Category Hub
AI Agents
Autonomous workflows that plan, call tools, and ship results.
Agent patterns that actually work: tool use, memory, planning loops, evals, and the frameworks worth your time.
Agents are the hyped end of AI, so this hub tries hard to be honest. We cover what agents can genuinely do today, what still needs a human in the loop, and the patterns that keep them from wandering off.
You'll find framework picks (LangGraph, CrewAI, OpenAI's Assistants, and lighter-weight options), how to give agents real tools without giving them your whole computer, memory patterns that don't blow up context, and evals so you know when your agent got worse.
If you're building your first agent, start with the single-task walkthrough. If you're already shipping agents, jump to evals and the multi-agent patterns.
Tools & apps in this category
We're publishing reviews here this week. In the meantime, browse the full App Directory.
Frequently asked about AI Agents
›Are AI agents production-ready?
Single-task agents with narrow tools, clear success criteria, and a human review step: yes, in production today. Fully autonomous multi-agent swarms with unbounded scope: no — still a research demo more than a product.
›Which agent framework should I start with?
For most people: OpenAI's Assistants API if you're already on OpenAI, LangGraph if you want fine control, CrewAI if you like a role-based mental model. For automation-heavy jobs, n8n's AI Agent node is often the fastest path.
›How do I stop an agent from wasting money or going off the rails?
Budget guardrails on tokens and tool calls, a strict tool allowlist, timeouts on every step, and evals that fail loud. If you can't answer 'how would I know this got worse?', you're not ready to leave it running.
›Agents vs plain automations — when do I actually need an agent?
If the steps are known ahead of time, use an automation. Use an agent when the model has to decide the steps — research, triage, reasoning across unstructured inputs. Most 'agent' projects turn out to be automations with an LLM step.