AI agents act as operational teammates, ranging from simple rule-based systems to learning systems that improve over time. They help organizations reduce alert noise, speed up incident response, and streamline workflows.
Five types of AI agents
AI agents fall into five types based on how they make decisions, from fixed rules to learning from experience. The table below compares them, followed by a closer look at each type.
|
Agent type |
Definition |
Benefit |
Limitation |
Business example |
|
Simple reflex agents |
These agents respond directly to predefined conditions with immediate action (if X happens, do Y automatically). They lack memory or learning capabilities. |
Fast responses to known issues |
No contextual understanding |
Alert routing to on-call staff, password reset autoresponders, auto-muting of known flaky alerts |
|
Model-based reflex agents |
These agents maintain an internal model of their environment (environmental awareness), enabling context-aware decisions while remaining rule-based. |
More accurate than simple reflex agents |
Still limited in dynamic situations |
Service dependency mapping for root cause analysis, context-aware notification routing, intelligent retry logic |
|
Goal-based agents |
These agents evaluate multiple options to achieve specific business outcomes, choosing actions based on defined objectives. |
Focused on measurable results |
Higher computational requirements |
Smart escalation logic, task redistribution to reduce response times, incident prioritization |
|
Utility-based agents |
These agents optimize for the best possible outcome rather than simply achieving a goal, evaluating outcome quality (not just goal achievement) to make strategic value-driven decisions. |
Enable multi-agent coordination |
Require clear utility models |
Customer impact scoring, cost-versus-stability analysis, dynamic resource allocation |
|
Learning agents |
These agents monitor outcomes and adjust their behavior based on experience and pattern recognition. |
Improve accuracy over time; enable self-optimizing systems |
Require quality training data and oversight |
Recurring incident prevention, adaptive alerting thresholds, post-incident recommendations |
1. Simple reflex agents
These agents respond directly to predefined conditions with immediate action. They lack memory or learning capabilities.
Key characteristics:
- If X happens, do Y automatically
- No contextual understanding
- Fast responses to known issues
Business applications: Alert routing to on-call staff, password reset autoresponders, auto-muting of known flaky alerts
2. Model-based reflex agents
These agents maintain an internal model of their environment, enabling context-aware decisions while remaining rule-based.
Key characteristics:
- Include environmental awareness
- More accurate than simple reflex agents
- Still limited in dynamic situations
Business applications: Service dependency mapping for root cause analysis, context-aware notification routing, intelligent retry logic
3. Goal-based agents
These agents evaluate multiple options to achieve specific business outcomes.
Key characteristics:
- Action selection based on defined objectives
- Higher computational requirements
- Focused on measurable results
Business applications: Smart escalation logic, task redistribution to reduce response times, incident prioritization
4. Utility-based agents
These agents optimize for the best possible outcome rather than simply achieving a goal, making strategic value-driven decisions.
Key characteristics:
- Evaluate outcome quality, not just goal achievement
- Require clear utility models
- Enable multi-agent coordination
Business applications: Customer impact scoring, cost-versus-stability analysis, dynamic resource allocation
5. Learning agents
These agents monitor outcomes and adjust their behavior based on experience and pattern recognition.
Key characteristics:
- Improve accuracy over time
- Require quality training data and oversight
- Enable self-optimizing systems
Business applications: Recurring incident prevention, adaptive alerting thresholds, post-incident recommendations
Bottom line
AI agents automate tasks that once required significant human effort, from managing alert fatigue to building self-optimizing systems.
PagerDuty puts these ideas to work in incident response. SRE Agent analyzes error logs, diagnostics, runbooks, and past incidents to recommend the right fix, and Scribe Agent transcribes incident calls and posts summaries with action items. Start a free trial to see them in action.