A8gent
AI Agent PlatformsAI agent builder

Zapier Agents Pricing and Review (2026)

AI teammates that act across 7,000+ connected apps.

Reviewed by A8gent
Contents
  1. Verdict
  2. What is Zapier Agents?
  3. Best for, and not for
  4. Strengths and limits
  5. Pricing
  6. What people actually build with it
  7. What Zapier Agents actually costs at small-business volume
  8. The gotcha: the activity definition is wider than the word suggests
  9. What breaks in production, and how you find out
  10. What it connects to that you already run
  11. Zapier Agents alternatives: what to pick instead, and when
  12. Migration and lock-in: can you get your work out?
  13. FAQ
VerdictIs Zapier Agents worth it?

Zapier Agents is the only product in this incumbent-suite batch you can buy entirely on its own, and it is priced accordingly: free for 400 activities a month, or $400 a year for 1,500. It does not need a paid Zaps plan, which quietly makes it the cheapest agent in this directory to try. The limitation is the same number: 1,500 activities is roughly 50 a day, and an activity is any billable step the agent takes, so a busy agent exhausts a month in a fortnight. Treat it as a capable assistant for low-volume, judgement-heavy tasks and not as infrastructure. For anything running thousands of times a month, ordinary Zaps are cheaper and more predictable.

What is Zapier Agents?

Zapier Agents are AI teammates built on Zapier's automation platform that can browse the web, look things up in attached knowledge, and take actions across the thousands of apps Zapier connects to. Unlike a Zap's fixed trigger-action recipe, an agent works from instructions and decides which of its allowed tools to use. It is Zapier's bet on moving from deterministic workflows to autonomous ones, billed on a separate activities model from regular Zap tasks.

Best for

Teams already on Zapier who want AI handling fuzzy, judgment-based tasks their rigid Zaps cannot express.

Not for

High-volume production automation. At 1,500 activities a month the Pro cap is about 50 a day, and there is no published overage rate to spill into.

Strengths

  • Genuinely standalone: Agents plans are sold separately and do not require a paid Zaps subscription, which is unique in this batch
  • The free tier is 400 activities a month forever, enough to evaluate an agent properly rather than for 14 days
  • Activities are a separate meter from Zap tasks, so an agent cannot silently consume the automation budget your Zaps depend on
  • Per-run activity caps of 10 on free and 40 on paid plans stop a single runaway agent draining the month
  • Inherits Zapier's connector breadth, which remains the widest in the category and is the main reason to choose it over an agent-first tool
  • Live data sources, web browsing and a Chrome extension surface mean the agent works where you already are rather than only in a builder
  • Pricing is published in full with no seat component, no minimum commitment and no prerequisite licence

Limitations

  • 1,500 activities a month on Pro is the binding constraint, and it works out at roughly 50 a day
  • No overage rate is published for activities, so exceeding the cap means the agent asks permission to continue rather than billing on
  • The activity definition is broad - a trigger, a knowledge lookup, an action, a web search and a browse each count - so activity counts run ahead of intuition
  • Team accounts share a single activity pool, so one enthusiastic colleague can consume the organisation's allowance
  • Zapier states Agents does not currently support the app and action restrictions that may be in place on an Enterprise account, which is a governance gap
  • Non-determinism is acknowledged in Zapier's own FAQ: the same input does not guarantee the same output, which is awkward for anything auditable
  • For deterministic, high-volume work, ordinary Zaps are cheaper and more reliable, so the product competes with its own parent

Zapier Agents pricing

The billing unit is an activity, which Zapier defines as actions the agent takes in behaviours or in chat, browsing the web, or looking up information from attached knowledge. It is a separate economy from Zap tasks and does not draw on your task allowance. Agents Free gives 400 activities a month forever; Agents Pro is $400 billed annually, quoted as $33.33 a month, for 1,500 activities; Agents Enterprise is contact-for-pricing with a custom activity allowance, agent sharing across the organisation, audit logs and restricted apps support. On a Zapier Team account the whole team shares one pool of activities. Zapier does not publish an overage rate for activities, so the cap is a cap rather than a meter.

