--- title: "Agentic workflows (2026): how to set them up" description: "Agentic workflows in 2026: how to set them up, with Zapier Agents as the top pick of tools graded 0 to 3 by depth, plan and price, plus steps for every tool." canonical: "https://thenotetakers.com/agentic-workflows" last_updated: "2026-11-06" --- # Agentic workflows (2026): how to set them up **To set up an agentic workflow, start with Zapier Agents. Tools here are graded 0 to 3 for depth, then shelved by plan and price, and it is grade 3 on Free at $0 (400 activities a month): 9,000+ apps, schedule and app triggers, agents that call other agents, and a view of each run. The steps below apply to every tool on this page.** ## Agentic workflows: Zapier Agents is the top pick Zapier Agents is grade 3 for agentic workflows, on Free at $0 (400 activities a month; [Zapier Agents pricing](https://zapier.com/l/agents-pricing), checked Nov 2026). Zapier is moving Agents to AI by Zapier, and says "Zapier has not set a date to turn off Agents yet." n8n also grades 3 and sits second because it starts on a paid plan, Starter at €20/month billed annually. CrewAI grades 3 and sits third because its app connectors are shelved on Enterprise, which has no published price. LangGraph, AutoGen, the OpenAI Agents SDK and the Claude Agent SDK are code you write, so they shelve lower. ## How the tools are graded The scale below is for this page. A grade counts what the vendor's own pages say ships in the product. | Grade | Meaning | |---|---| | 0 | Code only: a library or framework. You write the app connections, triggers and hosting | | 1 | Hosted runs start on a schedule or API call, with no prebuilt app actions | | 2 | Prebuilt app actions plus a trigger | | 3 | Grade 2, plus one agent handing work to another, plus logs or traces for each run | Code you write yourself is printed beside the grade and does not move the entry. "2 in code" means the library gives you tools and multi-agent patterns, with no trigger documented. "3 in code" adds a documented trigger and traces. A self-hosted edition or open-source library is printed beside the plan and does not move the entry either. Entries are shelved by depth grade, then plan required (Free, then entry paid, then higher, then sales-only), then the lowest list price for that plan. A published price shelves ahead of no published price. Ties after that are alphabetical. A feature the vendor's pages do not document does not raise a grade, and it is printed as Not documented, not as missing. It means the pages listed under Sources do not state it. ## Agentic workflow tools compared Shelved by depth grade, then plan required, then lowest list price, with ties listed alphabetically. Last checked: Nov 6, 2026. | Rank | Tool | Depth grade | Plan required | Lowest list price | How you define it | Tools and connectors | Multi-agent or handoff | Triggers and scheduling | Observability | Last checked | |---|---|---|---|---|---|---|---|---|---|---| | 1 | Zapier Agents | 3 | Free | $0, 400 activities a month | Agent created with help from Zapier Copilot | 9,000+ apps | Call an agent tool, fewer than 5 calls per main agent advised | On Demand, Schedule by Zapier, Zap workflows and MCP, app triggers | Activity view with expandable runs; Observability API and Log Streams are Enterprise only | Nov 6, 2026 | | 2 | n8n | 3 | Starter (self-hosted version on GitHub) | €20/month, billed annually | AI Agent node inside a workflow | 600+ native integrations, your workflows, MCP | Workflows as tools; Message an Agent node | Webhooks, queues, scheduling | Step inputs and outputs, replay, execution logs (7 days on Starter) | Nov 6, 2026 | | 3 | CrewAI | 3 | Enterprise for connectors (Basic is Free, 50 workflow executions/month; connectors on Basic Not documented) | Custom, no published price | Agents, crews and flows | OAuth app connectors: Gmail, Slack, HubSpot and more | Sequential, hierarchical or hybrid processes | App triggers, cron scheduling | Tracing, OpenTelemetry | Nov 6, 2026 | | 4 | LangGraph | 1 (3 in code) | Plus for hosted runs (MIT library is free) | $39/seat/month, then pay as you go | Graph of state, nodes and edges in code | In your code | Single, multi-agent, hierarchical | Cron and API trigger on LangSmith Deployment | LangSmith traces | Nov 6, 2026 | | 5 | AutoGen | 0 (2 in code) | None, MIT, archived | $0 | Python agents | McpWorkbench extension | Two-agent chat, group chats | Not documented | Not documented | Nov 6, 2026 | | 6 | OpenAI Agents SDK | 0 (2 in code) | None, MIT | $0 for the library; model prices Not documented | Python agent with instructions and tools | Python functions, remote MCP | Handoffs | Not documented | Tracing on by default | Nov 6, 2026 | | 7 | Claude Agent SDK | 0 (2 in code) | API key | Token rates: Opus 5.5 $4 input and $20 output per million tokens; Haiku 5.5 $0.10 and $0.50 on prompts up to 100K | Python or TypeScript | Built-in tools, MCP | Subagents | Not documented | Not documented | Nov 6, 2026 | ## How to set up an agentic workflow 1. **Pick one job with a clear finish.** "Answer each new helpdesk question" works. "Handle support" does not. 2. **Choose what starts a run.** n8n triggers from webhooks and queues. CrewAI triggers listen for new emails, calendar updates and CRM status changes. Zapier Agents can run on demand, on a schedule, or from an app event. LangSmith runs cron jobs, and "All cron schedules are interpreted in UTC." 3. **Choose a tool from the table.** Grade tells you how far it reaches as shipped. Plan required tells you what that reach costs. If you need one agent and no trigger yet, see [how to build an AI agent](/how-to-build-an-ai-agent). For one more builder outside this table, see the [Lindy AI review](/lindy-ai-review). 4. **Define the agent.** In n8n, you "connect a chat model and one or more tools". In LangGraph, you define "graphs of shared state, nodes, and edges". The OpenAI Agents SDK defines agents as "LLMs equipped with instructions and tools". 5. **Connect its tools.** n8n states: "You must connect at least one tool sub-node to an AI Agent node." CrewAI connects apps such as Gmail and HubSpot over OAuth. In Zapier Agents, you add them under "Tools this agent can use". The Claude Agent SDK and the OpenAI Agents SDK reach outside tools through MCP. 6. **Add a second agent only when the job needs it.** The options are the Zapier Call an agent tool, OpenAI handoffs, Claude subagents, CrewAI hierarchical processes and the n8n Message an Agent node. 7. **Add approvals before anything writes.** n8n has human approval for tool calls. LangGraph has human-in-the-loop checks. The Claude Agent SDK controls which tools "need approval". OpenAI has guardrails for agent inputs and outputs. 8. **Watch the runs and cap the spend.** n8n lets you replay any run. LangSmith and the OpenAI Agents SDK record traces. Zapier Agents shows each run under Activity, and Free stops at 10 activities per run. ## 1. Zapier Agents Zapier Agents is grade 3 for starting free across 9,000+ apps, on Free at $0 (400 activities a month; [Zapier Agents pricing](https://zapier.com/l/agents-pricing), checked Nov 2026). - **Grade:** 3. It has prebuilt app actions, several triggers, a Call an agent tool and an Activity view of each run. Neither the agent call page nor the activity page names a plan requirement, so the entry shelves on Free. - **Plan required:** Free. - **Price:** Free is $0 for 400 activities a month. Pro starts at about $33.33/month billed annually, with 1,500 activities a month. - **Define:** you create agents with help from Zapier Copilot and add company knowledge. - **Connects to:** 9,000+ apps. - **Multi-agent:** "you can ask your agent to call another agent." You add Call an agent as a tool, and Zapier advises "fewer than 5" calls per main agent. - **Triggers:** On Demand, Schedule by Zapier, Zap workflows and MCP, and app triggers such as Gmail, Google Sheets or Slack. - **Observability:** click Activity on the agent's page, then "Expand