The Second Meter: What AI Agents Really Cost on Zapier, Make & HubSpot (2026)

admin
By admin
14 Min Read

Affiliate disclosure: This post contains affiliate links. If you buy through them, we may earn a commission at no extra cost to you. Read this before the numbers: this post argues that Make is the value pick for agents — and we earn a 35% commission for 12 months on Make referrals. The math below is fully documented so you can check our work; if the numbers ever stop favoring Make, the recommendation changes. Every price checked October 2026.

Here’s the pricing problem nobody covers: when you add an AI agent to your automation stack, you don’t just pay for the agent. You pay for the agent plus the platform underneath it — and on some platforms, the agent runs on an entirely separate meter with its own units, its own plan, and its own bill.

Call it the second meter. Zapier’s agents bill in activities, separate from tasks. HubSpot’s agents bill in credits, on top of per-seat subscriptions. Make’s agents run inside your existing credit pool — no second meter at all.

This post documents each platform’s agent pricing from published price sheets, then models the true monthly cost for a 1-person business — base plan plus agent layer, itemized, no hand-waving.

The agent pricing table (documented, dated)

Zapier AgentsMake AI AgentsHubSpot Breeze agentsn8n AI Agent node
How agents are soldSeparate product (agents.zapier.com), separate subscriptionIncluded in every paid plan — no separate SKUOutcome-based pricing on top of HubSpot seatsIncluded in every plan; node in the workflow canvas
Base plan costProfessional: $19.99/mo annual (750 tasks/mo)Core: $9/mo annual (10,000 credits/mo)Service Hub Professional: ~$90/seat/mo annual (+ ~$1,500 onboarding)Cloud Starter: €20/mo annual (2,500 executions)
Agent meterActivities — Free: 400/mo; Pro: 1,500/moCredits — same pool as everything elseHubSpot Credits — ~$10 per 1,000Executions — one execution per agent run, however many tool calls inside
Agent add-on price$400/yr ≈ $33.33/mo (Pro, annual)$0 — bundled$0.50 per resolved conversation (Customer Agent, 50 credits); $1.00 per recommended lead (Prospecting Agent, 100 credits)$0 — bundled; self-hosted is $0 platform cost
Free tier400 activities/mo1,000 credits/mo on Free plan28-day trial; 500 credits on Starter (no Customer Agent access)Cloud trial; unlimited self-hosted
Included agent allowanceNone beyond Free tierFull 10,000-credit pool~3,000 credits/mo on Professional (≈ 60 resolutions)Full 2,500-execution pool

All figures from vendor pricing pages and documentation, re-checked October 2026. HubSpot’s outcome-based pricing took effect April 14, 2026, per HubSpot’s own announcement.

What each “second meter” actually meters

Flat-lay: four pastel envelopes arranged like an ascending bar chart with wooden tokens spilling out, symbolizing cost comparison

Zapier: activities are not tasks, and not runs. One agent run can consume several activities. When the monthly allowance runs out, agents stop running actions (chat access continues). Note Zapier is also documenting a migration path: Agents moving toward “AI by Zapier,” where agentic tool use lives inside the Zap editor — AI steps billing 1 task (Standard), 3 tasks (Advanced, the default since June 15, 2026), or 5 tasks (Premium) per step, every tool call at the same multiplier, MCP tool calls at 2 tasks each. So Zapier has two agent-adjacent meters today. Price both before you build.

Make: one meter, everything draws from it. AI Agents, the AI Toolkit, the MCP server, and 350+ AI apps all run on your plan’s credit pool. Two honest caveats from Make’s own documentation: AI Provider actions can consume more than 1 credit per step, and the Code app burns 2 credits per second — so agent-heavy scenarios need their own credit math. Connect your own AI provider and Make charges the operations while your provider bills tokens separately.

HubSpot: pay per outcome, on top of a $90/mo floor. The Customer Agent costs $0.50 only when a conversation is resolved — no human handoff within 72 hours. But agents require Service Hub Professional or Enterprise, the 3,000 included monthly credits cover roughly 60 resolutions, and those credits are shared with the Data Agent and other Breeze actions. Extra credits run about $9–10 per 1,000.

n8n: the flat-rate outlier. The AI Agent node, MCP server/client, and human-approval steps ship in every cloud plan. Agent loops cost tokens, not steps — roughly 40 tool calls still count as 1 execution. Self-hosted, you bring your own model keys and the platform cost is $0 (server costs aside).

n8n

True monthly cost: one solo founder, one support agent, 500 resolutions

The scenario: a 1-person business runs a support agent that resolves ~500 customer conversations a month. Itemized, base plan plus agent layer:

Base planAgent layerLLM/token spendTrue monthly total
ZapierProfessional $19.99/mo annualAgents Pro $33.33/mo (1,500 activities)Included in activities~$53.32/mo
MakeCore $9/mo annualIncluded ($0)Separate (BYOK provider bills you directly)~$9/mo + your token bill
HubSpotService Hub Pro $90/seat/mo3,000 credits incl. (≈60 resolutions); 440 × $0.50 = $220 overageIncluded in credits~$310/mo
n8n CloudStarter €20/mo annual (~$22)Included ($0)Separate (your own keys) or bundled assistant credits~$22/mo + your token bill
n8n self-hosted$0 (software)Included ($0)Your own keys~$5–15/mo VPS + token bill

