Learning Resources

Collects 75 real AI agent setups, 20 that people made money with, each with source, prompt, and steps.

The benchmark paper for evaluating LLMs as agents across diverse environments.

Collects sourced playbooks for real agent jobs with prompts, steps, human checkpoints and known failures.

Short course on building production agents with LangGraph by Andrew Ng's platform.

Comprehensive guide on AI systems design and deployment covering agent architecture patterns.

Collection of Claude agent recipes and integration patterns from Anthropic.

Anthropic's guide on agent design patterns, evaluation strategies, and production best practices.

Open-source course on building AI agents using Hugging Face tools and models.

EMNLP 2025 paper introducing PersonaEvolve, an LLM-based optimizer that refines agent personas so crowds of LLM agents behave realistically against expert benchmarks.

Universal suffix that manipulates text-embedding similarity to bypass safety guardrails across ChatGPT, DeepSeek, and Qwen.

Combines Monte Carlo tree search with LLM reasoning for complex multi-step planning tasks.

Survey of LLM agents organized by function, covering planning, memory, tool use, multi-agent systems, evaluation, and safety.

Deep breakdown of LLM-powered agent components: planning, memory, and tool use.

A 21-lesson course on generative AI concepts and agent development from Microsoft.

Practical guides and recipes for building with OpenAI APIs including agent patterns.

Community-maintained guide covering prompt engineering techniques and agent strategies.

The foundational paper behind the ReAct prompting pattern used in most agent frameworks.

Research framework letting agents learn from past mistakes via iterative verbal self-reflection loops.

Survey of the agent loop: control strategies, trained loops, termination and recovery, skills, and harness design.

Explores multiple parallel reasoning paths before committing to a final answer for complex problems.

Systematizes 56 autonomous research systems across seven axes, showing most can generate but few can defend research artifacts.