Zapier Agents pricing tiers
PlanPriceWhat you get
Agents Free$0400 activities a month forever, live data sources, web browsing, Chrome extension. No paid Zapier plan required
Agents Pro$400/yr ($33.33/mo)1,500 activities a month, plus everything in Free
Agents EnterpriseContact salesCustom activity allowance, agent sharing across the organisation, enterprise audit logs, restricted apps support
Zaps Free$0The separate core platform: 100 tasks a month, two-step Zaps, unlimited Zaps, Tables and Forms
Zaps Professionalfrom $19.99/moMulti-step Zaps, unlimited premium apps, webhooks. Tasks and activities are separate meters
Zaps Teamfrom $69/mo25 users, shared connections, SAML SSO. Team accounts share one pool of Agents activities

Pricing reflects public plans as of August 6, 2026 and can change. Check Zapier Agents for the latest.

What people actually build with it

Zapier Agents earns its place on tasks where the next step depends on reading something a fixed workflow could not have anticipated. That is a narrower category than the marketing implies, and being clear about it is the difference between an agent that saves an hour a day and one that burns 1,500 activities on work a Zap would have done for a fraction of the money.

Where the judgement is real

Inbound lead research and routing. A form submission fires the trigger, the agent searches the web for the company, reads what it finds, decides which of your service lines fits, and writes an enriched record with a recommendation. Four to six activities per lead. At 1,500 activities a month that is roughly 250 to 375 leads, which suits a professional services firm or an agency and does not suit a high-volume ecommerce operation.

Inbox triage that goes beyond rules. The agent reads a shared mailbox, decides whether a message is a support issue, a sales enquiry or noise, drafts a reply for review and files it. This is the canonical agent task because the classification genuinely needs comprehension. Keep a human approval step on the send, because Zapier's own documentation is candid that the same input will not always produce the same output.

Competitive and supplier monitoring. On a schedule, the agent browses a list of pages, notices what changed against what it saw last time, and posts a summary to Slack. Web browsing and scraping each count as activities, so a daily check across ten pages is about 300 activities a month before the agent does anything with them. Budget accordingly.

Meeting follow-up for a small sales team. After a call transcript lands, the agent extracts the commitments, drafts the follow-up email, and updates the CRM record. Three or four activities per meeting makes this comfortably affordable at Pro for a team doing a few hundred meetings a month.

Where a Zap is the right answer instead

Anything with fixed steps and high volume. Moving a record from a form to a CRM, sending an appointment reminder, posting a notification: these are deterministic, they run thousands of times, and they belong in ordinary Zaps where the task allowance is far larger and the behaviour is identical every time. Running them through an agent costs more, introduces variance, and eats a cap that should be reserved for judgement.

Customer support deflection at volume. A support conversation typically consumes several activities, so 1,500 a month does not stretch far. Intercom Fin bills $0.99 per resolved outcome with no cap, and HubSpot Breeze $0.50 per resolution. Both are the right shape for that job and Agents is not.

Anything needing an auditable, repeatable decision trail. Zapier acknowledges non-determinism in its own FAQ. For compliance-shaped work, use a deterministic path and keep the agent out of it.

Bottom line: Zapier Agents is worth its activity cost on lead research, inbox triage and monitoring where comprehension is required, and is a poor substitute for an ordinary Zap on anything deterministic or high volume.

What Zapier Agents actually costs at small-business volume

Agents Pro is $400 billed annually, quoted as $33.33 a month, for 1,500 activities. Divide that and an activity costs about 2.2 cents, which sounds trivially cheap until you count how many an agent uses. The honest way to budget is per run rather than per month: a four-activity agent run costs about nine cents and you get roughly 375 of them.

