AI agent builder platforms: what each one offers and how to choose one

An AI agent builder is the platform or toolkit a business uses to make AI agents: software that reads a request, works out the steps and acts in your other systems, such as turning an emailed order into a draft sales order. Microsoft, Google, OpenAI, Amazon, Anthropic, n8n and Zapier each offer one. Some let staff build an agent without code inside tools they already use, and others are code libraries and cloud services for developers. This guide compares what each vendor says its builder offers and shows how a smaller manufacturer or distributor can choose.

Derik Lawlis Founder, ThriveAI

What an AI agent builder does

An AI agent pairs a language model with instructions and tools. A language model is the kind of AI behind ChatGPT and Claude. A tool is an action the agent may call, such as searching a mailbox, reading a stock level or creating a draft order. An agent builder supplies the parts around the model:

• Instructions describe the job and its rules in writing.
• Connectors are prebuilt links to apps such as Outlook, SharePoint or an ERP, so the agent can read and write there.
• Triggers start a run, for example a new email, a schedule or a person asking in chat.
• Approvals pause the agent so a person can accept or reject an action before it happens.
• Logs record each run, the tools it called and what it changed.

Many builders also support MCP, the Model Context Protocol, an open standard for connecting an AI application to outside tools and data. Microsoft, OpenAI, Anthropic and AWS each list MCP support on the pages cited below. What is an MCP server? explains how it works.

The steps for building one agent, from choosing the job to running a pilot, are in How to build an AI agent. This page compares the platforms you would build it on.

The main AI agent builders in 2026

Each entry below reports what the vendor says its builder offers, from the vendor’s own pages as they stood on September 28, 2026. Vendors rename and retire these products often, so check the linked page before you commit.

Microsoft Copilot Studio

Microsoft calls Copilot Studio “a platform for building and managing agents” that you connect to your business data, create with natural language and publish “across the channels your teams and customers use” (Microsoft). The same page lists Model Context Protocol servers and “more than 1,400 external connectors,” and it names Microsoft Teams, SharePoint and Microsoft Copilot as places to deploy agents. Microsoft’s plan comparison includes on-premises data transfer for Power Platform connectors, the feature that reaches data on your own servers.

How Microsoft charges: agent use is counted in Copilot Credits. You buy credits as monthly packs for the whole organization, as a pre-purchase plan, or through a pay-as-you-go meter linked to an Azure subscription. Microsoft says usage of agents published to Microsoft 365 Copilot is included in the licence of anyone licensed for Microsoft 365 Copilot.

Google: Gemini Enterprise and its Agent Platform

Google offers a builder at two levels. The Gemini Enterprise app lets staff “build custom agents with our no-code Workflow Builder” and connects to Google Drive, Microsoft OneDrive, SharePoint, HubSpot, Jira and other sources. Its Business edition covers up to 300 seats. The Standard and Plus editions add controls such as customer-managed encryption keys and data residency, and they accept agents built with Google’s Agent Development Kit (ADK), a code library for developers.

For developers, Google describes the Gemini Enterprise Agent Platform, formerly Vertex AI, as “Google Cloud’s comprehensive platform for developers to build, scale, govern and optimize agents.” Google lists more than 200 Google and third-party models and tools on it, including Anthropic’s Claude models.

How Google charges: the Gemini Enterprise app is priced per seat per month. On the Agent Platform, Google’s pricing page bills an agent’s runtime by the compute and memory it uses, and bills model use separately.

OpenAI: the Agents SDK and workspace agents

OpenAI’s Agents SDK is a code library for Python and TypeScript. OpenAI says it has “a very small set of primitives,” meaning basic building blocks, and that it “comes with built-in tracing” so a developer can follow and debug each run.

Staff who do not write code can use workspace agents in ChatGPT Business and Enterprise workspaces. According to OpenAI’s help centre, you describe the job, the builder drafts a plan, and you add apps such as Google Drive, SharePoint and Slack, or your own MCP servers. By default, write actions in apps and connectors are set to Always ask during a run.

OpenAI’s Agent Builder is being retired. OpenAI announced its deprecation on June 3, 2026, scheduled the shutdown for November 30, 2026, and points users to the Agents SDK or ChatGPT workspace agents. ChatKit, which adds an embedded chat experience to your own product, remains available.

How OpenAI charges: an agent built with the SDK calls OpenAI’s API, which is priced per million tokens. A token is a small piece of text, often part of a word. Workspace agents run inside ChatGPT workspaces, and OpenAI prices ChatGPT Business per user per month.

Amazon Bedrock AgentCore

AWS calls AgentCore “the platform for production AI agents” and says you can start with LangChain, the OpenAI Agents SDK, the Claude Agent SDK, its own Strands SDK or your own framework, “and deploy with any model” (AWS). It is a set of services for developers, listed on the AgentCore pricing page. Runtime hosts the agent. Memory keeps what the agent should remember. Gateway connects agents to tools, models and other agents. Identity handles how agents sign in to other services. Policy keeps agents within the actions you allow, and Observability and Evaluations trace and score what agents do.

