ref/WhatIsAnAIAgent.com
menu
Last verified: April 2026

AI Agent Vendors: A Neutral Landscape Overview (2026)

This page is not a ranking, a top-ten, or a recommendation. It is a neutral overview of the vendor landscape organised by category, with representative names per category and a one-sentence note on what each one is. Vendor pricing changes monthly. Capability rankings shift quarterly. "Best" is decided by fit-to-use-case, not by absolute rank.

Five categories, fifty or so named vendors. Each named vendor is described in one sentence to help a reader place it on the landscape. Where Digital Signet has an affiliate relationship with a vendor named on this page, the relationship is disclosed in the methodology page. As of April 2026 there are no active affiliate relationships in the agent-vendor space.


01

Foundation model providers

The companies that train the underlying language models. An agent built on top of any of these will inherit the model's capabilities and licence terms.

  • Anthropic
    Claude family. Strong on long context, tool use, and structured output. Anthropic also publishes the Model Context Protocol specification.
  • OpenAI
    GPT family. The largest model marketplace and the broadest tool ecosystem. Also ships the OpenAI Agents SDK.
  • Google
    Gemini family. Tight integration with Google Cloud and Workspace. Vertex AI Agent Builder is Google's agent-platform layer.
  • Meta
    Llama family. Open-weights models that can be self-hosted. Production agents on Llama are typically run via inference platforms like Together or Fireworks.
  • Mistral
    European foundation-model provider with both open-weights and closed models. Strong on multilingual and on cost-efficient deployment.
  • Cohere
    Enterprise-focused models with strong retrieval and command-following capabilities. Less consumer presence than the leaders.
  • AWS Bedrock
    Meta-provider that exposes models from Anthropic, Cohere, Mistral, Meta, and Amazon's own models behind a unified API.
02

Agent-builder platforms

General-purpose platforms for building agents. Differ in language preference, hosted-versus-self-hosted, and how much of the agent loop the platform owns versus the developer.

  • LangChain / LangGraph
    Open-source Python framework with strong community adoption. LangGraph is the agent-orchestration layer. Used heavily by builders comfortable with code.
  • CrewAI
    Multi-agent framework with role-based abstractions. Open-core; the hosted product adds observability and deployment.
  • AutoGen
    Microsoft Research framework for multi-agent conversation patterns. Strong academic adoption, evolving rapidly.
  • Mastra
    TypeScript-first agent framework. Useful when the rest of the application stack is also TypeScript.
  • OpenAI Agents SDK
    OpenAI's first-party agent framework, tightly integrated with GPT models and OpenAI's tool ecosystem.
  • Anthropic Claude Agent SDK
    Anthropic's first-party agent framework. MCP-native, tightly integrated with Claude.
03

Vertical agent platforms

Agent products targeted at a specific business function. The agent loop is hidden behind a vertical-shaped UI; the buyer pays for the integration and the prompt engineering, not for the model.

  • Sales: Clay, Apollo, Common Room
    Lead enrichment, outbound personalisation, pipeline intelligence. The agent layer rides on top of established sales-tech platforms.
  • Support: Intercom Fin, Zendesk Resolve, Decagon
    Customer-support deflection and routing. Generally a pure-LLM agent with retrieval and a hand-off-to-human escalation path.
  • Engineering: Cognition Devin, Cursor, GitHub Copilot Workspace
    Coding agents with various levels of autonomy. Cursor in agent mode and Devin sit at the more autonomous end; Copilot Workspace is closer to a copilot than an agent.
  • Operations: Glean, Moveworks, Atomicwork
    Enterprise-search-plus-agent platforms aimed at internal IT and ops workflows.
04

RPA vendors with agent layers

Legacy RPA platforms have all added agent layers since 2024. Quality varies; the architectural starting point shows in the failure modes.

  • UiPath
    Largest RPA vendor. Has shipped agent products that combine LLM-driven decision points with the deterministic RPA execution layer.
  • Automation Anywhere
    Agentic Process Automation (APA) framing. Similar architectural pattern to UiPath.
  • Blue Prism (now SS&C Blue Prism)
    Older codebase; the agent pivot has been slower than competitors.
05

Open-source frameworks

The open-source landscape moves fast. A 2026 reference will be partly stale by 2027. Names below are representative of categories rather than recommendations.

  • LangGraph
    The most-used open-source agent orchestration library. Mature ecosystem, strong observability tools.
  • AutoGen
    See above. Open-source counterpart of Microsoft's research work.
  • CrewAI
    Open-core, with a hosted control plane available.
  • MetaGPT
    Multi-agent framework focused on software-engineering team simulations.
  • AGiXT
    Plugin-oriented agent platform. More experimental than the leaders.

Section 6

How to shortlist

The procurement question is not "which vendor is best". It is "which vendor fits this use case at this scale on this stack". Three filters narrow the field quickly. First, language and platform: a Python-shop with cloud-agnostic infrastructure has a different shortlist from a Microsoft-shop committed to Azure. Second, build-versus-buy: a team of senior engineers can use a low-level framework; a team that wants an agent in production this quarter probably needs a vertical platform. Third, integration depth: the agent will be as useful as the systems it can talk to. A vendor with strong integrations into your CRM and your data warehouse beats one with stronger raw capabilities and weaker integrations.

The full procurement-grade evaluation framework, with capability, reliability, cost, latency, and a 12-item checklist, is on how to evaluate an AI agent.

Concrete agent use cases by business function are on AI agent examples. Use the examples to clarify what you actually want the agent to do before talking to a vendor.