Applied to this batch's reference workload of 1,000 support conversations a month, Agents simply does not reach. Even at a conservative two activities per conversation - the trigger and one knowledge lookup, before the agent replies or writes anything - that is 2,000 activities against a 1,500 cap, and a realistic support exchange uses more. There is no published overage rate to spill into, so the honest answer is that this workload requires Agents Enterprise at unpublished pricing. We are not going to invent that number.

That is a genuine finding rather than a criticism. Agents is priced and capped for tens of runs a day, not thousands, and the table below shows what the same money buys elsewhere.

The same reference workload across this batch: 1,000 support conversations a month, 700 resolved without a human. Rates read 2026-08-06.
PlatformBilling unitPrerequisite licenceAll-in monthly total at the reference workload
Zapier AgentsActivity, 1,500 for $33.33/moNoneNot viable; the Pro cap is exceeded and Enterprise pricing is unpublished
Intercom FinOutcome, $0.99None$693 standalone
HubSpot BreezeResolution, $0.50HubSpot Professional~$590 plus $1,500 one-time onboarding
Agentforce20 Flex Credits per action, or $2 per conversationSalesforce Service Cloud~$825 to ~$2,525
Copilot StudioCopilot Credits, $200 per 25,000Microsoft 365 tenant plus the pack$200

Where Agents does win on price

At the volume it is designed for, nothing here is close. A firm running 200 lead-research agent runs a month at five activities each uses 1,000 activities and pays $33.33. The equivalent on Agentforce Flex Credits, assuming five actions, would be $100 in credits plus Salesforce licences. On Fin it would not be a support outcome at all. Agents is the cheapest way to put judgement into a low-volume workflow, and the free tier at 400 activities means the evaluation costs nothing.

The second saving is structural. Because activities are a separate meter from Zap tasks, an experiment with an agent cannot consume the task budget your production Zaps depend on. That sounds like a small thing and it removes the single most common reason teams refuse to try an agent inside an automation platform they already rely on.

Bottom line: Agents Pro works out at about 2.2 cents an activity and roughly 375 four-step runs a month, which is excellent for judgement-heavy work and insufficient for a support deflection workload.

The gotcha: the activity definition is wider than the word suggests

Everyone reads 1,500 activities as 1,500 things the agent does for you. Zapier's own definition is wider: activities include the actions your agent takes in behaviours or in chat, browsing the web, and looking up information from attached knowledge. The published pricing guidance is wider still, naming the trigger itself, answering from a knowledge source, running an action, web browsing or scraping, a web search, and a message sent via the Chrome extension as separately billable.

So an agent that is triggered by a new email, checks your knowledge base, searches the web for context and writes a CRM record has used four activities to do one job. A monitoring agent that visits ten pages daily uses roughly 300 a month just looking, before deciding anything. Under those definitions, 1,500 a month is not 1,500 jobs. It is closer to 300 to 400 jobs of ordinary complexity, or about a dozen a day.

The compounding factor is that the number of activities is not fixed by you. Agents decide how many lookups and searches they need, so the same nominal job can cost three activities on one run and seven on another. Zapier's per-run caps of 10 on Free and 40 on paid plans exist precisely because of this, and they are the reason a single agent cannot destroy your month in one execution. They do not stop a fleet of agents doing it over three weeks.

The mitigation is to measure a real agent for a week on the free tier before committing, and to strip out everything that does not need judgement. Every deterministic step you move into an ordinary Zap is an activity you keep. That single discipline is usually worth more than upgrading a plan.

Bottom line: A trigger, a knowledge lookup, a web search and an action each bill as separate activities, so 1,500 a month is nearer 300 to 400 real jobs rather than 1,500.

What breaks in production, and how you find out

Zapier is unusually honest about the failure mode that matters most here, which is worth crediting: its own Agents FAQ states plainly that you do not get the same outcome from AI every time, and attributes it to non-determinism. Everything else that breaks follows from designing as though that were not true.

