What is an Agent?

What is an Agent?

An agent is a configurable AI-powered assistant that can perform tasks, automate workflows, and interact with tools (MCPs). Each agent is defined by a set of instructions, optional arguments, and access to tools like version control, the file system, CI pipelines, or cloud environments.

Agents act as task-specific operators. You can think of them as highly focused copilots, each with a clear objective and strategy. Rather than manually guiding an AI through each step, you define the desired behavior once and reuse it reliably.

Why should I use Agents?

Using agents has several benefits:

• Customization: Define detailed instructions, tool access, and behavior tailored to your workflow.

• Repeatability: Agents behave consistently across runs, environments, and users.

• Automation: Agents can replace repetitive manual work: code reviews, testing, deployments, and more.

• Collaboration: Teams can share, version, and build on each other’s agents for unified engineering workflows.

---

Agent Types

Agents and Task Agents share the same creation flow (name, instructions, tool access, rules), but they differ significantly in usage:

• Modes are conversational and persistent.

• Workflows are command-style and stateless.

Modes

Modes are persona-driven agents and assistants designed for ongoing, multi-turn conversations. You can interact with them in a way that aligns with a specific role or mindset.

Examples:

• Architect Mode: Helps with system design and deep architectural questions.

• Lite Mode: Provides fast, lightweight responses using minimal tools.

Modes are selected from a dropdown menu in the chat, and remain active across multiple messages, adapting to your workflow during a session.

Workflows

Workflows are single-task executors. They are not meant for back-and-forth interaction but are triggered to perform a specific job, similar to Qodo’s slash commands.

Examples:

• Test Agent: Automatically writes tests based on predefined logic.

• Review Agent: Runs a static code review with structured output.

Workflows are launched via slash commands, run independently of chat context, and return the chat to its previously active agent after completion.

Workflows optionally include:

• Plan Mode: A preliminary step to decide how the task will be executed.

• Output Schema: A structured format for consistent, predictable results.