Skip to content
AI Tools · AI Agent Builder

Top 5 AI Agent Builders of 2026: Copilot Studio vs Zapier vs n8n vs Gemini Enterprise vs Lindy

No-code and low-code platforms that let business teams build AI agents, judged on what the agent may touch, who approves its actions, and what gets logged.

By ·Nov 15, 2026·13 min·5 tools compared
AI AgentsNo-CodeCopilot StudioZapiern8nAgent Security

Quick Comparison

Built forStarting price (as of Nov 2026)Human approval stepAudit logs and SSOBest For
Microsoft 365 organisationsCopilot Credit packs at $200/month for 25,000 credits, or pay-as-you-go; Microsoft 365 Copilot is $30/user/monthRequest for information action pauses an agent flow until a named reviewer answers in OutlookMaker audit logs in Microsoft Purview, alerts in Sentinel, Entra sign-in; admin data policies per connectorRegulated companies already on Microsoft 365
Small and mid-size teams with existing ZapsFree plan with 100 tasks/month; Pro from $19.99/month billed annuallyPer-tool Require approval before running toggle (off by default)SAML SSO and audit log on Team (6 months history) and Enterprise (1 year)Adding an agent step to automations you already run
Technical operations teamsSelf-hosted Community Edition from GitHub; Cloud Starter from 20 euros/month billed annuallyHuman review on selected AI Agent tool calls, via Chat, Slack, Teams, Gmail and moreSSO from Business; audit logging and log streaming on EnterpriseTeams that want to self-host and inspect every step
Google Workspace organisationsBusiness edition from $21/seat/month, 30-day trialWorkflow Builder Approval and Request info steps; paused runs show Needs ReviewCloud Audit Logs, VPC Service Controls and CMEK listed for Gemini EnterpriseGoogle-first companies that want agents over company data
Individuals and small teamsFree with $50 of credits for 7 days; Team from $29.99/month per active userActions with outside impact wait for your approval by defaultSSO, SCIM and audit logs on Enterprise onlyInbox, calendar and meeting assistants without building a workflow

Microsoft 365 organisations

Starting price (as of Nov 2026)
Copilot Credit packs at $200/month for 25,000 credits, or pay-as-you-go; Microsoft 365 Copilot is $30/user/month
Human approval step
Request for information action pauses an agent flow until a named reviewer answers in Outlook
Audit logs and SSO
Maker audit logs in Microsoft Purview, alerts in Sentinel, Entra sign-in; admin data policies per connector
Best For
Regulated companies already on Microsoft 365

Small and mid-size teams with existing Zaps

Starting price (as of Nov 2026)
Free plan with 100 tasks/month; Pro from $19.99/month billed annually
Human approval step
Per-tool Require approval before running toggle (off by default)
Audit logs and SSO
SAML SSO and audit log on Team (6 months history) and Enterprise (1 year)
Best For
Adding an agent step to automations you already run

Technical operations teams

Starting price (as of Nov 2026)
Self-hosted Community Edition from GitHub; Cloud Starter from 20 euros/month billed annually
Human approval step
Human review on selected AI Agent tool calls, via Chat, Slack, Teams, Gmail and more
Audit logs and SSO
SSO from Business; audit logging and log streaming on Enterprise
Best For
Teams that want to self-host and inspect every step

Google Workspace organisations

Starting price (as of Nov 2026)
Business edition from $21/seat/month, 30-day trial
Human approval step
Workflow Builder Approval and Request info steps; paused runs show Needs Review
Audit logs and SSO
Cloud Audit Logs, VPC Service Controls and CMEK listed for Gemini Enterprise
Best For
Google-first companies that want agents over company data

Individuals and small teams

Starting price (as of Nov 2026)
Free with $50 of credits for 7 days; Team from $29.99/month per active user
Human approval step
Actions with outside impact wait for your approval by default
Audit logs and SSO
SSO, SCIM and audit logs on Enterprise only
Best For
Inbox, calendar and meeting assistants without building a workflow
1

Microsoft Copilot Studio

Best Overall

Best for: Organisations on Microsoft 365 that need agents to pass a security and compliance review

“Copilot Studio ranks first because its governance is the most complete of the five, and it is documented. Admins set data policies that decide which connectors, knowledge sources, HTTP calls and publishing channels a maker may use. Maker activity lands in Microsoft Purview audit logs, and Sentinel can alert on agent behaviour. The trade-off is cost and complexity: billing runs on Copilot Credits, and the full control set assumes you already live in Power Platform and Entra.”

