// verdict · checked 2026-08-11 · CrewAI 0.x crews and flows (docs.crewai.com, 2026)
CrewAI
CrewAI
A framework can't be an agent, but it can sell the dream wholesale.
- CAPABILITY
- 5/5 AGENT
- PROOF
- OPERATOR INSPECTABLEYou can run it and read the steps yourself — the record exists for whoever runs it
- TRACE ACCESS
- OPERATOR-OWNEDWhoever runs it holds the record; nothing is published
- LEASH
- Not measuredno execution record scored — measure it yourself
CrewAI is a framework, but unlike most it ships its own reasoning loop and tools, so it scores as what it runs out of the box.
| LOOPSruns until doneyes | Out of the box each CrewAI agent runs a framework-provided (not developer-authored) reasoning/tool loop that varies steps with the task until done; max_iter (default 20) is a ceiling, which the rule says does not disqualify. |
| CHOOSESpicks its own toolsyes | Agents pick tools at runtime; the hierarchical process has a manager LLM allocate tasks to agents. |
| ACTSchanges the worldyes | CrewAI ships first-party FileWriterTool (writes/overwrites files on disk, creates dirs) and code-execution tooling that runs agent Python in Docker — the framework itself executes these state-changing tools, satisfying ACTS ("writes a file, runs a command"). |
| RECOVERSfixes its own errorsyes | Docs: failed guardrail sends the error back so "the agent attempts to fix the issue" (max 3), and on a code exception the agent "can attempt to correct the code or provide alternative solutions" — different action, no human. |
| UNSUPERVISEDnobody watchingyes | Live docs confirm crew.kickoff() runs all tasks to completion with human_input defaulting to False — a documented default mode that completes a full task with no per-step approval. That is what the framework does out of the box. |
Capability decisions combine documented behaviour and execution evidence. The proof label on each row shows which of the two decided it, and Leash is reported separately, only when it was measured from an execution record. Primary source: docs.crewai.com. Tested against: CrewAI 0.x crews and flows (docs.crewai.com, 2026). Our confidence in this dossier: 7/10.
Can you see it work?
the weakest of the five grades aboveOPERATOR INSPECTABLE — You can run it and read the steps yourself — the record exists for whoever runs it. Trace access: OPERATOR-OWNED — Whoever runs it holds the record; nothing is published.
built-in tracing plus Langfuse/Phoenix/MLflow exporters; you run the process
Where we looked: docs.crewai.com/llms.txt: observability overview, built-in tracing, plus Langfuse/Arize Phoenix/MLflow/Opik/Weave integrations and streaming runtime contract. · see the evidence →
- 2026-08-114/5 AGENT → 5/5 AGENTRECOVERS corrected to yes: Docs: failed guardrail sends the error back so "the agent attempts to fix the issue" (max 3), and on a code exception the agent "can attempt to correct the code or provide alternative solutions" — different action, no human.editorial correction — no case was filed
- 2026-08-113/5 WRAPPER WITH AMBITION → 4/5 AGENTUNSUPERVISED corrected to yes: Live docs confirm crew.kickoff() runs all tasks to completion with human_input defaulting to False — a documented default mode that completes a full task with no per-step approval. That is what the framework does out of the box.editorial correction — no case was filed
- 2026-08-112/5 WRAPPER WITH AMBITION → 3/5 WRAPPER WITH AMBITIONACTS corrected to yes: CrewAI ships first-party FileWriterTool (writes/overwrites files on disk, creates dirs) and code-execution tooling that runs agent Python in Docker — the framework itself executes these state-changing tools, satisfying ACTS ("writes a file, runs a command").editorial correction — no case was filed
- 2026-08-111/5 CRON JOB WITH EXTRA STEPS → 2/5 WRAPPER WITH AMBITIONLOOPS corrected to yes: Out of the box each CrewAI agent runs a framework-provided (not developer-authored) reasoning/tool loop that varies steps with the task until done; max_iter (default 20) is a ceiling, which the rule says does not disqualify.editorial correction — no case was filed
Badges
earned, never for sale
These render from the live dataset, so a badge cannot outlive the verdict it claims. If the verdict changes the badge changes with it, and an unearned one returns 409 rather than an image. No sponsor can buy one, at any price.
Think this is wrong?
bring evidence, not opinionName the criterion and link a trace, a doc, or a recorded run that shows the behaviour. A verdict changed by evidence is the best thing that can happen to this site, and the change goes on the public record with your reason attached. Vendors are welcome, and a vendor's own filing is labelled as such rather than buried.