The same input, a different answer

An agent that classified a message correctly ten times will occasionally classify it differently on the eleventh. In a triage workflow that is a mis-filed email. In a workflow that writes to a system of record, it is bad data that nobody notices until a report looks wrong. The structural fix is a human in the loop on any write that matters, and a deterministic Zap for any step that must be identical every time.

Detection is by sampling, not by alerting, because nothing errors. Review a sample of agent decisions weekly against what you would have decided. If you cannot describe what a correct decision looks like well enough to sample against, the task is not ready to be automated by an agent.

The activity cap arriving mid-month

When the allowance runs out, the agent asks permission before continuing rather than billing on, and no overage rate is published. That is a budget-safe design and an availability risk: your automation stops. On a Team account the pool is shared across everyone, so the colleague who built an enthusiastic monitoring agent can stop your lead routing. Watch consumption weekly rather than monthly, and keep production-critical work in Zaps where the task economy is separate and larger.

Governance gaps on larger accounts

Zapier states that Agents does not currently support the app and action restrictions that may be in place on an Enterprise account. If your organisation has deliberately restricted which applications automation may touch, Agents is a hole in that policy rather than a participant in it. That is a procurement conversation, not a technical one, and it is better had before deployment than after. Combine it with prompt injection risk on any agent that reads untrusted inbound content, and the case for keeping agents away from sensitive systems is straightforward.

Bottom line: Zapier acknowledges its agents are non-deterministic, so anything that writes to a system of record needs a human approval step and anything that must be identical every time belongs in an ordinary Zap.

What it connects to that you already run

This is the reason to choose Zapier Agents over an agent-first tool, and it is a good one. Agents inherit the Zapier connector catalogue, which remains the broadest in the category, so the agent can reach the accounting package, the booking system and the CRM you already run without any of them needing an API project. For a business with an ordinary mixed stack, that breadth beats a deeper integration with a suite you do not own.

Beyond connectors, Agents adds three surfaces the core platform lacks: live data sources that the agent can consult during a run, web browsing and scraping, and a Chrome extension so you can talk to your agents in the browser where the work is happening. Each of those bills as activities, which is the trade. Zapier also ships Zapier MCP, so agents can be exposed to and consume tooling through the Model Context Protocol rather than only through Zapier's own action library.

What it does not connect to is anything requiring deep, permission-aware knowledge of a single vendor's data model. Agentforce reads Salesforce objects with your sharing rules; HubSpot Breeze grounds answers in CRM records; Copilot Studio can ground across your whole Microsoft tenant through Graph. Zapier Agents knows how to call thousands of APIs and nothing about what your data means. That is the correct trade for a mixed stack and the wrong one for a single-suite business.

Connector breadth against integration depth, across this batch.
PlatformReachDepth in one systemBest when
Zapier AgentsThousands of apps via the Zapier catalogue, plus web browsing and MCPShallow by designYour stack is mixed and no vendor owns it
AgentforceSalesforce plus integration projectsNative objects and sharing modelSalesforce is the system of record
HubSpot BreezeHubSpot plus its app marketplaceNative CRM recordsHubSpot is the system of record
Copilot StudioMicrosoft estate plus Power Platform connectorsTenant-wide Graph groundingEverything is inside Microsoft 365
Intercom FinAny major helpdesk plus custom actionsHelp centre content and your APIsSupport is the job and the helpdesk stays

Bottom line: Zapier Agents trades data-model depth for connector breadth, which makes it the right agent for a mixed stack and the wrong one for a business whose truth lives in a single suite.

Zapier Agents alternatives: what to pick instead, and when

Zapier Agents sits between two things that are both better at their own job: ordinary Zaps, which are cheaper and deterministic, and agent-first platforms, which are more capable per run. Knowing which side of that line your workload falls on saves more money than any plan choice.

