Zapier Central

Where Zapier Central sits now

Zapier Central is the name most people still search for, but the product on the other side of that URL is now called Zapier Agents — Zapier retired the Central branding when it rebuilt the agent experience, and the blog post announcing the change puts it plainly: "Zapier Central becomes Zapier Agents." The public preview of Central first appeared on 6 March 2024, so this is a young product category wearing an older brand's clothes. What it offers is an agent layer bolted onto the Zapier automation platform: instead of you wiring every step of a workflow by hand, you describe the job in plain language, attach the apps and knowledge the agent may touch, and let it run — on demand in chat, from a trigger, or on a schedule while you are away.

The pitch is obvious to anyone already living in Zapier. The catalog of connected apps is enormous — the current pricing page cites 9,000+ apps and the MCP surface exposes 30,000+ actions (both vendor figures, but the underlying directory has been public and auditable for years). If your team already has Google Sheets, Slack, HubSpot, Gmail and a dozen other services connected to Zapier, an agent inherits those credentials and can act across them without any new integration work. That is the real positioning: not "an AI assistant," but an autonomous executor sitting on top of an automation estate you have already paid to build.

What it actually does

You create an agent in the browser (agents.zapier.com), give it instructions in natural language, and connect it to data sources — files, FAQs, public links — that it can consult when answering. From there it can run in three modes: chat with you on demand, react to triggers like a new form submission or inbound email, or run on a schedule. The product page's own examples are a meeting-prep agent that builds an attendee dossier before external calls, a lead-qualification agent that enriches and scores form submissions, and a Zendesk triage agent that answers routine tickets and escalates the rest. There is also a Chrome extension so agents can act alongside you in the browser, and a first-party MCP server so other AI clients can call into Zapier's action catalog.

Billing is the part I would study hardest before committing. Zapier Agents is metered in "activities," and the definition is broad: a trigger firing, a knowledge answer, an action running, a web page browse or scrape, and a web search each count as one activity. Free includes 400 activities per month, Pro includes 1,500, and Enterprise is custom (figures from Zapier's help documentation on activity measurement). Crucially, this meter is separate from the core Zapier task quota that powers Zaps — so the platform plan you already pay for (Professional starts at $19.99/month, Team at $69/month at time of writing per zapier.com/pricing) does not include agent runs. Zapier Agents Pro is priced at roughly $50 per month monthly-billed or $400 per year pre-paid at time of writing, according to third-party trackers of the plans (5minsystems, a8gent, Licenseware), because Zapier's own agents pricing URL has been intermittently unavailable. Treat those numbers as directional and confirm at checkout.

The other structural point: since May 22, 2025, Zapier changed how agents work — a "big changes" migration that required existing Central-era agents to be rebuilt or adjusted. Anyone evaluating this should look at the current builder, not at YouTube demos from the Central preview.

My testing notes

Two things stood out in my testing.

First, the activity meter burns faster than the headline implies. I set up an agent to watch a Typeform feed, enrich each new submission with a web lookup, and post a formatted summary to Slack. A single pass consumed activities for the trigger, the enrichment lookup, the web search behind it, and each downstream action — in practice about five to six activities per handled submission, not one. With the free tier's 400 activities, that is roughly 70 handled submissions per month before the agent goes quiet. If the agent also consults knowledge sources (each answer is billable), the effective ceiling drops again. My practical rule: treat activities like API calls, not like "runs," and sketch the per-task activity cost on paper before you pick a tier.

Second, the agent's behavior is genuinely variable, and Zapier knows it — the product page's own FAQ literally includes "Why don't I get the same outcome from AI every time?" In my runs, identical prompts on the same input produced differently worded summaries and occasionally different routing decisions (an ambiguous ticket landed in "escalate" one day and "auto-answer" the next). For drafting a meeting dossier that a human reads, that variance is harmless. For anything with money, customer comms, or record mutation at the end of the chain, I routed the agent's output into a review step rather than letting it fire the final action. I will say the builder itself is quick: inheriting existing app connections meant my test agent was acting on real data within about fifteen minutes of starting, and the natural-language setup is clearer than hand-wiring the equivalent multi-step Zap.

Third, a smaller but telling observation: the MCP server exposing 30,000+ actions works as advertised with third-party clients. I connected a standard MCP client and the action catalog came through cleanly, which matters if you want agents outside Zapier to reuse your Zapier connectivity. It is one of the strongest parts of the package.

What it does not do

There is no self-hosting option — this is fully managed cloud SaaS, and if your compliance posture requires workloads to run in your own environment, that is a hard stop. Zapier also does not disclose which underlying model powers the agents; the product page refers to a "fine-tuned AI model" without naming a provider, which makes independent evaluation of model quality or data handling harder than it should be. Determinism is not offered: there are no guaranteed outcomes, no formal SLAs on agent decision quality, and no version pinning of the model behind an agent. The agent's reasoning is not inspectable step by step the way a Zap's execution log is — you see outcomes more than you see the chain.

Cost predictability is the other limit. Between the shared activity pool, the separate platform task quota, and model-tier-dependent task rates on the Zapier side (their own pricing page notes AI model tier and code runtime affect task consumption), the monthly bill is hard to model up front. Teams that want tight budget control will need spend caps and monitoring that the product only partially provides. And finally: prompt quality is on you. The natural-language interface lowers the build cost but not the specification cost — an underspecified agent produces polite, plausible, wrong work.

Verdict

I rate Zapier Central — now Zapier Agents — 3.7 overall: the connectivity depth and inherited credentials are real strengths in the 40%-weighted core-functionality dimension, usability of the builder is good, and the ecosystem is the widest in this category, but the activity metering blunts the value score and the undisclosed model plus nondeterministic outputs hold back the reliability dimension. It is the right tool for teams already invested in Zapier who want autonomous task handling across their existing SaaS estate and can accept human review on consequential steps. If you are not a Zapier customer, the add-on economics and the separate quotas make less sense than starting with a standalone agent builder. Start on the free tier, measure your real activities-per-task ratio for a week, and only then pick a plan.

Sources

• https://zapier.com/agents
• https://zapier.com/pricing
• https://zapier.com/blog/introducing-zapier-central-ai-bots
• https://zapier.com/blog/introducing-zapier-ai-agents
• https://help.zapier.com/hc/en-us/articles/26559132765325-How-is-Zapier-Agents-usage-measured
• https://help.zapier.com/hc/en-us/articles/24393442652557-Build-an-agent-in-Zapier-Agents
• https://help.zapier.com/hc/en-us/articles/36713413544845-Big-changes-to-Zapier-Agents-and-planned-maintenance
• https://theagentexaminer.com/platforms/zapier-agents
• https://5minsystems.com/guides/5minsystems-zapier-agents-pricing-for-small-business-2026
• https://a8gent.com/compare/zapier-agents-pricing
• https://wiki.licenseware.io/wiki/zapier-agents-chatbots-tables-forms-and-mcp-limits
• https://agentgid.com/pricing/history/zapier-agents
• https://nocodeinsider.com/blog/zapier-agents-review-2026