Per-Task Pricing: Examples & Companies

What is it

Per-Task Pricing is a billing unit where customers are charged per task an automation or agent executes — Zapier’s historical unit, now spreading to AI agents. The task is work-shaped: not a request served or a token generated, but a discrete action completed on the user’s behalf, which makes it the unit that most directly answers the buyer’s question “what did I get for the money?”

Zapier defined the genre: every action step a Zap executes consumes one task, plans are sold as monthly task tiers, and the published prices ($19.99/month Professional and up, billed annually) are explicitly the 100-task entry rung of a ladder that climbs with volume. Rows embeds the unit inside seat pricing — its Plus tier bundles 200 AI Tasks and a million cell-enrichment tasks per month into a $8/user seat — and was acquired by Superhuman in February 2026 with the structure intact.

The unit’s frontier is AI agents, and the frontier is contested. Lindy sold agent work through task-credits before pivoting in 2026 to flat $49.99/$99.99/$199.99 subscriptions whose usage differences appear only as tier multipliers — while Mercor and micro1 price task-shaped expert and data work entirely behind sales quotes, with only the payout side (what contributors earn per task) publicly visible.

How it works

Worked example — the definition is the price. A five-step Zap (one trigger, four actions) firing 30 times a day consumes ~3,600 Zapier tasks a month — but the same workflow on a platform that counts one goal-level run per execution would meter 900. Neither vendor is wrong; they’ve defined the unit at different altitudes, and the 4x spread is invisible until you price your actual workflows. This is the sharpest instance of the rule in the usage-metric guide: the unit’s definition, not its rate, is where task quotes diverge.

Worked example — two tasks, 5,000x apart. On Rows Plus, an “AI Task” (an LLM-powered analysis step, 200/month included) and a “cell-enrichment task” (a table-data lookup, 1,000,000/month included) share a name and differ in allowance by 5,000x. The shared label works because each meter is sized to its own cost reality — but it means “tasks included” is meaningless across vendors without the qualifier.

Companies using this

5 in-corpus companies meter tasks: Zapier’s canonical action-step meter, Rows’ bundled task allowances, Lindy’s post-pivot usage tiers, and the quote-gated expert-task marketplaces Mercor and micro1.

Patterns observed

The task meter survives best where tasks are discrete and countable: Zapier’s automations have crisp step boundaries, and Rows’ spreadsheet operations are naturally enumerable — both vendors publish their counting rules and let the tier ladder carry the volume discount. Where the work is ambient and continuous, the unit retreats behind abstractions: Lindy’s multiplier tiers and the entitlement-style allowances that wrap most agent pricing both exist because an always-on assistant’s task burn is unforecastable for the buyer.

The second pattern is the human-AI blur: at Mercor and micro1, the “task” being priced is performed by vetted human experts, AI systems, or both — and buyer-side rates are quoted, not published, because the task mix itself is the negotiation.

Counterexamples & variants

Lindy is the counterexample from inside the cohort: it ran task-credit pricing for its agents and walked it back in 2026, landing on flat subscriptions where usage exists only as tier multipliers — evidence that for ambient AI assistants, the task meter’s precision reads as unpredictability to the buyer. The agent-platform world Lindy left behind is split between that flat model and the per-task conviction Zapier still embodies; which side wins is one of the open questions in agent pricing. And the marketplaces (Mercor, micro1) are less a counterexample than a different genus: the task is real, but it’s a unit of labor procurement rather than software metering, priced per engagement behind a sales process.

What this means for buyers vs vendors

For buyers

Price your three highest-volume workflows, not the rate card: count the action steps the way each vendor counts them, because the task definition moves quotes by 4x before any discount conversation starts. On bundled models (Rows), check which task meter your workload actually hits — the generous-sounding million-task allowance may not be the one your AI usage draws from. And on flat agent plans (Lindy), ask what “3x usage” denominates; a multiplier without a unit is a fair-use policy you haven’t read yet.

For vendors

The task is the most buyer-legible unit in automation — one task, one thing done — so spend that legibility carefully: publish the counting rule with worked examples, keep one definition across endpoints, and put volume discounts in visible tiers the way Zapier does. If your product acts continuously rather than discretely, consider whether you’re selling tasks at all: Lindy’s pivot suggests ambient agents sell better as capacity (flat tiers with allowances) than as itemized work, and a meter your buyer can’t predict is a churn driver regardless of how fair it is.