Agents

• Agents can plan, reason and execute multi-step tasks autonomously or with minimal guidance.
• They chain together tool calls, memory retrieval and decision making to interact and take real-world actions.
• Core loop is: Observe → Think → Act → Repeat

Current State

• Agents are finally becoming good enough for day-to-day use.
• With good models they're already quite productive with minimal nudging.
• Model reasoning quality has crossed a threshold where agents can recover from mistakes mid-task.
• Without cost subsidization through subscriptions they get insanely expensive though.

Agent Frameworks

The leading frameworks for agents are LangGraph (Python) and Mastra (JavaScript).