AI agents, explained: 19 tools from the directory
An AI agent plans and carries out a multi-step task on its own, instead of answering one prompt at a time. Below: what that means, what to check before you hand one real work, and the 19 tools in our directory whose own descriptions say they are agents or agentic.
Listed by type, not ranked. These are research-based profiles, not hands-on tests. Descriptions follow what each vendor says, and prices are as listed in the directory, so confirm on the vendor's site before you buy.
Agent or assistant?
An assistant
Answers when you ask, one step at a time. You stay in the loop for every step and decide what happens next.
An agent
Is given a goal, picks its own steps, may use tools such as browsing, running code, editing files or calling other apps, and checks its own progress. The industry has not settled on one exact definition.
Terms in our glossary: AI agent, tool use, computer use, MCP, multi-agent system and orchestration.
Where are agents heading? See AI agents: state and outlook, a dated snapshot of what agents can do today, what is still shaky and what has been reported lately.
Skills, tools and MCP: how an agent gets new abilities
Three words come up often when people talk about what an agent can do. They are related but not the same thing.
Tools
The actions an agent can take: searching the web, running code, editing files, calling another app. Giving a model these actions is called tool use.
MCP
An open standard for connecting an agent to tools and data, so one connection can work across different apps. See MCP in the glossary.
Skills
Reusable folders of instructions, scripts and reference files that an agent loads only when a task calls for them. See agent skills in the glossary.
In short: a tool or an MCP connection gives an agent access to something, and a skill tells it how to do a kind of task. Anthropic describes it the same way, saying connections give Claude access to data and skills teach it what to do with that data.
A skill is a folder with a file called SKILL.md (a name, a description of when to use it, and the instructions), plus optional scripts and reference files. The agent reads only each skill's name and description at first and opens the rest when a task matches. Anthropic introduced the format in late 2025 and published it as an open standard, described at agentskills.io. Anthropic documents skills for Claude, and GitHub and Visual Studio Code publish pages about skills in GitHub Copilot and in VS Code. Support differs from tool to tool, so check your tool's own page.
Before you add a skill or a connection
- Treat it like any program you download. A skill can include scripts that run on your machine. Prefer sources you trust and read what it does first. This is our advice, not a vendor rule.
- Check what it reaches. A connection can expose files, mail or accounts. Give it only what the task needs.
- Check that your tool supports it. Names, folders and extra features differ between products.
- Remember instructions inside files can be wrong or hostile. Text in a document or web page can try to steer an agent; see prompt injection.
Want to learn how these work? Our Learn AI guide lists courses and guides, many of them free, including ones on agents, MCP and agent skills.
Before you hand one real work
- What can it touch? Check which accounts, files or payments it can reach, and whether you can limit that.
- What does a run cost? Many agents charge by credits, usage or executions, so a long task can cost more than the plan price suggests.
- Can you see what it did? Look for a log or step-by-step history you can read afterwards.
- Who reviews the result? Agents can be wrong. Keep a person checking anything that is sent, paid for or published.
- Start small. Try a low-risk task first, on the free plan if there is one.
Agents that do tasks for you3 tools
You give a goal and the agent plans the steps. Some work in a browser, others operate a virtual desktop.
- Manus AIAI agent that plans and carries out multi-step tasks such as research, slides and website builds. Credit-based plans with a free tier; paid plans start at $20 a month.$20/mo
- GensparkAI workspace with a Super Agent for multi-step tasks, plus AI Slides, Sheets, Docs, chat and image tools. Free plan with daily credits; Plus is $19.99 a month billed annually ($24.99 monthly).Free / $19.99+
- Simular AISai is a computer-use agent that runs on a Mac, a cloud Windows or Mac computer, or your own device; no free plan, paid plans from $50 a month, billed monthly.$50/mo
Coding agents5 tools
Agents for writing, changing and shipping code.
- DevinCognition's AI software engineer that plans, codes, tests and ships. Free plan with limited usage; Pro is $20 a month, with usage quotas plus pay-as-you-go credits.Free / $20+
- GitHub CopilotGitHub's AI coding assistant: completions, chat, code review and agents in VS Code, JetBrains, Visual Studio, Xcode and on GitHub. Free plan; Pro is $10 a month.Free / $10+
- Replit AgentAI agent in Replit that builds, tests and publishes apps from plain-language prompts. Free Starter plan; Core is $18 a month billed annually ($20 monthly).Free / $20+
- Amazon Q DeveloperAWS coding assistant with chat, suggestions and Java/.NET upgrades. AWS has blocked new sign-ups since 15 May 2026 and ends IDE plugin support on 30 April 2027, pointing to Kiro.$19/mo
- WindsurfAI coding editor from Cognition, renamed Devin Desktop in June 2026 per the vendor. Runs local and cloud agents in a VS Code-compatible IDE. Free plan; Pro is $20 a month.Free / $20+
Build your own agents and workflows10 tools
Platforms for building agents or multi-step automations. Several price by workflow runs, executions or credits.
- Lindy AIAI teammate that connects to a team's tools and does tasks like inbox, CRM and reporting work. Credit based, with $50 of free credits for 7 days; the pricing page lists Team at $49.99 a month.$49.99/mo
- Relevance AINo-code platform for building AI agents and multi-agent workforces. Pro is $19 a month billed annually ($29 monthly); the docs say the free plan is retired.$19+/mo
- Zapier AgentsNo-code AI agents that work across 9,000+ apps in Zapier. Free plan of 400 activities a month; Pro is $400 a year (about $33 a month).Free / $33+
- GumloopNo-code builder for AI agents and workflows, with MCP support. Pro starts at $37 a month with a 14-day trial; no free plan is listed on its pricing page.$37+/mo
- Beam AIPlatform for building and running AI agents that automate business processes. Free is 20 tasks a month; Pro is $50 a month, billed monthly.Free / $50+
- n8nWorkflow automation and AI-agent platform. Self-host the Community Edition or use n8n Cloud from €20/mo billed annually; billing counts workflow executions.From €20/mo
- CrewAIOpen-source (MIT) Python framework for multi-agent workflows, plus the hosted CrewAI AMP platform with a visual Studio editor. The free tier allows 50 executions a month; Enterprise is custom.Free / Custom
- DifyPlatform for LLM apps and agentic workflows with visual workflows and RAG, as hosted Cloud or self-hosted. Free Sandbox plan; Professional is $590 per workspace a year ($59 monthly).Free / $49.17+
- ActivepiecesOpen-source workflow automation and AI agent platform, cloud or self-hosted. Free includes 1,000 credits; Plus is $20 a month (annual price not listed).Free / $20+
- AutoGPTHosted platform for hiring AI Experts for recurring work, plus free self-hosting. Pro is $42.50 a month billed annually ($50 monthly).$42.50/mo
Voice agents1 tools
Agents that make and receive phone calls.
- VapiDeveloper platform for voice AI phone agents. Pay as you go at $0.05 per minute hosting plus provider costs, no free plan; optional Core package is $29 a month (billing basis not stated).$29/mo
No agent tools match that search.
More: the Autonomous Agents category, a head-to-head of n8n and Zapier Agents, or the Tool Battle to compare any two.
