasons. Pricing volatility: OpenAI, Google, and Anthropic have all revised pricing multiple times. Usage restrictions: OpenAI doesn’t allow full fine-tuning of GPT-4 Turbo. Ecosystem dependency: switching providers is costly and disruptive.
Open standards solve this. Hybrid approaches are emerging—Salesforce lets users switch between proprietary and open models, while Oracle and SAP now support LLaMA models. The future is multi-platform, and enterprises want freedom to move between them.
Open Wins by Influence
OpenAI has quietly adopted structurally identical architecture to Agent Skills in both ChatGPT and its Codex CLI tool, using the same file naming conventions, metadata format, and directory organization. Microsoft is building Skills into VS Code. Enterprise software giants are racing to support the standard. Anthropic is winning not just by adoption, but by influence—even competitors are building on their specifications.
Platform wars aren’t won by the best technology. They’re won by strategy. Android didn’t beat iOS on features; it won on openness. Linux didn’t beat Windows on usability; it won on freedom. Web standards didn’t beat proprietary alternatives on polish; they won on interoperability. Anthropic is betting on developer values—freedom, no lock-in, open standards—and the market is responding.
The enterprise AI agent war is just beginning, but the playbook is centuries old: open standards win by ecosystem, not by control.
