- Is LangChain or CrewAI better for multi-agent systems?
- CrewAI is the more opinionated of the two about multi-agent specifically, and it says so: it is a "fast and flexible multi-agent automation framework" whose two primitives are Crews, for role-based autonomous collaboration, and Flows, for event-driven control. LangChain's core is broader. "a framework for building agents and LLM-powered applications". And its answer for orchestration is a separate package, LangGraph, described on its own README as the framework "for building controllable agent workflows". So the honest comparison for multi-agent work is CrewAI against LangGraph rather than against LangChain, and the choice comes down to whether you want role-based crews or a low-level graph.
- Are LangChain and CrewAI still actively maintained?
- Neither repository publishes an answer to that question, which is why it needs measuring rather than reading. Neither README carries a maintenance statement, a deprecation notice or a successor. Unlike AutoGen, which states plainly that it "is now in maintenance mode" and "will not receive new features or enhancements". StillShipping reads those figures from the GitHub API every night. Commit recency, release cadence, issue response time and maintainer count. And publishes them at /is/langchain-maintained and /is/crewai-maintained, both scored on the same night so the numbers are comparable rather than merely present.
- Are they free?
- Both frameworks are MIT-licensed and free to use, and both companies sell something beside them. CrewAI publishes the AMP Suite "for organizations that need a commercial control plane around CrewAI", adding managed deployment, observability, governance, security and enterprise support, with the Crew Control Plane free to try. LangChain publishes LangSmith for "agent evals, observability, and debugging" and a separate deployment platform. In both cases the paid product is operational tooling around the framework rather than a gate on the framework itself. Which matters for licensing and matters for maintenance, because it is the paid product that funds the free one.
- How do you install each one?
- Both are Python, both install with uv, and both publish a one-line quickstart. LangChain: `uv add langchain`, then `init_chat_model` and invoke. CrewAI installs as a CLI rather than a library. `uv tool install crewai`, then `crewai create crew <project_name>` to scaffold a project from YAML configuration, `crewai install` for its dependencies and `crewai run` to start it. LangChain additionally publishes langchain.js as a JS/TS equivalent; CrewAI is Python only.
- What does StillShipping measure that a comparison article does not?
- Whether the project is still moving, on the night you read the page. A comparison article is a snapshot of the day it was written, and the thing most likely to have changed since is not the API. It is whether anyone is still merging. Every tool on this site is rescored nightly from the GitHub API on commit recency, release cadence, issue response time and maintainer count, and each tool's own page says which of those moved. That is the axis a listicle structurally cannot cover, and it is the one that decides whether a framework you adopt today is a liability in eighteen months.
- What is the difference between Spec Kit and BMAD-METHOD?
- Spec Kit is a toolkit that makes the specification the executable artifact. "define what to build before building it. With any AI coding agent". And it is explicitly agent-agnostic and extensible, shipping a ready-to-use process you can replace with your own. BMAD is a method with skills attached: you run `npx skills add bmad-code-org/BMAD-METHOD`, then invoke `bmad-build` with what you want to change inside your existing AI coding tool, and its process expands or contracts with the size of the change. The practical difference is where the work lands: Spec Kit produces a spec your agent implements; BMAD runs a conversation that keeps the decisions explicit and carries them forward as context.
- Which one is still being maintained?
- Neither repository answers that on its front page, which is precisely why this site scores them. Neither README carries a maintenance notice, a deprecation warning or a successor. The contrast worth knowing is AutoGen, which states outright that it "is now in maintenance mode" and "will not receive new features or enhancements". The figures under this verdict are read from the GitHub API every night: commit recency, release cadence, issue response time and maintainer count, both projects measured on the same night so they can be held against each other.
- Do either of them cost anything?
- No. Both are MIT-licensed. BMAD states that it is "free and open source, with no paywalled workflows or gated community". Spec Kit publishes no paid tier, hosted plan or commercial suite on its README. Neither uses the arrangement LangChain and CrewAI use, where an MIT framework sits beside a paid control plane.
- What do you need installed to run each?
- Spec Kit needs uv, and its install is version-pinned by design: "Replace vX.Y.Z with the latest release tag from Releases. Keep the leading v (for example, v0.12.11, not 0.12.11)", installed as `uv tool install specify-cli`, or from PyPI. It is the only one of the five frameworks tracked here whose published install instructions pin a version. BMAD's Skills CLI route needs Node.js, npm, and Git, and installs with `npx skills add bmad-code-org/BMAD-METHOD`; BMAD setup scripts use uv. After that the two diverge: `specify init my-project --integration copilot` scaffolds a project, where BMAD expects you to open your project in your AI coding tool and invoke its skills there.
- Can you use them together?
- Nothing in either README forbids it, and BMAD's own framing invites the arrangement: "start anywhere. Use BMAD end to end, or carry its briefs, specifications, and architecture into your existing delivery workflow." Spec Kit is agent-agnostic and describes its process as replaceable. "a ready-to-use spec-driven process (or bring your own)". Neither publishes an integration with the other, so this is a statement about what their own documentation permits rather than a documented path.
