Workshop | Getting Started with AI Agents

The Agentic Shift Accelerates

AI agents represent a fundamental transition in artificial intelligence: from passive responders like chatbots to proactive, goal-oriented systems that plan, decide, and act across tools and environments. This shift crystallized in 2025 with key protocols—Anthropic's Model Context Protocol for tool integration and Google's Agent2Agent for inter-agent communication—donated to open standards, enabling scalable, interoperable agents. By February 2026, companies like Anthropic and OpenAI have released models with near-human capabilities in tasks such as navigating spreadsheets, coding, and coordinating parallel workflows.

The timing matters because 2026 marks the pivot to widespread adoption. Microsoft executives have dubbed it 'the year of the agent,' with enterprises moving from pilots to fleets of production-ready agents. Surveys show nearly 70% of executives anticipate transformative operational changes, while Gartner forecasts 40% of enterprise applications incorporating task-specific agents. Early impacts appear in software development, where up to 90% of code at firms like Anthropic is AI-generated, and Microsoft reports 30% AI-written code, with projections reaching 95% industry-wide by decade's end.

Real-world stakes center on productivity and economics. Agentic systems promise to handle complex workflows in customer support, operations, data analysis, and decision-making, acting as digital coworkers that amplify small teams. McKinsey values potential annual additions at $2.6–4.4 trillion across use cases, while KPMG quantifies $3 trillion in global gains, akin to a 5% EBITDA lift for Fortune 1000 companies. Businesses slow to integrate face rising relative costs, slower processes, and competitive erosion as rivals reduce cycle times, errors, and manual labor.

Non-obvious angles include emergent risks and tensions. Agents with broad system access introduce insider-threat vectors, amplifying vulnerabilities like prompt injection—deemed unlikely to be fully solved—or data exfiltration. Multi-agent orchestration offers parallel efficiency but demands new governance for accountability, as autonomous decisions blur lines between human intent and machine action. Stakeholders clash over pace: rapid deployment risks failures and reputational damage, yet hesitation cedes advantage in a race where model owners increasingly control new operational systems. Security teams must balance capability gains against threats, while platforms adapt interfaces for agent interaction without ceding control to unchecked autonomy.