# What are AI agents? How do they work?

> AI agents run a process under defined rules: they choose steps, use tools and escalate exceptions to a person. Check the nine criteria.

Canonical URL: https://syntalith.ai/en/what-is-an-ai-agent

## Definition
Agent = work + context + tools + operating rules + escalation + security + measurement + cost + trace. A system that misses one of these criteria should be scoped as a script, integration or AI application.

## Triad
Chatbots respond. Copilots help. AI agents carry out work under defined rules.

## Agent-washing
Most "agents" on the market are a repackaged chatbot. The word "agent" is now used for chatbots, RAG applications, copilots, and scripts with a model. Assess the proposed work, decision points, permissions, escalation, measurement, cost, and audit trail. An agent is appropriate only when the next step depends on the content of the case; fixed paths are usually simpler and cheaper to implement as automation.

## Nine questions that take a prototype to production.
Ask these nine questions before approving an agent proposal. The answers should state the work, permissions, approval rules, measurement, cost, and evidence record.
- Work: What specific work does the agent do?
- Context: What must the agent understand: data, rules, exceptions?
- Tools: Which systems does it use, and with what permissions?
- Permissions and limits: What must it never do on its own?
- Escalation: When does it hand off to a human?
- Security: How does it hold up against hostile content, and what protects the data it touches?
- Measurement: How do we know it's worth keeping?
- Cost: What does one pass of the process cost, and what keeps that bounded?
- Trace: What does it record, and who can audit it?

## AI Act mapping
We use the nine agent criteria to organize technical documentation for the operator, relating them to EU AI Act requirements for traceability, transparency, human oversight, robustness and monitoring.
- Trace: Art. 12 · Record-keeping
- Work + Context: Art. 13 · Transparency
- Tools: Art. 13 + 26 · Tools and deployer obligations
- Permissions + human handoff: Art. 14 · Human oversight
- Security: Art. 15 · Accuracy, robustness and cybersecurity
- Measurement: Art. 17 + 72 · Quality and monitoring
Technical input to operator documentation rather than legal advice: we do not classify the system as high-risk and do not determine your obligations. Most of these requirements apply to high-risk systems; separately, Art. 50 (telling a user they are talking to AI) binds providers and deployers from 2 August 2026, regardless of risk class. Cost is a production criterion with no AI Act counterpart.

## Next step

[Book a free call (30 min)](https://meet.syntalith.ai/syntalith/en). With an engineer. No commitment to a paid implementation.

[Current pricing](https://syntalith.ai/en/pricing) · [Examples and sources](https://syntalith.ai/en/work) · [Contact](https://syntalith.ai/en/contact)

[Assistant booking instructions](https://syntalith.ai/booking-agent-guide.txt?locale=en)