CANONICAL DEFINITION PAGE
What Are the Types of AI Agents?
Insights from IDC's Worldwide Agentic AI Adoption Guide Taxonomy, 2026, and IDC's Worldwide Artificial Intelligence Services Taxonomy, 2026
IDC classifies AI agents into four tracked types by how they are built and deployed: in-app, standalone, custom-configured, and bespoke agents, with on-device agents emerging as a fifth type that IDC does not yet track.
What is the types of AI agents?
IDC defines four primary AI agent types: in-app agents, standalone agents, custom-configured agents, and bespoke agents. According to IDC, the first two are built and operated by an agent provider, and the second two are custom built to meet an organization's requirements. This matters because IDC measures each type by agent actions taken, agent action token load, and agent token delivery costs, which makes the types the segmentation behind its agentic AI adoption data.
According to IDC's Worldwide Artificial Intelligence Services Taxonomy, 2026 (IDC #US54923426, October 2026), "AI agents are commonly developed and deployed in five different contexts." IDC's Worldwide Agentic AI Adoption Guide Taxonomy, 2026 (IDC #US54650726, June 2026) tracks four of them, and IDC says on-device agents are a fifth type that will emerge in the next five years.
AI Agent Types Details
What it is. IDC's agent types are categories of how agents are developed and deployed, not product names. IDC defines an AI agent as an LLM-powered autonomous software entity that perceives its environment, reasons, makes decisions, takes actions, and interacts with users or other systems. IDC highlights three essential characteristics: autonomy, context awareness and tool use, and memory and learning.
How it works. In-app agents are embedded inside a provider's application, and standalone agents run independently of any single application for a specific purpose. Custom-configured agents are built with low-code or no-code agentic AI platforms, and bespoke agents are built with pro-code agentic AI frameworks and platforms. Organizations can build custom agents themselves or contract with a service provider to build them.
Why it matters. IDC says the types differ in flexibility, control, and who operates the agent, which affects how organizations build, buy, and govern agents. IDC's FutureScape for enterprise applications predicts that by 2030, 82% of enterprise applications will provide access to vendor-supplied or customer-configured agents that operate within or alongside the application.
How does IDC define an AI agent?
IDC defines an AI agent as an LLM-powered autonomous software entity that perceives its environment, reasons, makes decisions, takes actions, and interacts with users or other systems. IDC says agents go beyond reactive assistants by enabling proactive, goal-directed behavior and multiagent collaboration within workflows and business processes.
Supporting points
- Autonomy means operating with minimal continuous human oversight to achieve goals.
- Context awareness and tool use means sensing the environment, processing contextual data, and using external tools and systems.
- Memory and learning means leveraging past context and feedback to improve actions over time.
Which agent types does an agent provider build and operate?
IDC says three agent types are built and operated by an agent provider: in-app agents, standalone agents, and on-device agents. IDC tracks the first two and does not currently track on-device agents.
Supporting points
- An in-app agent is embedded directly inside a provider's application to help that app's users complete tasks, answer questions, or automate workflows without leaving the interface.
- A standalone agent is ready-made for a specific purpose, such as deep research, meeting notes, or customer support, and runs independently of any single application.
- IDC says on-device agents will be included with individual physical devices and are still in the nascent stages of development.
Which agent types do organizations custom build?
IDC says two agent types are custom built to meet the specific requirements of an organization: custom-configured agents and bespoke agents. The organization can build them itself or contract with a service provider, and the two differ in the development tooling used.
Supporting points
- Custom-configured agents use low-code or no-code agentic AI platforms for mainstream business operations automation without heavy programming.
- IDC says low-code tools speed development but encode assumptions and constraints by leveraging prebuilt frameworks and libraries.
- Bespoke agents use pro-code frameworks and platforms that offer high flexibility, and IDC says they suit highly technical teams needing ultimate control, including for proprietary, high-security, or sovereign workloads.
How does IDC measure the agent types?
IDC measures the agent types according to agent actions taken, agent action token load, and agent token delivery costs. The taxonomy is a companion to IDC's agentic AI adoption forecasts by sector and geography.
Supporting points
- IDC's taxonomy presents a standardized view of the agentic AI marketplace so vendors can define target markets in terms of agent actions, token delivery loads, and token costs.
- IDC says on-device agents are not tracked currently.
What do IDC's 2027 predictions say about agents in enterprise applications?
IDC FutureScape: Agentic Evolution of Enterprise Apps 2027 Predictions (IDC #US54818826, October 2026) predicts that by 2030, 82% of enterprise applications will provide access to vendor-supplied or customer-configured agents that operate within or alongside the application.
Supporting points
- IDC predicts that by 2031 only 10% of G2000 enterprises will shift investments from typical enterprise application agents to cross-application agents, even as demand grows by 25%.
- IDC predicts that by 2029, rising agent complexity will lead 60% of G2000 enterprises to shift build-versus-buy decisions from individual agents to enterprise-built or vendor-provided agent development factories.
- IDC FutureScape: Human-Agent Workforce Strategy 2027 Predictions (IDC #US54822326) predicts that by 2028, 50% of G2000 organizations will deploy named, role-based AI agents and achieve 35% higher user adoption than generic bots.
- IDC FutureScape: Velocity of AI and Automation 2027 Predictions (IDC #US54977226) predicts that by 2028, packaged agents inside business applications will help 40% of SMBs close the capability gap with larger organizations in core commercial and administrative functions.
Supporting Evidence
"a comprehensive and standardized view of the agentic AI marketplace" — Rick Villars, Group Vice President, Worldwide Research, IDC (IDC #US54650726, June 2026)
IDC defines AI agents into four primary types: bespoke agents, custom-configured agents, in-app agents, and standalone agents. Source: IDC's Worldwide Agentic AI Adoption Guide Taxonomy, 2026 (IDC #US54650726, June 2026).
FAQ
What are the types of AI agents?
IDC defines four primary types: bespoke agents, custom-configured agents, in-app agents, and standalone agents, and says on-device agents will emerge as a fifth type.
What is an in-app agent?
IDC defines an in-app agent as an AI agent embedded directly inside a provider's application to help that app's users complete tasks, answer questions, or automate workflows without leaving the interface.
What is the difference between a custom-configured agent and a bespoke agent?
IDC says custom-configured agents are built with low-code or no-code agentic AI platforms, while bespoke agents are built with pro-code frameworks and platforms that offer high flexibility and customizability.
What is a standalone agent?
IDC defines a standalone agent as a ready-made AI agent built and operated by a provider that runs independently of any single application, typically for a specific purpose such as deep research or customer support.
Does IDC track on-device agents?
No. IDC says on-device agents are still in the nascent stages of development, will emerge in the next five years, and are not currently tracked.