Introduction

"The Rise of the Agent: Rails in the AI Era" is a talk by Kinsey Durham Grace presented at Rails World 2025. The session explores the growing significance of intelligent agents in software development, with a particular focus on integrating these agentic systems within the Ruby on Rails ecosystem.

Main Theme

The main topic is the transition from basic AI prompting toward more autonomous, agent-based systems in software, highlighting how these agents can be built, integrated, and utilized effectively in production—especially using Rails.

Key Points

• Defining Agentic AI and Agents:

  • Agentic AI refers to AI systems with the ability to act autonomously, setting and executing goals without continuous human prompting.
  • Agents differ from traditional workflows; they adapt, make decisions, and chain tasks together, providing more flexibility than rigid, rule-based workflows.

• Workflows vs. Agents:

  • Workflows excel at regimented, predefined tasks but lack adaptability.
  • Agents are model-based, capable of handling open-ended, unpredictable problems through learning and adaptation.

• Types and Use Cases for Agents:

  • Various agent types: learning agents, utility-based, model-based, goal-based, and simple reflex agents.
  • Real-world applications include customer support, security threat detection, travel concierges, and especially coding agents, such as those used at GitHub.

• Building Agents in Rails:

  • Essential building blocks discussed:
  • Tooling Layer: Recommends using MCP (Model Context Protocol) for secure interactions with external tools, using an example integrating Zenesk via MCP.
  • Memory: Separating short-term (state store) and long-term memory (timeline of agent thoughts), aiding context-awareness and resumability.
  • Agent Orchestration: Planning, executing, and supervising agent actions, implementing guardrails, and handling business logic, retries, and observability.
  • Planning: Techniques like subgoal decomposition, reflection, self-critique, and chain-of-thought are crucial for agent decision-making.

• Best Practices for Agent Development:

  • Modular and maintainable design for rapid iteration.
  • Implementation of gates (guardrails) and validation, often with tools like dry-schema.
  • Keep prompt changes minimal to avoid unpredictable agent behavior.
  • Extensive testing using evaluations and test harnesses to maintain reliability.
  • Observability and explainability—tracking every action/decision for accountability and debugging.

• Future Trends:

  • Agents are expected to become more human-like, with improved memory, personality, and the capacity to form user relationships.
  • Emergence of sub-agents (hierarchical responsibility) and workflow-native orchestration, blending traditional workflow tools with AI-driven steps.
  • Emphasis on using agents for augmentation rather than replacement of human developers.

• Ethical Considerations:

  • Need for transparency, observability, and regular audits.
  • Addressing bias in LLMs and promoting ethical usage.
  • Implementation of security and retention policies, plus thorough dogfooding.

Takeaways

• Intelligent agents in Rails represent a paradigm shift from procedural, prompt-based AI toward goal-oriented, context-aware systems.
• Successful agent development requires robust architecture, best practices for reliability, and strong ethical safeguards.
• Rails and Ruby ecosystems need to evolve and collaborate to remain relevant in the age of AI agents.
• Agents should augment, not replace, human capability—enhancing productivity and allowing more focus on high-value activities.