AWS re:Invent 2025 - A leader's guide to agentic AI (SNR201)

Leading in the Era of Agentic AI

The Shift to High Agency

• Agentic AI systems are goal-driven, resourceful, and able to learn and adapt, unlike traditional automation
• The capability of AI to handle more complex tasks is doubling every 7 months, while the cost to access this intelligence is continuously dropping
• This creates a "sweet spot" where highly capable and affordable agentic AI systems are becoming viable for real-world applications

Key Leadership Mental Model Shifts

Governance

• Move from gate-based governance to a policy engine that provides strategic direction and guardrails, similar to how a board of directors works with a CEO

Risk Management

• Shift from fixed risk thresholds to a "trading floor" model with real-time visibility, control, and circuit breakers

Organizational Structure

• Transition from vertically optimized silos to a more fluid, cross-functional "immune system" structure that can swarm problems

Culture

• Evolve from a culture of precision and obedience to one that embraces new discoveries and adapts, like a research lab

Reinventing Business Processes

• Example: Accounts Payable (AP)
  • Traditional AP focused on paying invoices on time and accurately, with vertical, siloed workflows
  • ERP implementations enabled more horizontal, cross-functional processes
  • Agentic AP can have a higher-order goal of optimizing cash flow, with agents dynamically handling tasks like forex trading or involving humans for exceptions

Technical Capabilities for Agentic AI

Intelligence

• Agents need access to a variety of AI models optimized for different tasks and trade-offs (speed, accuracy, cost)
• Example: Amazon Pharmacy using models to reduce prescription fulfillment time by 90% while reducing error rate by 50%

Context

• Agents require an understanding of data relationships, semantics, and memory (priming, procedural, semantic, episodic)
• Techniques like knowledge graphs, vector databases, and machine-readable content are key
• Example: Senta's "Cropwise AI" agents that leverage multiple data sources to provide farmers with optimized action plans

Trust

• Agents need well-defined guardrails and the ability to prove their actions are correct through techniques like automated reasoning checks
• Example: Amazon's text and compliance team using agents to benchmark policies across 600 companies

Getting Started with Agentic AI

• Ideal use cases involve dynamic tool selection, adaptability, pattern recognition, and exception handling
• Examples of high-impact areas include software development, customer support, and knowledge work
• Preparing the organization with training and access to expert guidance is crucial

Key Takeaways

• The shift to agentic AI requires fundamental changes in leadership mental models, business processes, and technical capabilities
• Agentic systems offer the potential for greater agility, adaptability, and business impact compared to traditional automation
• Successful implementation requires rethinking governance, risk management, organizational structure, and culture
• Specific technical capabilities around intelligence, context, and trust are essential to enable agentic AI at scale
• Early adopters are already seeing benefits in areas like software modernization, customer service, and agricultural optimization