- What are the best alternatives to LangChain?
- CrewAI is the strongest direct alternative if your problem is multi-agent: it is a "fast and flexible multi-agent automation framework" with two explicit primitives. Role-based Crews for autonomy, event-driven Flows for control. And it states it is standalone rather than a layer on something else. AutoGen was the other obvious answer and is no longer one for new projects: its README puts it in maintenance mode and points new users at Microsoft Agent Framework. If what you actually want is orchestration rather than a different framework, LangChain's own answer is LangGraph, which its README describes as the framework "for building controllable agent workflows".
- Is AutoGen deprecated?
- Its own README describes it as being in maintenance mode, which is close to but not the same as deprecated: "AutoGen is now in maintenance mode. It will not receive new features or enhancements and is community managed going forward. New users should start with Microsoft Agent Framework. Existing users are encouraged to migrate using the AutoGen → Microsoft Agent Framework migration guide." So it still installs, it is still MIT (alongside a CC-BY-4.0 licence for non-code content), and the code you already depend on has not been withdrawn. But nothing new is coming, and Microsoft has said where to go instead. That is a far better position to be in than an unmaintained project that says nothing.
- Which agent framework is most actively maintained?
- That is a question about this week rather than about the documentation, and none of the three publishes an answer. The table above records what each project says about itself; StillShipping's board records what each repository actually did, rescored nightly from the GitHub API on commit recency, release cadence, issue response time and maintainer count. AutoGen is the only one of the three whose README says anything about its own status at all. Treating the other two's silence as evidence of health is the specific mistake this site exists to make unnecessary.
- Do any of these lock you into a paid product?
- Not at the licence, and it is worth knowing where the money is anyway. All three are MIT-licensed. LangChain sells LangSmith. "agent evals, observability, and debugging for LLM apps". And a deployment platform beside it. CrewAI sells the AMP Suite, "a commercial control plane around CrewAI" adding managed deployment, observability, governance, security and enterprise support. AutoGen publishes no paid tier at all, which in its case is less reassuring than it sounds: what funds a free framework is usually the paid product beside it, and AutoGen's successor is the thing Microsoft is now investing in.
- How is this list different from a roundup article?
- A roundup is a snapshot of the day someone wrote it, and the thing most likely to have changed since is not the API surface. It is whether anyone is still merging. Every claim in the table above is a quotation from the project's own repository with the date it was read, and every figure on the board is measured from the GitHub API on the night you are reading. Where a project publishes nothing on an axis, the cell says "Not published" and names the page that was read rather than inferring an answer. Which is how a table stays a reference rather than becoming an opinion.
- What is a Claude Code orchestrator?
- A layer that decides what the coding agent does next, rather than doing the coding itself. In practice two different things get called that. One is an agent runtime. LangChain, CrewAI, AutoGen. Which you import and which runs agents against a model. The other is a process framework. Spec Kit, BMAD-METHOD. Which produces specifications and decisions and then hands them to whatever coding agent you already use; Spec Kit is explicit that it works "with any AI coding agent", and BMAD expects you to "open your project in your AI coding tool" and invoke its agents there. This page ranks both, because both are things people install on top of Claude Code, and both stop being maintained the same way.
- Which orchestrators have stopped shipping?
- Several, and the ranking below names them with the date each was last measured rather than describing them in prose that ages. The one that says so itself is AutoGen: "AutoGen is now in maintenance mode. It will not receive new features or enhancements and is community managed going forward." That candour is rare. Of the five frameworks in the table above, it is the only one whose README makes any statement at all about its own maintenance. The rest are scored from the GitHub API: commit recency, release cadence, issue response time and maintainer count, rescored every night.
- Why rank by freshness instead of by stars?
- Because stars measure the past and freshness measures the present. A project that collected forty thousand stars in a good year keeps them after the last maintainer walks away, and every roundup that ranks by stars will keep recommending it. The failure this creates is specific and expensive: you adopt something on a signal that cannot fall, and discover eighteen months later that your issues are unanswered. Commit recency, release cadence and issue response time all fall the moment a project stops, which is the property a health signal needs.
- Should I use a runtime or a process framework with Claude Code?
- Match it to what is actually failing. If the agent produces code that does not do what you meant, the problem is upstream of the runtime and a process framework is the answer. Spec Kit if you want the spec to be the executable artifact and to stay agent-agnostic, BMAD if you want the depth of planning to scale with the size of the change and to keep the decisions explicit as context. If instead you are building a product where several agents coordinate at runtime, you want a runtime: CrewAI for role-based crews and event-driven flows, LangChain for the largest integration surface. Adopting a runtime to fix a specification problem is how teams end up with both and neither working.
- Are these free to use?
- All five are MIT-licensed, and what differs is what sits beside them. LangChain sells LangSmith and a deployment platform; CrewAI sells the AMP Suite as "a commercial control plane". Spec Kit and AutoGen publish no paid tier on their READMEs. BMAD publishes the absence directly: "BMad is free and open source, with no paywalled workflows or gated community".