Best AI agents and automation tools
This list is for people wiring models into jobs that already have steps: crawl a site, wait on a queue, call an API, then write back. It sits next to the broader AI agent list, but the emphasis is automation: LangGraph, n8n, Make, AutoGen, browser runtimes, and scrapers such as Apify.
What is AI agent automation?
AI agent automation is the combination of a model that can choose steps with a system that can run those steps reliably. The model proposes an action. The automation layer retries, waits, and records what happened. Without the second half, agents stall on timeouts and half-finished tasks. Without the first half, you only have a fixed zap.
This shortlist includes LangGraph, n8n, Make, CrewAI Studio, AutoGen, Flowise AI, and 16 more.
Who this shortlist is for
- Ops and GTM teams who already live in n8n or Make and want a model in the loop
- Engineers who need graphs, queues, or browsers, not only a chat SDK
- People comparing scrapers and hosted browsers as tools for agents
How we picked
Products land here if they run workflows, browsers, or long-running jobs that an agent can call. Chat-only LLM apps stay on the chat list. Support-only agents stay on the support-agent list.
- Exposes workflows, graphs, queues, or a browser or scraper API
- Can be triggered by an event, not only a person typing in a box
- Has a public product page we can attach to a structured profile
- Does not duplicate the support-agent list without an automation angle
The shortlist
Open a profile for pricing, alternatives, and a product page.

Visual automation platform for complex multi-app scenarios and data transforms.
Alternative to n8n

How to choose an agent automation stack
Separate the planner from the runner. Many teams keep the model in one product and the retries in another.
| If this is the job | Look at |
|---|---|
| Long-running or flaky jobs | Temporal or Windmill for retries and state. Do not keep that logic in the prompt. |
| Non-technical builders | n8n, Make, Lindy, or Dify. Check export and who can edit live flows. |
| The agent needs a real browser | Browserbase, Browserless, or a scraper such as Apify. Read the agentic-browser risk notes first. |
| Multi-agent research code | AutoGen, CrewAI Studio, LangGraph, or LangFlow if you will own the graph. |
Related research
Best AI agent tools
Frameworks, coding agents, and hosted builders for multi-step AI work.
Best developer tools
Editors, repos, cloud, APIs, and observability for engineering teams.
AI agent software
Full category directory, not this editorial shortlist.
MCP tool poisoning risk
Bookmarkit research note
Agent security checklist
Bookmarkit research note
Frequently asked questions
What is the difference between this list and the AI agents list?
The AI agents list covers frameworks, coding agents, and general builders. This page focuses on products you use when the agent must run a workflow, a browser, a scraper, or a durable job. If your question is “which library should I import?”, start on the agents list. If your question is “how does this keep running at 2am?”, start here.
Can n8n or Make replace an agent framework?
They can replace a framework for narrow jobs: watch a trigger, call a model, write to a sheet. They do not replace a framework when you need custom tools, evals, and a graph you test in CI. Many teams use both: n8n for glue, LangGraph or AutoGen for the planner.
Why are scrapers and browsers on an automation list?
Agents that research the public web need a fetch layer that can render pages, rotate sessions, and store results. Apify, Bright Data, Oxylabs, Browserbase, and Browserless are that layer. They are not agents. They are tools an agent calls, and they carry legal and security constraints.
What is LangGraph used for?
LangGraph is a way to define agent steps as a graph you can test, interrupt, and resume. Teams pick it when a linear chain is not enough and they want checkpoints. It is a developer tool, not a no-code studio.
How do I keep automation agents from looping forever?
Set a max step count, a wall-clock timeout, and a human approval on writes. Log every tool call. If you use a browser, isolate cookies and credentials from the model prompt. Durable runners such as Temporal help you fail a job instead of retrying blindly.



































