Field
AI Agents
Models that take actions, and the gap between the demo and production.
An AI agent is a model given the ability to do things rather than just say them — call a search, read a database, send a message. It's the most exciting area in AI and the one with the widest gap between what demos promise and what production delivers.
That gap is the theme of this field, and the entries don't soften it. Multi-agent systems fail in ways that are well documented and rarely mentioned. Agents don't reliably catch each other's errors, because they share training data and therefore share blind spots. A system that's 90% reliable per step is a coin flip after seven steps.
The field covers what makes agents work (tool use, agent memory) and what keeps them safe (guardrails) — where the honest lesson is that the strongest guardrail isn't a clever prompt, it's not granting the capability.
Start with AI Agent, then Guardrails before you build anything that can act.
16 concepts in this field
AI Agent
Software that pursues a goal by taking its own steps — deciding, acting, and reacting — instead of answering once and stopping.
Tool Use
Letting a model call real software — a search, a calculator, your database — instead of trying to answer everything from memory.
Agent Memory
Giving an AI a way to remember across conversations, since the model itself forgets everything the moment a session ends.
Multi-Agent Systems
Several AI agents working together on one problem, each with a role — powerful in demos, awkward in production.
Guardrails
The checks around a model that decide what it's allowed to receive, say, and do — the part that stops a demo becoming an incident.
Function Calling
How a model asks your code to do something — the mechanism underneath every agent, and it's the model requesting, never executing.
Structured Output
Making a model return JSON that always parses — solved at the format layer, and still wide open at the correctness layer.
ReAct
Think, act, look at what happened, think again — the loop underneath essentially every agent, and it's four lines of pseudocode.
Planning
Working out a sequence of steps before taking them — the thing agent demos imply models can do, and the evidence says they mostly can't.
Task Decomposition
Breaking a big job into small ones — which reliably helps, and reliably multiplies your failure rate.
Reflection
Asking a model to critique and fix its own output — which works when there's external feedback, and mostly doesn't when there isn't.
Model Context Protocol
An open standard for connecting models to tools and data — solving a real integration problem, and a standards fight is a political event, not a technical one.
Agent Evaluation
Measuring whether an agent actually works — much harder than evaluating a model, and the reason agent demos and agent products are different things.
Human-in-the-Loop
Putting a person at the decision point — the only reliable safeguard for agents, and it fails quietly when the person becomes a rubber stamp.
Sandboxing
Running an agent where it can't do damage — the only agent safety measure that doesn't depend on the model behaving.
Agent Governance
The controls around a system that acts rather than answers — a different problem from model safety, and the field is early.