AI Applied Wrapped: Innovation, Collaboration And Constraints In Enterprise AI In 2025

Looking back on 2025, the most striking feature of enterprise AI was not any single release, but the speed at which the landscape shifted. In the space of 12 months, the model market moved from Gemini 1.5 to Gemini 3, and from GPT-4o through GPT-4.1, o3 and ultimately to GPT-5. Throughout the year, Verdantix tracked the rapid evolution of the AI market – below, we surface some of our main takeaways heading into 2026.

Model development itself remained intense. Early 2025 was marked by heavy emphasis on scale, where parameter counts and training hours were openly marketed. Over the year, the limits of this approach became clearer. Larger models continued to deliver gains, but with diminishing returns and persistent constraints around cost and auditability. This drove a shift toward smaller, more flexible deployments and greater attention to behaviours that matter for downstream use, particularly in agent contexts. This included OpenAI’s o3 and GPT-5, which established reasoning and multimodality as core capabilities, while Google advanced Gemini through successive releases, leveraging its integrated control of Google Cloud and TPU hardware. Models remained central but increasingly served as components rather than standalone products.

This reframing was closely tied to the rise of agents as a dominant theme. As agents moved from concept to early deployment, interoperability and coordination became more important than raw model benchmarks. Anthropic’s Model Context Protocol (MCP) gained traction in early 2025; Google introduced its Agent2Agent (A2A) protocol to support discovery and cooperation across external frameworks; and IBM’s Agent Communication Protocol (ACP) was later aligned with A2A under the Linux Foundation umbrella, consolidating fragmented standards. Parallel efforts, such as the Agent Network Protocol (ANP), explored decentralized interaction models. The donation of MCP to the Agentic AI Foundation, a fund under the Linux Foundation, in December reinforced the direction of travel. Agent systems are still constrained, particularly by memory and semantic consistency, but the foundations for multi-vendor ecosystems advanced materially.

2025 also marked a recalibration of enterprise expectations. Many early pilots struggled to demonstrate value, and even successful deployments rarely scaled cleanly. The 2025 Verdantix AI survey highlighted the key constraints as data readiness, governance gaps and limited in-house AI capability. However, enterprise AI platforms gained traction by anchoring AI use cases in existing systems of record or curated data layers, allowing utility agents to be deployed relatively easily for applications such as knowledge retrieval and task assistance. Vendors of varying pedigrees emerged as credible enterprise AI platforms – including the evolution of some CRM and ERP providers into these platforms – giving buyers a wide range of options. The promise of comprehensive workflow automation remains largely unrealised for most firms, but this pragmatic deployment model, which enables effective AI use without first completing large-scale data transformations, represented an important step forward.

Geopolitics continued to shape the market. Initial provisions of the EU AI Act came into force mid-year before political reassessment slowed momentum. In the US, a more permissive regulatory stance contrasted with Europe’s emphasis on sovereignty. China advanced a largely self-contained stack spanning open-source models such as DeepSeek-R1 and Alibaba Cloud’s Qwen3, alongside domestic hardware initiatives like Huawei’s Ascend chips. Innovation remained high priority, but ecosystems became more nationally defined.

Overall, 2025 was characterized by rapid change, intense vendor activity and a clear shift from model spectacle to execution, where agents, platforms and enterprise integration increasingly mattered more than incremental gains in model scale. To understand how these dynamics are likely to evolve in 2026, join the Verdantix 2026 AI predictions webinar and explore the AI Applied insights page for deeper, ongoing analysis of developments across the AI market.