Tasks

Without tasks, a crew is a group chat. Tasks make output inspectable and order explicit.

A sprint ticket: owner, description, definition of done. Context is the linked ticket below it.

Visual Workflows

Start here — scroll inside each diagram frame to explore, then use + / − to zoom up to 200% if needed.

100%

Loading diagram...

Scroll inside the frame to explore · use + / − to zoom up to 200%

Key Takeaways

1.A Task is a unit of work: description, expected output, assigned agent, optional context. expected_output is the definition of done — write it like a schema, not be helpful.
2.context from a prior task forwards the previous result so the next agent does not start from zero. Tasks also support structured output, output files, guardrails, and human review.
3.A task names the work, the DoD, and the owner. Placeholders like {topic} are filled from kickoff inputs.
4.Chain tasks with context so the writer sees the research. output_file writes the result to disk when the task completes.

Learn elsewhere

→Processes
→Build a Research Crew

Real Example

Scenario

Research → write. Expected output: markdown with headings + sources, not a nice summary.

What you would do

In CrewAI, apply Tasks to this scenario: Research → write. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Commands

Commands to Remember

Task has expected_output
Chain with context
Placeholders from kickoff inputs
output_file for artifacts

Cheat Sheet

Quick recap

quick ref

•Tasks not vibes
•Chain with context
•Schema the output
•One owner per task

Common Mistakes

✕Skipping evaluation for Tasks before production
✕No logging or tracing around crewai tasks process steps
✕Ignoring cost and latency implications