The Mash · three stories, one thread
Agent coordination, developer controls, and funding
Reported game tests, two announced funding rounds, and shipped framework updates raise different questions for people building agents.
Explore
Look at the reported roles and hierarchies in the game tests alongside CrewAI’s shipped flow inputs and empty-answer handling. Broccoli AI’s announced funding rounds show investment in agents, but these claims do not connect its work to those tests or to CrewAI.[1][2][3][4][5][6]
Create
Make a small role-based agent test inspired by the reported game tests. Use CrewAI’s shipped run --inputs option for a Python flow, then check how the test behaves when an LLM returns an empty answer.[1][2][3]
Theorize
Hypothesis: Explicit flow inputs and a typed error after an empty-answer retry could help builders compare agent tests more consistently. The reported coordination patterns do not establish that CrewAI reproduces them.[1][2][3]
In the mash
YC Labs details multi-agent tests in games like Minecraft
YC Labs detailed game tests for readers tracking how multiple AI agents coordinate.
Broccoli AI raises Series A and Series B funding
Broccoli AI raised funding for its business selling AI agents to plumbing, HVAC, and electrical companies.
CrewAI releases 1.15.34 with flow and runner updates
CrewAI 1.15.34 adds CLI options and changes agent output handling for developers using the framework.