Pros

  • Admin data policies in the Power Platform admin center govern authentication, knowledge sources, connectors, HTTP requests, triggers and channel publishing, per Microsoft's security and governance page
  • Maker audit logs are visible in Microsoft Purview, and Microsoft Sentinel can monitor and alert on agent activity
  • A Request for information action pauses an agent flow and waits for a human reviewer before later steps run
  • Through Microsoft Agent 365, agents can be represented as Microsoft Entra identities and governed with Conditional Access
  • A built-in security scan warns makers before publishing when default security settings were changed

Cons

  • Pricing is credit-based: a pack of 25,000 Copilot Credits is $200 per month, and an Azure subscription is required to use agents
  • Request for information goes through Outlook only and cannot be sent to users outside your tenant, per Microsoft's documentation
  • Most controls assume Power Platform environments, Purview and Entra are already set up and staffed
Honest Weakness: Copilot Studio is the safest choice only if someone owns the Microsoft admin side. The controls exist, but they live across the Power Platform admin center, Purview, Sentinel and Agent 365. A small company without a Microsoft 365 admin will pay for governance it never switches on. Microsoft also notes that its Customer Lockbox does not cover the security audit telemetry sent to Purview or Agent 365.

Governance You Can Show an Auditor

Microsoft's security and governance page for Copilot Studio, updated November 2026, lists the controls in one table. Data policies limit maker authentication, knowledge sources, connectors, HTTP requests, triggers and publication channels. Connector dependency insights show where data moves before deployment. Customer-managed encryption keys are available per environment. Maker audit logs flow to Purview, and Sentinel handles alerting. For a security team, that list is the difference between approving an agent programme and blocking it.

Where a Human Signs Off

Agent flows can include a Request for information step under Human review. The flow pauses, emails the assigned reviewers through Outlook, and resumes when the first reviewer submits. Inputs can be text, yes or no, email, number or date, and later steps can use the answers. Microsoft is explicit that only the first response counts, which matters when you assign a group of approvers.

Copilot Credit pack $200/month (25,000 credits) or pay-as-you-go; Microsoft 365 Copilot $30/user/month paid yearly

Visit Microsoft Copilot Studio
2

AI by Zapier (formerly Zapier Agents)

Best Value

Best for: Small and mid-size teams that already run Zaps and want an agent step with per-tool approvals

“Zapier is moving its standalone Agents product into AI by Zapier, a single AI step inside the Zap editor. That gives an agent the same triggers, filters, paths and run history as any other Zap. Each tool the agent can call has a Require approval before running toggle, so the agent can read freely while writes wait for a person. Billing moves to ordinary Zapier tasks, which is simpler to budget than a separate agent meter.”

Pros

  • Per-tool approval: switch on Require approval before running for tools that create, update or delete data, and leave read-only tools automatic
  • Each run has a task limit, 75 by default and adjustable up to 500, and a run that exceeds it pauses for review
  • Task usage is broken down by tool call, model tier and step in Zap history
  • A free plan exists with 100 tasks per month, and Pro starts at $19.99 per month billed annually for 750 tasks

Cons

  • Approval is off by default for every tool, so an unreviewed agent can write data until someone changes the setting
  • SAML SSO and the audit log are not in Pro; they start on Team, with 6 months of audit history, and Enterprise keeps 1 year
  • The migration from Agents is in progress: Zapier has not set a shutdown date for the old product, and its pricing page still lists the separate Agents add-on
Honest Weakness: Zapier's safety model depends on the builder flipping the right toggles. Nothing forces approval on a tool that sends email or deletes records. A team that hands agent building to many people should set a rule that every write tool requires approval, and buy Team or Enterprise so there is an audit log to check it against.