individual runs to see exactly which actions consumed activities." Zapier lists its Observability API and Log Streams for Enterprise only. - **Best for:** starting free across a large app catalog. - **Limitation:** Zapier is moving Agents to AI by Zapier, and "Zapier has not set a date to turn off Agents yet." A migrated agent becomes a Zap with one AI by Zapier step. Free allows 10 activities per run, and test runs count toward the monthly allowance. - **Last checked:** Nov 6, 2026. ## 2. n8n n8n is grade 3 for agents inside scheduled workflows, on Starter at €20/month billed annually (2.5K workflow executions; [n8n pricing](https://n8n.io/pricing/), checked Nov 2026). - **Grade:** 3. Agents get prebuilt integrations as tools, scheduling is built in, and run logs ship with the product. An agent calls a workflow as a tool, and a workflow can call a published agent with the Message an Agent node. - **Plan required:** Starter, which lists agents, MCP and AI nodes. A standard self-hosted version is on GitHub. - **Price:** Starter €20/month and Pro €50/month, both billed annually. "One turn with an agent is one execution." - **Define:** the AI Agent node is "one step inside of a workflow that runs to completion". See the [n8n review](/n8n-review) for the node in more depth. - **Connects to:** "600+ native integrations and your own workflows as tools", plus MCP Client and MCP Server Trigger. - **Multi-agent:** workflows as tools, plus the Message an Agent node. - **Triggers:** webhooks and queues on Starter, plus built-in scheduling. - **Observability:** "Open any step to see the inputs and outputs." You can also replay any run. - **Best for:** teams that want agents inside scheduled workflows, with app connections ready. - **Limitation:** Starter caps runs at 5 minutes and keeps execution logs for 7 days. The license is fair-code (Sustainable Use License and n8n Enterprise License). Asked whether external systems can call its autonomous agents, n8n answers "Not at the moment." - **Last checked:** Nov 6, 2026. ## 3. CrewAI CrewAI is grade 3 for role-based crews with app triggers, on Enterprise at custom pricing ([CrewAI pricing](https://www.crewai.com/pricing), checked Nov 2026). - **Grade:** 3. It has OAuth app connectors, app triggers and cron, sequential or hierarchical processes, and tracing. - **Plan required:** Enterprise for connectors. The pricing page names the row "Enterprise connectors", integrations in code need an Enterprise Token, the docs put app connections under "Enterprise journey", and no page cited lists connectors on Basic. - **Price:** Basic is Free, with 50 workflow executions/month. Enterprise is custom, with no published price. The framework is MIT licensed. - **Define:** "agents with tools, memory, knowledge, and structured outputs using Pydantic", arranged in crews and flows. - **Connects to:** Gmail, Google Drive, HubSpot, Slack, Salesforce, Jira, Notion, Linear, GitHub and more. - **Multi-agent:** "sequential, hierarchical, or hybrid processes with guardrails". - **Triggers:** triggers "listen for changes" in apps such as Gmail, HubSpot and Salesforce. Cron scheduling and standard triggers are on Basic and Enterprise. - **Observability:** tracing and OpenTelemetry, on Basic and Enterprise. - **Best for:** Python teams that want role-based crews with app triggers. - **Limitation:** the plan that names connectors has no published price, and no page cited lists connectors on Basic. - **Last checked:** Nov 6, 2026. ## 4. LangGraph LangGraph is grade 1 (3 in code) for code-level control of state, with hosted runs on Plus at $39/seat/month, then pay as you go ([LangChain pricing](https://www.langchain.com/pricing), checked Nov 2026). - **Grade:** 1. LangSmith Deployment runs cron jobs and API triggers. Prebuilt app actions are Not documented on the pages cited, so connections count as your code. Multi-agent patterns and traces take it to 3 in code. - **Plan required:** Plus for hosted runs. The library itself is MIT licensed