Three things this table is honest about:

  1. Zapier’s $53 assumes 1,500 activities cover 500 resolutions. If your agent averages more than 3 activities per resolution — very plausible for multi-step support work — you outgrow Pro’s allowance with no published mid-tier between Pro and Enterprise. Model activities-per-resolution before committing, not after.
  2. Make and n8n exclude the LLM token bill because it’s genuinely separate — your provider key, your usage, billed by them. For 500 support resolutions this is typically single-digit dollars on efficient models, but it’s your variable to control.
  3. HubSpot’s $310 includes a seat you’re probably already paying for. On Service Hub Professional anyway, the marginal agent cost is the $220 in overage credits, not the full $310. But if you’re buying HubSpot for the agent, the $90 floor plus ~$1,500 onboarding is the real entry price.

AWL verdict: For a solo founder, Make wins this scenario at $9/mo plus tokens — agents bundled, no second subscription, no per-outcome meter. n8n self-hosted is cheaper still if you’ll run a server. Zapier is the price of approachability: fine for non-technical teams, but ~6× Make for the same 500 resolutions. HubSpot’s per-resolution pricing is the fairest unit in the industry and the most expensive total for a one-person shop.

Make

The bundling verdict: who actually includes what

Strip away the marketing and there are only three packaging strategies:

  • Bundled (Make, n8n): agents are a feature of the plan you already pay for. Your existing usage pool covers them. The bill goes up only if your usage does.
  • Add-on (Zapier): agents are a second product with a second subscription ($33.33/mo) and a second unit (activities). The base plan doesn’t cover a single agent run.
  • Pay-per-outcome (HubSpot): agents are metered by results ($0.50/resolution), which sounds cheapest per unit — until the subscription floor ($90/mo), the onboarding fee, and the shared credit pool enter the chat.

The pattern to watch: every vendor is converging on usage-based agent billing, but from opposite directions. The solo founder’s move is the same either way: price the agent layer separately from the base plan before you build, because the base plan’s marketing page won’t mention it.

One more honest note: Zapier has no public affiliate program — it’s partner-only. We earn nothing if you pick Zapier after reading this. We earn 35% for 12 months if you pick Make. We’ve shown you the full table anyway. Check the math, not our incentives.

FAQ

Do I need Zapier Agents Pro if I already pay for Professional?

Yes — they’re separate products. Professional’s 750 tasks cover Zaps; Agents run on activities (400/mo free, then $33.33/mo Pro for 1,500). One subscription does not cover the other. This is the entire “second meter” problem in one question.

Are Make’s AI agents really “free”?

They’re included, not free — included in every paid plan starting at Core ($9/mo annual), drawing from the same 10,000-credit pool as your scenarios. Heavy agent use burns credits faster, and AI Provider actions can cost more than 1 credit per step. Plus your own LLM token bill if you bring your own keys. “No second subscription” is the accurate claim; “free” is not.

Can I just use ChatGPT/Claude instead of a platform agent?

For answering questions, yes. For an agent — software that takes actions across your tools (refund the order, update the CRM, file the ticket) — you need the orchestration layer: the app connections on Zapier, the scenario canvas on Make, the CRM context on HubSpot. The chatbot is the brain; the platform is the hands. Price both.

What’s the cheapest legitimate way to run an AI agent as a solo founder?

Two documented options: self-hosted n8n (~$5–15/mo VPS plus your own model keys, $0 platform cost) or Make Core ($9/mo annual plus your token spend). Both land under ~$15/mo before tokens. The real trade is your time — n8n costs ops hours, Make costs the subscription.

Methodology

This is a documented price-sheet analysis, not a hands-on test. What was verified: October 2026 agent pricing for Zapier Agents (Free 400 activities/mo, Pro $400/yr ≈ $33.33/mo for 1,500 activities), Make AI Agents (included in all paid plans; Core $9/mo annual), HubSpot Breeze (outcome-based pricing effective April 14, 2026: Customer Agent $0.50/50 credits per resolved conversation; credits ~$10/1,000; Service Hub Professional ~$90/seat/mo with 3,000 included credits), and n8n (AI Agent node in all plans; Cloud Starter €20/mo annual; self-hosted free). Affiliate rates re-verified at draft time: Make 35% recurring for 12 months; Zapier has no public affiliate program (partner-only).

What was NOT done: no agent was built or run on any platform; no activities, credits, or resolutions were consumed; no accounts were created. The 500-resolution cost model is arithmetic on published rates and stated allowances — it does not account for activities-per-resolution variance, per-step credit multipliers, or BYOK token costs, all flagged inline. Prices change — confirm on each vendor’s pricing page before you buy.

Sources

Share This Article
Leave a Comment