Zapier Agents against four alternatives, priced on their own natural workload rather than on support deflection.
PlatformBilling unitEntry costWho it suitsDoes better than Zapier Agents
Zapier AgentsActivityFree at 400/mo; $400/yr for 1,500Low-volume judgement tasks on a mixed stackConnector breadth and a genuinely free tier
ZapierTaskFree at 100 tasks; Professional from $19.99/moDeterministic, high-volume workflowsCheaper and repeatable for fixed steps
LindyCreditPublished per-plan credit allowancesAssistant-style agents with a richer agent-first UXPurpose-built agent behaviour rather than an add-on
GumloopCreditPublished credit tiersAgentic data workflows and enrichment at volumeBetter suited to bulk processing pipelines
MakeOperationCheaper per operation than Zapier tasksCost-sensitive, high-volume automationMaterially lower unit cost at scale

When to pick the alternative

Pick ordinary Zapier when the steps are known. This is the most common correct answer and the one Zapier's own marketing will not push you toward. Professional starts at $19.99 a month with a task allowance that dwarfs 1,500 activities, and a fixed workflow runs the same way every time. If you can write the steps down, you do not need an agent.

Pick Make.com if cost per unit of work is the constraint and volume is high. Make's operation pricing is materially cheaper than Zapier tasks at scale, and the visual flow builder is at least as capable. The trade is a steeper learning curve and a smaller connector catalogue.

Pick Lindy if you want an agent-first product rather than an agent bolted onto an automation platform. The agent experience is more coherent and the credit model is designed around agent runs rather than retrofitted around tasks. If agents are the point rather than an addition, start there.

Pick Gumloop for bulk agentic processing, such as enriching thousands of records, where Zapier's activity cap would bind immediately and a credit-based pipeline tool will not.

Pick Intercom Fin or HubSpot Breeze if the job is customer support. Zapier Agents is not a deflection product and its cap makes that obvious the moment you do the arithmetic.

Bottom line: If you can write the steps down, use ordinary Zaps or Make.com instead; reserve Zapier Agents for the small share of work where the next step genuinely depends on comprehension.

Migration and lock-in: can you get your work out?

Zapier Agents has moderate lock-in and, unusually, the lock-in is not really about the agents. Agent configuration is proprietary, there is no export format another platform reads, and rebuilding an agent elsewhere means rewriting its instructions, knowledge attachments and behaviours from scratch. But an agent is typically a page of instructions and a handful of connections, so that rebuild is days rather than months.

The heavier dependency is the connection layer beneath it. Every app connection is an OAuth grant held by Zapier under a specific person's account, and moving away means re-authorising every one of them somewhere else. Businesses discover this at the worst possible moment, which is when the person who authorised the connections leaves. Use shared connections on a Team plan and service accounts rather than individual logins, and that risk mostly disappears.

There is one genuine advantage worth stating: because Agents plans are standalone, leaving Agents does not mean leaving Zapier, and leaving Zapier does not mean losing a suite licence. Compare Agentforce, where the agent is fused to a CRM you also pay per seat for, or HubSpot Breeze, where dropping the agent means either keeping a Professional subscription you no longer need or migrating a CRM. Zapier's separation of the two plan families is the cleanest exit position in this batch.

The practical hedge is the same one that applies everywhere: document what each agent is meant to achieve in plain business language, keep the deterministic parts of the workflow in a form you could rebuild anywhere, and keep a handful of real inputs with correct outputs so that any replacement can be measured rather than trusted. Where the alternative is a rebuild, being able to start from written intent is the difference between a fortnight and a quarter.

Bottom line: Because Agents is sold separately from Zaps, leaving it costs you a page of instructions and a set of re-authorisations rather than a suite licence, which is the cleanest exit in this batch.

Zapier Agents FAQ

How much does Zapier Agents cost?

Agents Free is $0 forever with 400 activities a month. Agents Pro is $400 billed annually, quoted as $33.33 a month, for 1,500 activities a month. Agents Enterprise is contact-for-pricing with a custom activity allowance plus agent sharing, audit logs and restricted apps support. Agents plans are separate from Zaps plans, so you do not need a paid Zapier subscription to use them. There is no seat component and no minimum commitment.

What counts as an activity in Zapier Agents?