From Agents to an AI Step

Zapier's migration guide says Agents (agents.zapier.com) is moving to AI by Zapier, with no shutdown date set yet. The new step bills in tasks rather than a separate agent subscription. Default models are free, and higher model tiers apply a multiplier, so one step can cost 1, 3 or 5 tasks. Connecting your own model API key can reduce that multiplier.

Approval Granularity

The older Agents product was all or nothing on approvals. AI by Zapier sets approval per tool: the agent can look up a CRM record on its own and still stop before it updates one. A paused run sends an email link, and the reviewer approves or rejects from the Activity tab. Approve continues the run with the proposed fields; Reject stops it.

Free (100 tasks/month); Pro from $19.99/month and Team from $69/month, billed annually; Enterprise custom

Visit AI by Zapier (formerly Zapier Agents)
3

n8n

Best Open Source

Best for: Technical operations teams that want to self-host agents and see every step they take

“n8n is the only builder here you can run on your own servers from public source code, which keeps agent data and credentials inside your network. Its AI Agent node can require human review before it calls a specific tool, with the request routed to Slack, Teams, Gmail, Telegram or n8n's own chat. It is low-code rather than no-code: you wire nodes on a canvas and will meet expressions and JSON. Business users can run what an ops team builds.”

Pros

  • Self-hosted Community Edition is available on GitHub under n8n's fair-code Sustainable Use License, and the repository had about 210,600 stars when we checked in November 2026
  • Human review can be set on all tools of an AI Agent node or only selected ones; the reviewer approves the call with the AI's input or denies it
  • Review requests can go to n8n Chat, Slack, Discord, Telegram, Microsoft Teams, Gmail, WhatsApp Business Cloud, Google Chat or Outlook
  • Cloud plans start at 20 euros per month billed annually for 2,500 workflow executions

Cons

  • Audit logging and log streaming are Enterprise-only on n8n's pricing page; Starter and Pro have neither
  • The Business plan (667 euros per month billed annually) is listed as self-hosted only, so SSO on n8n's cloud means Enterprise
  • The Sustainable Use License is fair-code, not an OSI-approved open-source licence, which matters if you plan to resell or embed it
Honest Weakness: n8n gives you the most control and the most homework. Self-hosting means you patch the server, rotate the credentials it stores and back up its database. The audit trail a security team expects arrives only with Enterprise. Below that, the execution history is your record, which is useful for debugging but not a compliance log.

Human Review on Tool Calls

n8n's documentation describes human-in-the-loop for AI tool calls as a gate on individual tools. When the agent decides to call a gated tool, the workflow pauses until a reviewer approves or denies. n8n advises telling the agent in its system prompt which tools need approval and how to respond to a denial. That is the right pattern for agents: let them read, and make them ask before they write.

Workflow Automation vs Agents

n8n also appears on our automation tools comparison, where it is judged as a deterministic workflow engine. Here it is judged on its AI Agent node: tool calling, approvals and where the agent runs. If your process has fixed steps, a plain workflow is cheaper and easier to audit than an agent.

Community Edition self-hosted (Sustainable Use License); Cloud Starter 20 euros/month, Pro 50 euros/month, Business 667 euros/month, billed annually; Enterprise custom

Visit n8n
4

Gemini Enterprise (Workflow Builder)

Runner Up

Best for: Google Workspace organisations that want no-code agents grounded in company data

“Gemini Enterprise is Google's workplace agent platform, and its Workflow Builder lets non-admins build chat agents and multi-step workflows from a prompt or a visual editor. Admins keep the key control: an agent can only use the connectors, tools and permissions an administrator has provisioned. Approval and Request info steps pause a run until a person reviews it. The Business edition starts at $21 per seat per month with a 30-day trial.”

Pros

  • Business edition starts at $21 per seat per month, with a 30-day trial aimed at teams up to 300 people, per Google's Business edition page
  • Agents and workflows are limited to the connectors, tools and permissions an administrator has provisioned
  • Two human-in-the-loop step types: Request info asks the user questions, and Approval holds outputs for sign-off; paused runs show Needs Review in the Runs tab
  • Google's editions page lists VPC Service Controls, customer-managed encryption keys, data residency, Access Transparency and Cloud Audit Logs for Gemini Enterprise

Cons

  • Google's editions page labels custom no-code agent building as preview, and its feature matrix did not render which editions include it
  • An administrator must turn on the Workflow Builder feature toggle before anyone can use it
  • Frontline users can only use previously built agents, not create their own
Honest Weakness: Gemini Enterprise documentation is split by edition, and its pages disagree in places: the Business page targets teams up to 300, while the editions page lists Business at 1 to 500 users. Confirm with Google which security features your edition includes before you promise them to a compliance team.

