Agent Frameworks: LangChain Commands 82% of AI Discovery

When you ask ChatGPT, Perplexity, or Gemini "what framework should I use to build AI agents," one name appears in virtually every response: LangChain. Not occasionally. Consistently, across platforms, across query variants, across time.

Our audit ran 164 AI discovery prompts across the agent frameworks category. LangChain surfaced in 135 of them — an 82.3% shortlist rate. AutoGen and CrewAI both hit 75.0%. LlamaIndex reached 65.9%. After that, the cliff: every other brand sits below 30%.

The category concentration index is 12.2. For context, anything above 10 signals that one brand has built a structural moat in AI discovery. LangChain doesn't just lead. It owns the conversation.

The top four brands separate clearly from everything else. Below LlamaIndex, the next highest rate is 28.1%. Enterprise incumbents like UiPath and Haystack — both credible products with large install bases — barely surface. This isn't a product problem. It's a content footprint problem.

The paradox hiding in the AEO scores

Here's the number that should unsettle LangChain's marketing team: AutoGen and CrewAI each have an AEO score of 86/100. LangChain's is 53/100.

AEO score measures how well a brand's content is structured for AI citation — technical documentation quality, citation signals, content depth and breadth. AutoGen and CrewAI are better optimized. LangChain is less optimized. And yet LangChain leads by 7 percentage points in actual AI discovery.

What that gap tells you: LangChain's dominance is momentum-based, not optimization-based. It built a content footprint early, got cited heavily across the open-source and developer ecosystem, and that historical density became its advantage. AI engines weight authoritative, frequently-cited sources. LangChain has more of them.

The implication runs in both directions. LangChain is vulnerable — a lead built on momentum can erode when competitors catch up on optimization. AutoGen and CrewAI have upside — they're already better optimized, and if they increase their citation density through content velocity, they can close the 7-point gap and potentially flip the shortlist order.

What built LangChain's moat

Early mover advantage in content compounds. LangChain published tutorials, integration guides, and architecture walkthroughs before the category existed as a named category. Those pages aged, accumulated citations, and became the authoritative source that newer AI models trained on.

The GitHub presence reinforces it. LangChain's repository is one of the most starred in the AI ecosystem. Stars, forks, and GitHub discussions are signals AI engines use to evaluate community credibility. LangChain's social proof is massive and broadly distributed — Stack Overflow threads, YouTube tutorials, conference talks, third-party blog posts all pointing back to the same brand.

Content breadth matters too. LangChain covers more ground than any competitor. Use-case guides for customer service agents, data processing pipelines, RAG architectures, multi-agent orchestration. When an AI engine looks for an authoritative source on a specific agent framework use case, LangChain has a page for it.

The tier break below the top four

The 47-point gap between LlamaIndex (65.9%) and handoff.ai (28.1%) is the category's defining feature. It means AI engines have made a clear distinction between established frameworks and everything else.

handoff.ai and base.ai are technically capable products. Their sub-30% shortlist rates reflect limited content surface area, not product quality. AI engines cite what they can find and verify. If there aren't enough indexed, well-structured pages pointing to a brand, it doesn't surface — regardless of how good the product is. Microsoft Copilot Studio at 20.1% is the most surprising finding in the data. Microsoft has massive domain authority, developer mindshare, and distribution. But Copilot Studio isn't showing up in agent framework discussions at the rate you'd expect from a brand of that scale. The product's positioning — more workflow automation than framework — may explain why AI engines route away from it when the query is specifically about agent orchestration.

UiPath at 9.2% and Haystack at 5.5% tell a similar story. Both are credible, well-funded, with large customer bases. But neither has built significant content coverage in the agent frameworks framing. Enterprise buyers evaluating agent frameworks aren't finding them through AI discovery.

What AutoGen and CrewAI should do next

Both brands are in a strong position. Visibility at 75%, maximum model diversity (surfacing across all four major AI engines), and AEO scores of 86/100. The optimization foundation is there. The gap to close is citation density.

The move is to publish more specific, use-case-anchored content. Not "what is AutoGen" content — that ground is covered. The opportunity is "AutoGen for enterprise HR automation" or "AutoGen vs. LangChain for multi-agent customer support" or "how to migrate from LangChain to AutoGen." Specific, comparative, use-case-driven content fills gaps that LangChain's broad coverage misses.

Comparison pages carry disproportionate weight in AI discovery. When a buyer asks "LangChain vs. AutoGen," an AI engine looks for a well-structured page that answers that question directly. If AutoGen publishes that page with depth and technical accuracy, it gets cited. If it doesn't, LangChain's documentation gets cited instead — even in questions that favor AutoGen.

What LangChain needs to protect

An 82.3% shortlist rate on a 53/100 AEO score is an unstable equilibrium. The content advantage is real, but the optimization gap is a vulnerability.

LangChain should treat content refresh as a defensive priority. AI engines weight recency alongside authority. As the agent framework ecosystem evolves — new models, new orchestration patterns, new deployment targets — LangChain's older documentation loses relevance. Competitors with higher AEO scores are publishing fresher, better-structured content into the same topic areas.

The deeper threat is platform bundling. If OpenAI, Anthropic, or AWS decides to feature a specific framework as the recommended default in their developer documentation, that citation becomes a signal that propagates through AI model training. LangChain's lead was built through community citation. It can be displaced by platform-level citation from a single authoritative source.

The takeaway for the rest of the category

Twelve brands made the consensus shortlist across 164 prompts. Most of them are invisible to buyers relying on AI discovery. If your brand is in the agent frameworks category and you're not in the top four, the path forward is narrow: own a specific vertical or use case deeply enough that AI engines cite you first in that context.

The generic "agent framework" query is LangChain's. But "agent framework for healthcare data pipelines" or "agent framework for financial document processing" — those conversations are still open. The brands that publish the most specific, credible, well-structured content in those sub-categories will own them.

Category concentration of 12.2 is high. But every concentrated category was once wide open. The window to establish a defensible position in a specific niche closes fast.