Zapier defines an activity as the actions your agent takes in behaviours or in chat, browsing the web, or looking up information from attached knowledge. In practice that means a trigger firing, a knowledge lookup, an action running, a web search and a scrape each count as one. A single agent run that checks a knowledge source, searches the web and then writes to a CRM has used at least three. This is why activity consumption runs ahead of what people expect from the word.

Do I need a paid Zapier plan to use Agents?

No, and this genuinely distinguishes it from the rest of this category. Agents is sold as its own plan family with its own free tier, so you can run agents without any Zaps subscription. Compare HubSpot Breeze, which requires a Professional subscription, or Agentforce, which requires Salesforce licences. If you do have a Zapier Team account, note that the whole team shares one pool of Agents activities rather than each member getting their own.

What happens when I run out of activities?

The agent asks permission before continuing rather than billing you for an overage, and Zapier does not publish an overage rate for activities. That makes the cap a genuine ceiling rather than a soft limit, which is safer for your budget and worse for your reliability. There are also per-run caps of 10 activities on Free and 40 on paid plans, which stop a single runaway agent draining the month in one execution. Plan capacity rather than relying on spillover.

Zapier Agents vs Zaps: which should I use?

Use Zaps when you know the steps and they do not change, which covers most business automation. Use Agents when the next step depends on judgement about content the workflow cannot predict. The economics reinforce it: a deterministic multi-step Zap on the Professional plan starts at $19.99 a month with a much larger task allowance, while Agents Pro is $33.33 a month for 1,500 activities. Agents is the more expensive per unit of work by a wide margin, and that is correct, because it is doing something a Zap cannot.

Is Zapier Agents good for customer support deflection?

Not at any real volume, and we would steer you elsewhere. At 1,500 activities a month, and with a single support conversation typically consuming several activities across the trigger, the knowledge lookup and the reply, Agents Pro will not sustain even a few hundred conversations. Intercom Fin at $0.99 per outcome or HubSpot Breeze at $0.50 per resolution are built for that workload and bill in a unit that matches it. Agents is for internal, judgement-heavy, low-volume work.

Does Zapier Agents have a free plan?

Yes, and it is unusually good: 400 activities a month, free forever, including live data sources, web browsing and the Chrome extension, with no paid Zapier plan required and no card. The per-run cap on Free is 10 activities, which limits how ambitious a single run can be but is plenty for evaluation. This is the cheapest genuine trial in this batch by a wide margin, since Copilot Studio effectively needs a $200 pack and Agentforce needs Salesforce licences.

Is Zapier Agents reliable enough for production?

For low-stakes internal work, yes. For anything auditable, be careful, and Zapier says so itself: its own FAQ notes that you do not get the same outcome from AI every time because of non-determinism. Add the missing app and action restrictions on Enterprise accounts and the shared team activity pool, and the governance story is thinner than the rest of the platform's. Put a human approval step in front of anything that writes to a system of record, and use Zaps for the steps that must be identical every time.

Can you build and run this for us?

Yes, and the most valuable part is usually deciding which half of the job should be an agent at all. We map the workflow, put the deterministic steps in ordinary Zaps or n8n where they are cheaper and repeatable, and reserve agent activities for the genuinely judgement-shaped parts so the cap is not wasted. That starts with a free automation audit, and the handover includes documentation you own outright plus a recurring plain-language report of what ran, what it cost and what needs attention.

Looking at alternatives? Zapier Agents is the easiest thing in this directory to recommend trying and one of the harder ones to recommend depending on. 400 free activities a month with no card and no prerequisite plan is a genuinely generous way to find out whether an agent helps you, and Zapier's connector catalogue means it will reach the systems you actually run. Where it beats us outright is time to first working agent, which is minutes. Where it does not is anything that has to run reliably thousands of times a month, because the activity cap is low, the overage behaviour is a stop rather than a meter, and Zapier itself acknowledges the output is non-deterministic. Use it to discover the workflow, then build the workflow properly. See the Autonoly review.