Scope note: “agent” in this glossary means AI software that runs a task. It does not mean a real estate or construction agent.
- AI agent builder
- A product that turns a task into an AI agent that runs it, using outside tools. This award covers hosted and no-code builders.
- AI agent
- Software that works through a task in steps. It can look up data, use tools and wait for approval before outside actions.
- Agent framework
- A code library a developer installs in their own project to build agents. A framework has no setup screen unless a separate platform is offered.
- Tool access
- The outside services an agent can call, such as search, social platforms, sales data and media generation. The Institute counts only tools named on a vendor’s public pages.
- Integration
- A connection between the builder and an outside service. The integration list is the set of connections the vendor names publicly.
- MCP (Model Context Protocol)
- An open standard for connecting AI assistants to outside tools and data. Some builders expose an MCP server so that other assistants can use their tools.
- Human-in-the-loop approval
- A step where the agent pauses and waits for a person to approve, edit or reject an action before it runs.
- Trace
- A record of what a run did, step by step, including the tools called and the results returned.
- Run history
- A list of past runs, with their status, inputs and outputs, that a user can open.
- Credits
- A unit of usage that a builder charges for. Some products use one balance across all tools, and others charge tools separately.
- Go-to-market (GTM) work
- Tasks that support selling a product, such as market research, lead lists, competitor review, content and outreach.
- Documentation review
- A review that scores only what a vendor publishes. It is not a test of the product and does not measure how well the product works.
Published 2026‑10‑11 · Panel review of public documentation, October 2026