Back to Continue

The Cloud Agents Taxonomy

docs/guides/cloud-agents/cloud-agents-taxonomy.mdx

1.5.456.2 KB
Original Source

Why This Taxonomy Exists

“Cloud agents” is a term that gets used to describe everything from cron jobs to IDE copilots.

That ambiguity causes teams to:

  • choose the wrong abstraction
  • automate too early
  • or avoid cloud agents entirely because they feel risky

This guide exists to:

Define what cloud agents actually are, how they differ from other agents, and when they make sense for software teams.

A Clear Definition of Cloud Agents

<Info> A **cloud agent** is an AI-driven process that runs on remote infrastructure, is triggered by tasks, schedules, or external events, and uses reasoning over changing inputs to produce reviewable outcomes across shared engineering systems. </Info>

Key characteristics:

  • runs outside a developer’s local environment
  • operates on shared team context (repos, alerts, analytics, infra)
  • produces reviewable outcomes (PRs, reports, actions)
  • can run asynchronously and repeatedly

Cloud agents exist to handle work that benefits from reasoning, shared context, and reviewability.

What Cloud Agents Are Often Confused With

Cloud agents are often confused with other tools.

<CardGroup cols={2}> <Card title="Not IDE Agents"> IDE agents focus on individual productivity and local context. Cloud agents operate on shared systems and team-owned work. </Card> <Card title="Not Just Cron Jobs"> Cron is a trigger. Cloud agents reason over changing inputs and produce reviewable outcomes. </Card> <Card title="Not One-Off Scripts"> Scripts execute fixed logic. Cloud agents apply reasoning and shared context even for a single run. </Card> <Card title="Not Magic Automation"> Cloud agents still require review, ownership, and guardrails. </Card> </CardGroup>

The Two Axes That Matter

Cloud agents are best understood along two dimensions:

  1. What triggers them
  2. How they’re governed

Everything else is implementation detail.

mermaid
quadrantChart
  title Cloud Agents Taxonomy
  x-axis One-Off --> Event-Driven
  y-axis Manual --> Automated

  quadrant-1 Automated & Event-Driven
  quadrant-2 Manual & Event-Driven
  quadrant-3 Manual & One-Off
  quadrant-4 Automated & Scheduled

  "Weekly dependency updates" : [0.65, 0.85]
  "Security scan with PR review" : [0.75, 0.60]
  "One-off repo audit" : [0.25, 0.25]
  "Nightly analytics summary" : [0.45, 0.75]
  "Incident response triage" : [0.85, 0.40]

Most teams start bottom-left and move diagonally as trust increases to create their Continuous AI Workflow.

Axis 1: Cloud Agent Trigger Type

1. One-Off Cloud Agents

Manually triggered when a human explicitly dispatches work to an agent. One-off cloud agents are first-class agents that differ by trigger, not by capability.

<Columns cols={2}> <Card title="Examples"> - Analyze the top recurring Sentry errors - Audit a repo for security issues - Summarize system behavior </Card> <Card title="Best for"> - Unknown scope - Exploratory work - Building confidence before automation </Card> </Columns>

2. Scheduled Cloud Agents

Time-based triggers that run on a fixed cadence (daily, weekly, hourly).

<Columns cols={2}> <Card title="Examples"> - Weekly vulnerability scans - Nightly analytics summaries - Regular dependency updates </Card> <Card title="Best for"> - Predictable recurring work - Backlog prevention - Drift detection </Card> </Columns>

3. Event-Driven Cloud Agents

Signal-based triggers that react to real-time events from external systems.

<Columns cols={2}> <Card title="Examples"> - New high-severity alert - Failed deployment - Critical security finding </Card> <Card title="Best for"> - Operational workflows - Incident response - Time-sensitive fixes </Card> </Columns>

Axis 2: Cloud Agent Governance Level

This is where most teams get into trouble.

<Tabs> <Tab title="Manual"> - Human decides when the agent runs - Human reviews everything
*Safest starting point.*
</Tab> <Tab title="Assisted"> - Agent runs automatically - Human approves outcomes
*Where most teams should live first.*
</Tab> <Tab title="Automated"> - Agent runs unattended - Outcomes merge or act directly
*Earned over time — not assumed.*
</Tab> </Tabs> <Warning> Cloud agents don’t fail because they’re autonomous. They fail because teams skip the control phase. </Warning>

Cloud Agents vs Local / IDE Agents

Local / IDE AgentsCloud Agents
Individual contextTeam context
Short-livedLong-running
Developer-initiatedSystem-initiated
Productivity focusReliability & ownership focus

Both are useful. They solve different problems.

Why Cloud Agents Exist at All

Most engineering pain is recurrence.

  • alerts repeat
  • security issues resurface
  • operational work interrupts feature work

Cloud agent workflows exist to:

  • notice repetition
  • respond consistently
  • reduce future interruptions

The goal is not more automation. The goal is less work over time.

When Cloud Agents Make Sense

Use cloud agents when:

  • the same class of issue keeps returning
  • the cost is in handling, not solving
  • work spans multiple systems
  • review and traceability matter

Avoid them when:

  • the problem is novel
  • ownership is unclear
  • blast radius isn’t understood yet

The Architectural Reality

As teams adopt cloud agents, they discover:

Agents alone don’t scale without control and visibility.

Production cloud agents eventually require:

  • a shared place to review runs
  • a record of decisions
  • adjustable prompts and rules
  • gradual automation

Without this, cloud agents become brittle fast.


Where to Go Next

<CardGroup cols={2}> <Card title="Developer’s Guide to Cloud Agents" href="/guides/cloud-agents/guide-to-cloud-agents"> Learn how cloud agents are implemented and used in practice. </Card> <Card title="Mission Control" href="https://mission.continue.dev"> Explore how teams manage cloud agents safely at scale. </Card> </CardGroup>

One Sentence to Remember

<Callout type="success"> Cloud agents are how teams take ownership of work that happens after code ships. </Callout>