Just spent a few days prototyping with CrewAI and hit a major blocker: the pricing model.
The per-crew-execution cost adds up alarmingly fast for any real workflow. Even a simple two-agent crew with a basic task can burn through credits during development and testing. For production, where you might run hundreds of executions daily, the math gets scary.
Here's a quick breakdown of my test flow:
```python
# Simple crew: Researcher -> Writer
from crewai import Crew, Agent, Task
researcher = Agent(role='Researcher', goal='Find data')
writer = Agent(role='Writer', goal='Write report')
task = Task(description='Research X and write a summary', agent=writer)
crew = Crew(agents=[researcher, writer], tasks=[task])
result = crew.kickoff() # This is one execution
```
Every `kickoff()` is a credit charge. During iterative debugging, that's dozens of executions just to get things working.
Key concerns:
* No clear cost visibility during development in the local SDK.
* Scaling a multi-crew system for parallel processing would be prohibitively expensive.
* Compared to managing raw LLM calls (or even other orchestration tools), the per-execution fee feels like a heavy tax.
Love the framework's design, but the pricing needs a shift—maybe a tier with monthly active agents or included executions. This current model is a non-starter for any substantial project.
Anyone else running into this?
#savings