Admin-Bounded Agents

Workflow Builder's governance model is simple to explain to a security team. Any employee can build an agent, but the agent inherits only what an admin provisioned: data connectors, tools and permissions. That bounds the damage a badly written agent can do, because its reach is set by the admin, not the builder.

Approval Steps in Practice

To add a review step, a builder selects Add step, opens Human in the loop and chooses Approval or Request info. Approval presents outputs for review before downstream steps run. Request info pauses and asks the user one or more questions. Either way, the paused run appears with a Needs Review status, so reviewers have one queue to work from.

Business edition from $21/seat/month (30-day trial); Standard, Plus and pay-as-you-go editions via Google Cloud

Visit Gemini Enterprise (Workflow Builder)
5

Lindy

Honorable Mention

Best for: Individuals and small teams that want inbox, calendar, meeting and phone agents without designing workflows

“Lindy sells finished assistants rather than a blank canvas: an executive assistant for email and calendar, a meeting note-taker, and a phone receptionist. Its safety default is the best of the five for non-technical users. Any action with outside impact, such as sending an email or publishing a document, waits for your approval, while read-only lookups run on their own. Enterprise controls like SSO and audit logs require the Enterprise plan.”

Pros

  • Approval is on by default for actions with outside impact, per Lindy's pricing FAQ, so a new user does not have to find a setting
  • Lindy states SOC 2 Type II, audited by Johanson Group, and offers HIPAA with a signed BAA on Enterprise
  • Lindy lists 3,000+ integrations plus MCP, and works inside Slack, email, text and calendar
  • Team starts at $29.99 per month per active user with 3,000 pooled credits

Cons

  • The free plan is a trial: $50 of credits that expire after 7 days
  • SSO, SCIM and audit logs are listed for Enterprise only
  • Credit pricing makes heavy use hard to forecast; top-ups cost $10 per 1,000 credits
Honest Weakness: Lindy is built to act for one person across their inbox and calendar, which is exactly the access an attacker wants. The approval default helps, but an assistant that reads every email will also read the malicious ones. Keep approvals on for anything that sends, and do not connect it to shared or admin mailboxes without the Enterprise audit log.

Assistants, Not Workflows

Lindy's homepage frames each agent as a role. The executive assistant runs inbox, calendar and meetings. The meeting agent takes notes and sends follow-ups. The receptionist answers calls and books appointments. That suits a founder or office manager who wants an outcome, not a flowchart. It suits a process owner less well, because the logic is harder to inspect than a Zap or an n8n workflow.

Credits and Limits

Team pricing scales with pooled credits, from 3,000 credits for $29.99 a month up to 1,750,000 credits for $9,999.50. Lindy's FAQ says credits pool across the team. Estimate usage on the free trial before committing, since 7 days of light use will not show what a busy inbox costs.

Free trial ($50 credits, 7 days); Team from $29.99/month per active user; Enterprise custom

Visit Lindy

Which One Should You Pick?

Use CaseOur Recommendation
A Microsoft 365 company whose security team must approve every agentCopilot Studio. Admin data policies control connectors and publishing, maker activity lands in Purview audit logs, and Sentinel can alert on agent behaviour. Add a Request for information step before any action that changes a record.
A 20-person team that already automates with ZapierAI by Zapier. Add the AI step to an existing Zap and switch on Require approval before running for every tool that writes data. Move to Team if you need SSO and an audit log.
Agents that must run inside your own networkn8n self-hosted. Gate risky tool calls with human review routed to Slack or Teams, and budget for Enterprise if auditors need audit logs and log streaming.
A Google Workspace company that wants agents over Drive, Gmail and its SaaS appsGemini Enterprise. Start with the $21 Business edition trial, keep connector permissions narrow, and add Approval steps before anything that writes back.
A founder who wants help with email, calendar and calls this weekLindy. Its approval-by-default behaviour suits someone who will not configure guardrails, but keep it off shared mailboxes until you are on a plan with audit logs.