and free to use. - **Price:** Plus is $39/seat/month, then pay as you go, with 10k base traces a month. Developer is $0/seat/month. - **Define:** "graphs of shared state, nodes, and edges" in code. - **Connects to:** whatever your code connects to. - **Multi-agent:** "single, multi-agent, hierarchical" control flows. - **Triggers:** cron and API trigger on LangSmith Deployment. Cron schedules run in UTC. - **Observability:** LangSmith tracing. "Base traces have a shorter retention period of 14 days." - **Best for:** developers who want control of state and the option to resume after failures or pauses. - **Limitation:** you write the app connections, and Plus adds usage charges on top of the seat price. - **Last checked:** Nov 6, 2026. ## 5. AutoGen AutoGen is grade 0 (2 in code) for existing AutoGen projects, at $0 under the MIT license ([AutoGen on GitHub](https://github.com/microsoft/autogen), checked Nov 2026). - **Grade:** 0. It is a programming framework that requires Python 3.10+. - **Plan required:** none. - **Price:** $0, MIT license. - **Define:** conversational single and multi-agent applications in Python. - **Connects to:** the McpWorkbench extension. - **Multi-agent:** "two-agent chat or group chats". - **Triggers:** Not documented on the pages cited. - **Observability:** Not documented on the pages cited. - **Best for:** existing AutoGen projects. - **Limitation:** the project is archived. Microsoft provides no bug fixes or security fixes, and says "New users should start with Microsoft Agent Framework." - **Last checked:** Nov 6, 2026. ## 6. OpenAI Agents SDK The OpenAI Agents SDK is grade 0 (2 in code) for Python handoffs, at $0 under the MIT license ([Agents SDK on GitHub](https://github.com/openai/openai-agents-python), checked Nov 2026). - **Grade:** 0. Prebuilt app actions and triggers are Not documented on the pages cited. Function tools, MCP and handoffs are code you write. - **Plan required:** none. - **Price:** $0 for the library, MIT license. Model prices are Not documented on the SDK pages. - **Define:** "LLMs equipped with instructions and tools". - **Connects to:** any Python function as a tool, plus remote MCP tools. - **Multi-agent:** "Handoffs allow an agent to delegate tasks to another agent." - **Triggers:** Not documented on the pages cited. - **Observability:** "Tracing is enabled by default." - **Best for:** Python developers who want handoffs, with tracing on by default. - **Limitation:** "Handoffs stay within a single run." Tracing is unavailable to organizations under a Zero Data Retention policy. - **Last checked:** Nov 6, 2026. ## 7. Claude Agent SDK The Claude Agent SDK is grade 0 (2 in code) for agents that read and edit files and run commands, on an API key at token rates from Haiku 5.5 at $0.10 input to Opus 5.5 at $4 input per million tokens ([Claude pricing](https://claude.com/pricing), checked Nov 2026). - **Grade:** 0. It is "Claude Code as a library", with no prebuilt app actions or triggers documented on the pages cited. - **Plan required:** API key authentication. - **Price:** per million tokens, Opus 5.5 is $4 input and $20 output, Sonnet 5.5 is $2 and $10, and Haiku 5.5 is $0.10 and $0.50 on prompts up to 100K. An Agent SDK price is Not documented on that page. - **Define:** code in Python or TypeScript. - **Connects to:** built-in tools that read, write and edit files, run commands and search the web, plus MCP. - **Multi-agent:** subagents, by default up to 3 layers deep and 20 running at once. - **Triggers:** Not documented on the pages cited. - **Observability:** Not documented on the pages cited. - **Best for:** agents that read and edit files and run commands, in Python or TypeScript. - **Limitation:** "Each subagent makes its own API requests", and those count toward the query's cost. - **Last checked:** Nov 6, 2026. ## Best for - Best for most teams: Zapier Agents (grade 3, Free, $0) - Best for agents inside scheduled workflows: n8n (grade 3, Starter, €20/month billed annually) - Best