代理式AI的轉變：2025年代理式AI的真正突破

OneMCP

The Agentic Shift: 2025’s Real Breakout for Agentic AI

Table of Contents

Every year, AI has a defining narrative. For a long time, that narrative was intelligence—bigger models, better reasoning, and more fluent outputs. We spent years marveling at what AI could "think."

But in 2025, that focus shifted. The conversation moved from what AI could think to what AI could actually do. We are no longer debating whether an LLM can pass the bar exam. We are watching AI agents manage global supply chains, debug entire codebases, and negotiate enterprise contracts via specialized MCP servers.

As the tech industry rushes to "agentize" everything, a clear divide has emerged. Some are merely chasing the latest trend, while others are building the autonomous infrastructure of the next decade.

Let’s reflect on the key developments in 2025 that helped turn the "Generative AI" hype into an "Agentic AI" reality.

Major Launches and Advancements

The Rise of "Computer Use" and Browser Agents

Industry Standardization: The Agentic AI Foundation (AAIF)

Enterprise & Device-Native Advancements

Key Investments and Acquisitions

Investment in agentic AI reached a fever pitch in 2025, as capital shifted from "chatbots" to "agents" capable of autonomous action.

Explosion in Enterprise GenAI Spending

Global enterprise spending on generative AI applications surged to $37 billion in 2025, a 3.2x increase from 2024. This surge is part of a broader trajectory toward a $51.5 billion market by 2028, with 65% of enterprises now moving beyond pilots to full-scale adoption of agentic coding and workflow tools.

Major VC Funding Surge

AI startups captured over 50% of all global venture capital in 2025. Leading the charge in agentic ventures, WorkFusion raised $45 million to scale its AI agents for financial crime compliance, while Console secured $23 million to automate Slack-based IT support. Seed-stage investment in autonomous agents alone topped $700 million,  highlighting U.S. dominance.

Workday's Acquisition of Sana

On November 4, 2025, Workday completed its acquisition of Sana, an AI-native knowledge platform, for $1.1 billion. The deal aims to turn Workday into the "front door for work" by embedding agents that proactively manage HR and finance tasks.

Big Tech Infrastructure & M&A

Crypto-AI Intersections

Coinbase bridged the gap between AI and finance by launching the x402 protocol and AgentKit. Combined with tokenization platforms like Virtuals Protocol, the "Agentic Web" is now positioned to capture a projected $52.6 billion market by 2030.

Real-World Applications and Examples

Healthcare Agents

Non-diagnostic agents handled patient intake and reminders, while AI models like ECgMLP achieved up to 99.26% accuracy in detecting specific cancers such as endometrial pathology. In life sciences, LG AI Research’s EXAONE Path 2.0 accelerated drug discovery by integrating histopathological and genomic data.

Finance and Insurance

ERP-embedded agents enabled touchless operations, and multi-agent NLP systems reduced manual labor in claims processing by 80%. While industry averages vary, specialized AI trading agents utilized Financial Learning Models (FLMs) to deliver risk-adjusted returns exceeding 200% in volatile markets.

Software and IT

Autonomous coding agents now handle most routine bug fixes. In infrastructure, proactive IT agents boosted resiliency, while supply chain agents enabled self-healing through real-time modeling of global logistics data.

Marketing and E-Commerce

At Amazon Accelerate 2025, the company unveiled an Agentic Seller Assistant that proactively manages inventory, pricing, and compliance. Similarly, Box introduced Box Automate, enabling agentic workflows to extract and orchestrate data across enterprise content.

Productivity Impacts

According to LSE research, AI agents save professionals an average of 7.5 hours per week. A 2025 PwC survey revealed that 88% of executives are increasing AI budgets, with 77% of leaders redesigning their entire data strategies to support autonomous agent integration.

Regulatory and Legal Developments

As adoption grew, so did scrutiny, with new frameworks addressing the unique risks of autonomous agents.

EU AI Act Implementation

The European Commission launched the AI Act Service Desk and a Single Information Platform in late 2025 to support compliance. These tools provide a Compliance Checker specifically for providers of high-risk systems and General-Purpose AI (GPAI) models, ensuring traceability and human oversight as the Act’s core obligations become applicable.

U.S. Legislation and Trends

While state-level AI bills proliferated, the federal focus shifted toward a "minimally burdensome" national framework via a December 2025 Executive Order. In a landmark case for "Agentic Commerce." Amazon sued Perplexity AI over its Comet browser, alleging that the autonomous purchasing agent bypassed security measures and violated terms of service by masking its automated identity.

Global Risk Frameworks

Regulators in the UK and France advanced risk-based oversight through initiatives like the Digital Regulation Cooperation Forum (DRCF). These frameworks emphasize human-in-the-loop requirements and "self-healing" security protocols to mitigate bias and prevent agents from executing unauthorized financial transactions.

Trends and Future Outlook

The close of 2025 marked the emergence of Multi-modal Integration and the "AgenticOS" concept, a foundational layer where AI agents interact seamlessly with blockchain protocols to power autonomous, tokenized micro-economies.

The Human-Agent Workforce

While 2024 was defined by experimentation, McKinsey’s late-2025 research highlights a shift toward "skill partnerships." Early-year data showed agentic-related job postings surged by 985%, supported by $1.1 billion in initial equity investments. By Q4 2025, the focus evolved into the "Virtual Coworker" model, with 23% of enterprises now scaling agentic systems to handle complex, multi-step workflows.

Sector-Specific Momentum

Industry-specific growth was solidified at major 2025 summits. At the Singapore FinTech Festival, Dyna.Ai demonstrated agentic platforms for the 85% of financial institutions now deploying AI. Simultaneously, specialized forums like Agentic AI Davos and the Generative AI in Insurance USA conference provided the blueprints for autonomous underwriting and claims management.

Long-term Forecast

In summary, 2025 was the year agentic AI transitioned from a "hype cycle" into a structural reality. According to Precedence Research, the global agentic AI market is on a trajectory to reach $199.05 billion by 2034, expanding at a robust 43.84% CAGR. As IBM and Microsoft emphasize, the future lies in balancing rapid innovation with "accountability by design" within these autonomous systems.

Why 2025 Became the Breakout Year for Agentic AI

2025 will go down in history as the year that AI capabilities were wired into durable systems. Standards like AAIF and MCP, big infrastructure bets from hyperscalers, serious M&A, and early but real regulation all pulled in the same direction. Together they turned “agents” from demo material into a structural part of how work gets done.

That is the real agentic shift. The question is no longer “Can an agent do this task?” but “Under what conditions should it be allowed to… and how do we keep it within those lines?” If we use that lens as the default and build with auditability, shared standards, and clear limits, the next wave of agentic AI will feel less like a stunt and more like a new layer of the economy.

‍

Team Gentoro

Further Reading

Turn Your OpenAPI Specs Into MCP Tools—Instantly

Customized Plans for Real Enterprise Needs

Gentoro makes it easier to operationalize AI across your enterprise. Get in touch to explore deployment options, scale requirements, and the right pricing model for your team.

Cookie Settings

We use cookies to provide you with the best possible experience. They also allow us to analyze user behavior in order to constantly improve the website for you.