Skip to content

Exercises — Week 3 — Sequential vs hierarchical

Do these after reading Week 3. Sketch both process types. Do not implement a voting/CEO consensus loop that never runs.

from crewai import Agent, Crew, Process, Task

Predict before you run

Does building a Crew(..., process=Process.sequential) call an LLM? Who owns the final string in sequential vs hierarchical?

Runnable command

python your_crew_process.py   # construct only — no kickoff()

Each task has three hints, closed by default. Open only as far as you need.

1. Sequential crew

Two agents, two tasks, context=[t1] on the second, process=Process.sequential.

Checks:

  • crew.process == Process.sequential
  • t2.context == [t1]
  • No kickoff() required
Hint 1 — a nudge

In sequential mode, what decides the order the work happens in?

Hint 2 — the approach

Two Agents, two Tasks (the second with context=[t1]), and Crew(..., process=Process.sequential). The task list is the order; there's no manager.

Hint 3 — most of the code 
from crewai import Agent, Crew, Process, Task

researcher = Agent(role="researcher", goal="list churn reasons", backstory="extracts facts", verbose=False)
writer = Agent(role="writer", goal="five-line summary", backstory="writes from facts only", verbose=False)
t1 = Task(description="List three churn reasons from tickets.", expected_output="bullet list", agent=researcher)
t2 = Task(description="Write a 5-line summary.", expected_output="five lines", agent=writer, context=[t1])

seq = Crew(agents=[researcher, writer], tasks=[t1, t2], process=Process.sequential)
print(seq.process, t2.context == [t1])

2. Hierarchical flag

Build a second Crew with process=Process.hierarchical and a manager_agent whose goal names the manager's one job. CrewAI 1.x refuses to construct a hierarchical crew without a manager.

Checks:

  • hier.process == Process.hierarchical
  • In two sentences: who owns the final string in sequential vs hierarchical
Hint 1 — a nudge

Hierarchical adds a role that isn't in your agents list. Who is it, and what does it need that the others don't?

Hint 2 — the approach

Same agents and tasks, process=Process.hierarchical, plus manager_agent=Agent(..., allow_delegation=True). Keep the manager out of agents=. Try it once without a manager first and read the ValidationError — it fails at construction, before any tokens are spent. Your two sentences contrast "the last task's agent" with "the manager."

Hint 3 — most of the code 
try:
    Crew(agents=[researcher, writer], tasks=[t1, t2], process=Process.hierarchical)
except Exception as exc:  # CrewAI 1.x validates the manager at construction
    print("no manager:", type(exc).__name__)

manager = Agent(role="release manager", goal="Assign each ticket; do not rewrite the output.",
                backstory="Owns the order of work, not the words.", allow_delegation=True, verbose=False)
hier = Crew(agents=[researcher, writer], tasks=[t1, t2], process=Process.hierarchical, manager_agent=manager)
assert hier.process == Process.hierarchical

3. No ballot

Do not add vote/re-vote/CEO-aggregate code.

Checks:

  • The file has no votes = / consensus loop
  • One sentence: a human gate is LangGraph week 4, not a CrewAI poll
Hint 1 — a nudge

If two agents disagree, who should decide — a third agent, or a person with a pager?

Hint 2 — the approach

Grep your file for vote and consensus. Then write the sentence pointing at interrupt_before from LangGraph week 4.

Hint 3 — a skeleton 
When the crew's output needs a decision, <who> approves it via <LangGraph mechanism>, because <why a poll of models isn't accountability>.

Expected observation

Open after you run

crew.process == Process.sequential and t2.context == [t1]. Hierarchical flag set without a fake manager LLM.

Self-check

No vote/re-vote/CEO loop. A human gate is LangGraph week 4, not a CrewAI poll.