Organizations are building AI systems that combine models, agents and external tools instead of relying on standalone AI, according to Snyk’s latest State of Agentic AI Adoption report. The study analyzed 3,044 enterprise environments and 1.39 million code repositories to examine how enterprises are deploying AI.
Adopting agentic architectures
Of organizations using AI, 46.9% have adopted agentic architectures built on AI agents, model context protocol (MCP) servers, or both. More than half have deployed the full stack, combining AI agents with MCP infrastructure that enables access to enterprise data, applications, services and external tools.
Agentic adoption: agents vs. MCP servers (Source: Snyk)
These systems extend beyond chat interfaces by retrieving information, coordinating workflows and carrying out actions across enterprise environments.
Enterprise AI deployments include far more than models. When frameworks, MCP servers, retrieval systems, vector databases, datasets and supporting tools are included, the average AI footprint is about three times larger than model inventories indicate.
Nearly half of the companies analyzed had no declared AI models in their code repositories. They used AI through third-party services, packages and tools. At the same time, 17.2% operated large fleets of AI models, indicating AI is integrated across platforms and applications.
Organizations need visibility across this broader AI ecosystem because supporting components introduce additional dependencies, integration points and governance requirements.
The AI supply chain
OpenAI remained the most widely used model provider, although Anthropic and other vendors increased their share of enterprise deployments. The top four providers accounted for about 71% of identifiable model occurrences, showing a broader mix of AI vendors across enterprise environments.
Proprietary models represented 63.8% of deployed models, while open-source models accounted for 32.5%. Organizations use proprietary models for advanced reasoning and autonomous tasks. Open-source models are commonly deployed for embeddings, retrieval and other supporting workloads.
Third-party software plays a significant role in enterprise AI. External sources accounted for 77.4% of AI packages and tools, while 22.6% were developed internally. These dependencies shape how AI systems operate and introduce security, governance and software supply chain risks that organizations need to manage.
Capability, lineage and AI governance
Enterprise AI is becoming more capable, even though companies are not always deploying the latest models. They rely on established proprietary models for production workloads to balance performance with operational requirements. Open-source models are increasingly used for retrieval, embeddings and other specialized functions. The gap between proprietary and open-source models is narrowing, giving enterprises more options when building and operating AI applications.
About half of the organizations using AI models could not be linked to the datasets used to train or fine-tune them. This makes it harder to understand how models generate results, investigate incidents and demonstrate compliance.
AI is becoming deeply embedded in software development. Developers are working with more AI models, agents and supporting tools, increasing the amount of autonomous technology operating across enterprise environments. Enterprises need visibility into what AI systems can do, what data they use, what resources they can access and how they behave in production.
AI adoption across industries and regions
AI adoption varies widely across industries. Media and entertainment companies had the highest concentration of AI components per organization, followed by retail, consumer goods and education. Technology and IT companies deployed the largest overall volume of AI, while other sectors integrated AI extensively across individual businesses.
Industries where AI supports content creation, customer experiences and business processes recorded the highest levels of adoption. These sectors are deploying not only AI models but also agents, orchestration frameworks and supporting infrastructure that enable AI to perform tasks across multiple systems.
Businesses in North America and Europe are building similar AI architectures, with both regions increasing adoption of AI agents and MCP infrastructure. North American organizations generally deploy AI at a larger scale, while the underlying technologies and architectural patterns remain consistent across both regions.