How we evaluated

This comparison covers hosted no-code and low-code platforms that let a business team build an AI agent: software that reads data, decides, and takes actions in other apps. It does not cover code frameworks for developers, which have their own page, or AI agent security products, which are linked below.

Each platform was assessed on the criteria in the comparison table above:

  • Who can build: whether a business user can ship an agent without writing code, and what an admin must set up first.
  • Permissions: whether an admin can limit which connectors, data sources and actions an agent may use.
  • Human-in-the-loop: whether a run can pause for a named person to approve an action, and whether that is on or off by default.
  • Audit logs and identity: whether agent and builder activity is logged somewhere a security team can search, and which plan unlocks SSO.
  • Pricing model: the unit you pay on (seats, credits, tasks or executions) and the cheapest published starting price.

The security criteria follow the identity-first view of agents on the portal's Agent Security Map: an agent is a non-human identity, so the first questions are what it may access, who approves its writes, and what record it leaves.

What we reviewed

We reviewed each vendor's own product, pricing and documentation pages in November 2026. We did not run these products in a lab, and no figure on this page comes from our own testing. Prices change often; each is the published figure when we checked.

What we left out

  • OpenAI Agent Builder. OpenAI's deprecations page lists it as deprecated since June 3, 2026, with shutdown on November 30, 2026. The Agent Builder guide says the same and points to the Agents SDK or ChatGPT Workspace Agents.
  • Relevance AI. Its pricing documentation lists Pro from $19 a month and Team from $234 a month, billed annually. The Free plan is retired, and SSO, RBAC and audit logs are Enterprise-only. It is a strong fit for sales and go-to-market agents, but narrower than the five above.
Note

Editorial independence: this is a vendor-neutral comparison with no paid placements, sponsorships, or affiliate links. Rankings reflect fit for the stated use cases, not commercial relationships.

Frequently Asked Questions

How is this different from your Top 8 Agentic AI Frameworks comparison?
That page compares code frameworks such as LangGraph, CrewAI, the OpenAI Agents SDK and Google ADK, which developers use to write agents. This page compares hosted no-code and low-code builders that a business team can use without writing an application. If your agent needs custom code, version control and tests, start with a framework.
Why is OpenAI Agent Builder not on this list?
OpenAI's deprecations page lists Agent Builder as deprecated, with a notice date of June 3, 2026 and a shutdown date of November 30, 2026. OpenAI recommends the Agents SDK or ChatGPT Workspace Agents instead, and says ChatKit remains available. We do not rank a product that shuts down within weeks.
What is the single most important safety setting in an agent builder?
Human approval on any tool that writes, sends or deletes. Every platform here supports it, but Zapier and n8n leave it off until you turn it on. Approve writes, let reads run, and check that someone actually reviews the queue.
Do I need audit logs if a human approves every action?
Yes. Approval tells you a person said yes; an audit log tells you who built the agent, what it touched and when. In this comparison, audit logs come only with Copilot Studio's Purview integration, Gemini Enterprise's Cloud Audit Logs, or the Team or Enterprise tiers of Zapier, n8n and Lindy.
Can a no-code agent be tricked by content it reads?
Yes. Any agent that reads email, web pages or documents can meet hidden instructions written by an attacker, a problem called prompt injection. Microsoft's own documentation calls the Request for information step a way to add human oversight, and approval steps are the practical defence on every platform here.
Which agent builder is cheapest to start with?
Zapier's free plan includes 100 tasks a month, and n8n's Community Edition costs nothing to self-host beyond your server. Lindy's free plan is a 7-day trial with $50 of credits. Gemini Enterprise Business has a 30-day trial, then $21 per seat per month.

About the author

is the founder and creator of LoginRadius, a customer identity platform he built and scaled to over a billion users. He is now the founder of GrackerAI, a GEO platform for B2B SaaS and cybersecurity teams, and has spent more than 15 years building identity and security products.

Related Comparisons