Understand the spectrum from simple reactive agents to fully autonomous ones, and learn how to choose and combine agent types for your specific use case.
Reactive agents
A reactive agent listens for a specific trigger — a webhook, an incoming message, or a schedule — and responds with a predetermined set of steps. The LLM may still be involved in formatting a response or extracting fields from text, but the overall flow is largely fixed. Reactive agents are fast to build, easy to reason about, and ideal for tasks like triaging support tickets, summarizing incoming emails, or routing form submissions. On Cotonity, you configure a reactive agent by defining a single trigger, a short prompt, and one or two output actions.
Autonomous agents
An autonomous agent is given a goal rather than a script. It uses its LLM to decide which tools to call, in what order, and what to do with the results — repeating the perceive–reason–act loop until the goal is achieved or a stop condition is met. Autonomous agents are well suited for research tasks, complex data enrichment, and multi-system orchestration where the exact steps cannot be known in advance. They require more careful prompt design, error handling configuration, and monitoring because each run can follow a different path.
Choosing the right type for your use case
Most real-world use cases fall somewhere between purely reactive and fully autonomous. Cotonity supports a hybrid model: you define a skeleton workflow with fixed entry and exit points, but allow the agent to make decisions within certain steps. For example, a customer onboarding agent might always start with a CRM lookup (fixed) and always end with a welcome email (fixed), but let the LLM decide which resources to include in that email based on the customer's industry and plan tier. This bounded autonomy gives you predictability at the workflow level while still benefiting from the LLM's reasoning.
Upgrading a reactive agent to autonomous
If you already have a reactive agent and want to give it more flexibility, you do not need to rebuild it from scratch. In the Cotonity visual editor, you can add a planning step before your existing action sequence, provide the agent with additional tools it can call optionally, and set a maximum iteration count to bound its autonomy. Start with a low iteration limit (two or three) and expand it after observing the agent's behavior in sandbox mode. Always review the agent's reasoning trace in the run logs to confirm it is choosing tools for the right reasons before increasing its autonomy.