for role-based crews in Python: CrewAI (grade 3, Enterprise, custom pricing) - Best for code-level control: LangGraph (grade 1, 3 in code, Plus, $39/seat/month) - Best for Python handoffs: OpenAI Agents SDK (grade 0, 2 in code, MIT license) - Best for file and command agents: Claude Agent SDK (grade 0, 2 in code, API token rates) ## FAQ ### What is an agentic workflow? It is a set of steps in which a model decides which tools to call to finish a task. n8n describes its agent node the same way: "the agent decides which tools to call to complete a task." ### How do I set up an agentic workflow? Pick one job, choose what starts a run, choose a tool, define the agent, connect its tools, and add approvals before anything writes. Then watch the runs and cap the spend. ### What is the best tool for agentic workflows? Zapier Agents is the pick on this page, at grade 3 on Free for $0 with 400 activities a month. n8n is also grade 3, on Starter at €20/month billed annually. ### What is the difference between an agentic workflow and an AI agent? The agent is the part that decides, and the workflow is what surrounds it. n8n calls its AI Agent node "one step inside of a workflow that runs to completion", while its autonomous agents have "their own addressable identity, and persistent memory." ### Do I need code to build an agentic workflow? Not always: Zapier Agents are created with help from Zapier Copilot, and CrewAI Basic lists a visual editor and AI copilot. LangGraph, AutoGen, the OpenAI Agents SDK and the Claude Agent SDK are code you write. ### Can I run agentic workflows for free? Zapier Agents Free gives 400 activities a month, and CrewAI Basic gives 50 workflow executions a month. The LangGraph, AutoGen and OpenAI Agents SDK libraries are MIT licensed, but those licenses do not cover model prices. ## Sources All pages checked Nov 6, 2026. - n8n: [pricing](https://n8n.io/pricing/), [AI agents](https://n8n.io/ai-agents/), [AI Agent node](https://docs.n8n.io/integrations/builtin/cluster-nodes/root-nodes/n8n-nodes-langchain.agent/), [GitHub](https://github.com/n8n-io/n8n) - CrewAI: [pricing](https://www.crewai.com/pricing), [introduction](https://docs.crewai.com/en/introduction), [tools and integrations](https://docs-platform.crewai.com/platform/en/features/tools-and-integrations), [automation triggers](https://docs-platform.crewai.com/platform/en/guides/automation-triggers), [GitHub](https://github.com/crewAIInc/crewAI) - Zapier: [Agents](https://zapier.com/agents), [Agents pricing](https://zapier.com/l/agents-pricing), [how usage is measured](https://help.zapier.com/hc/en-us/articles/26559132765325), [set up your agent's trigger](https://help.zapier.com/hc/en-us/articles/45394909914381-Set-up-your-agent-s-trigger), [pricing](https://zapier.com/pricing), [call an agent from another agent](https://help.zapier.com/hc/en-us/articles/37902635257101-Call-an-agent-from-another-agent), [unexpected activity usage](https://help.zapier.com/hc/en-us/articles/45697420326285-Zapier-Agents-unexpected-activity-usage), [migrating from Agents to AI by Zapier](https://help.zapier.com/hc/en-us/articles/47402591569805-Migrating-from-Agents-to-AI-by-Zapier) - LangGraph and LangSmith: [LangGraph](https://www.langchain.com/langgraph), [overview](https://docs.langchain.com/oss/python/langgraph/overview), [pricing](https://www.langchain.com/pricing), [cron jobs](https://docs.langchain.com/langsmith/cron-jobs) - AutoGen: [docs](https://microsoft.github.io/autogen/stable/), [GitHub](https://github.com/microsoft/autogen) - OpenAI Agents SDK: [docs](https://openai.github.io/openai-agents-python/), [handoffs](https://openai.github.io/openai-agents-python/handoffs/), [tracing](https://openai.github.io/openai-agents-python/tracing/), [GitHub](https://github.com/openai/openai-agents-python) - Claude Agent SDK: [overview](https://code.claude.com/docs/en/agent-sdk/overview), [subagents](https://code.claude.com/docs/en/agent-sdk/subagents), [pricing](https://claude.com/pricing)