AWS’s region table lists Runtime, Memory, Gateway, Identity, Policy, Observability and Evaluations in its Canada (Central) Region. Where the model processes a request is a separate setting, which private AI for business covers.

How AWS charges: AgentCore has consumption-based pricing “with no upfront commitments or minimum fees,” and each service is billed on its own.

Anthropic: the Claude Agent SDK

Anthropic’s Claude Agent SDK is a code library for Python and TypeScript that gives developers “the same tools, agent loop, and context management that power Claude Code.” Its built-in tools read and edit files, run commands and search the web. It adds hooks that run your own code at set points in a run, subagents for focused subtasks, MCP connections, and permissions that control “which tools run automatically, which need approval.” Anthropic also offers Managed Agents, in which Anthropic hosts the agent loop and runs each session in its own cloud sandbox or in a sandbox on your infrastructure.

How Anthropic charges: the SDK signs in with a Claude API key, and its use falls under Anthropic’s Commercial Terms of Service. Model use is billed at Claude API rates, which Claude API pricing sets out.

n8n

n8n is a workflow automation tool with an AI Agent node, which n8n calls “the core of your AI Agent.” n8n says you can “Use any LLM or vector store” and “Add human-in-the-loop approval steps when your agent’s decisions need oversight” (n8n). A vector store is a database that finds passages by meaning, the kind of search that RAG uses. You can use n8n Cloud or host n8n yourself, and n8n publishes a self-hosted Community Edition on GitHub. ThriveAI vs n8n and Zapier compares n8n and Zapier in more detail.

How n8n charges: n8n Cloud plans are priced by workflow executions a month, “regardless of complexity.”

Zapier Agents

Zapier says its agents do work “across 9,000+ apps” using company knowledge you attach, and that you create them “with the help of Zapier Copilot” (Zapier). Its FAQ adds that the agents “can only take actions in the apps you’ve connected and use the triggers and actions you’ve set up.”

How Zapier charges: Agents plans are sized by activities a month. An activity is an action the agent takes, a web search or a lookup in attached knowledge. Zapier offers a free Agents plan.

AI agent builders side by side

The table sums up the entries above. It shows who does the building, where the agent runs and how each vendor charges, as each vendor described it on September 28, 2026.

How to choose an AI agent builder

Answer these questions for one job, such as entering orders that arrive by email, before you compare prices.

1. Where does the work already happen? Start with the builder closest to the data the job needs. Copilot Studio sits beside Microsoft 365, the Gemini Enterprise app beside Google Workspace, and workspace agents beside ChatGPT Business.
2. Who builds it, and who keeps it running? A no-code builder suits a staff member who owns the process. The SDKs, AgentCore and the Agent Platform need a developer for the build and again whenever a model or a connected app changes.
3. Can it reach your ERP? Check the connector for your ERP by name, and ask how the builder reaches a server in your building. If the ERP has no API, an agent may have to work its screens, as legacy ERP automation explains. AI for ERP covers the wider picture.
4. Where does a person approve? Check the default for write actions, and whether you can require approval for named actions such as sending an email or creating an order. Human in the loop covers which steps need a person.
5. What does it record? You need a log of every run, every tool call and every approval, kept where your team can search it.
6. Where does the data go? Ask where the builder stores data and where the model processes it. Private AI for business and sovereign AI cover the Canadian questions.
7. How does the bill grow? Seats, credits, activities, executions and tokens each grow differently as volume rises. Estimate one month of your real volume, as the cost of AI shows.
8. Does it pass a test on your own work? Run the agent on real past jobs with known answers before anyone relies on it. AI evals explains how.

What a builder leaves to you

A builder supplies the parts. Your team or a partner still chooses the job, collects past examples with the right answers, connects the ERP, decides who approves what and tests every change. How to build an AI agent sets out those steps with a timeline and notes that the depth of the integration sets the timeline. AI agents for business lists jobs that suit an agent at a manufacturer or distributor, and RPA vs AI agents covers where older screen automation still fits.

How ThriveAI helps

ThriveAI is an AI engineering company in Ottawa. It builds private AI systems on the client’s own data for manufacturers and distributors in Ontario and Quebec. Derik Lawlis, the founder, leads every project and stays close to the build.

The platform is designed to keep each client’s data on its own server in Canada. You choose the model: one that runs on that server, or a hosted model under a written zero data retention agreement, under which the provider keeps no copy of a request or its answer. A hosted model may process requests outside Canada, so the contract names the model. A named person at your company approves every action before anything is sent or saved. ThriveAI vs alternatives compares ThriveAI with building it yourself and with other options, and About ThriveAI covers the company.