You sat down to price your first AI agent. You opened a spreadsheet, looked at competitors, and instinctively wrote down some version of "€49 per user per month." Then you realized your agent doesn't have users. It has a Slack channel and a Zapier webhook and it runs at 3am whether anyone is at their desk or not.

That instinct — the seat-based reflex — is what's about to leave most agent builders charging a fraction of what their work is worth. Bessemer's data shows seat-based pricing dropped from 21% to 15% of SaaS companies in just twelve months. Hybrid models jumped from 27% to 41%. The market is voting. The vote is against seats.

The question isn't whether per-seat is dying. The question is what to replace it with — and how to do it on the agent you're shipping this month, not the one you'll perfect next year.

Why Seats Stopped Mapping to Value

Per-seat pricing was a clever proxy for a long time. More seats meant more usage meant more value. SaaS pricing teams could ignore the messy question of what the customer actually got and just count humans. That stopped working the second agents arrived.

Agents don't log in. They authenticate through service accounts, MCP servers, or API tokens. A single agent can do the work of fifteen humans without occupying a single seat in the systems it operates. If you price by seats, you're pricing the empty chairs left behind.

Agents compress headcount. MindStudio's analysis pegs the compression rate at 90% — meaning an agent replacing a team of ten leaves one human at the helm. Per-seat math leaves 67% of the captured value on the table.

Buyers know. SMB owners who have spent twelve months watching Salesforce, HubSpot, and Zendesk shift their own agent products to outcome-based pricing have learned the new vocabulary. When you walk in with a per-seat quote, you sound like a SaaS vendor from 2022.

The Three Pricing Models That Actually Work

The agent pricing landscape settled on three viable models in 2026. Each has a sweet spot. Picking the wrong one isn't just lost revenue — it's a churn liability the day a competitor picks the right one.

Per-seat (€30–€80 / agent / month). Still defensible when your agent is genuinely a teammate — a sales rep, a marketer, a researcher — used by a small number of named humans who treat it as a personal tool. Fine for solo-founder buyers. Falls apart at enterprise.

Per-task or per-ticket (€0.30–€1.00 / event). Best when volume is the value. Inbound support, lead enrichment, document processing. Customers love it because the cost scales with their business; you love it because volume is sticky.

Per-resolution or per-outcome (€0.50–€2.00 / resolved event). Where the market is going. You only get paid when the agent actually finishes the job. HubSpot's Customer Agent moved to €0.50 per resolved conversation in April 2026. Zendesk sits around €1.50. The pricing aligns incentives perfectly — you're paid to be good.

Where Each Model Wins in Practice

Sales agents. Per-seat usually wins. A sales agent is closer to a person than a process. Charge €79/agent/month, cap the number of agents at the company size, and let it ride.

Support agents. Per-resolution. Always. Customers compare you to humans at €15/hour and to other agents at €1/resolution. Anything other than per-resolution looks insane on a procurement deck.

Marketing or content agents. Hybrid. A flat platform fee (€199/month) plus a per-output charge (€0.10 per draft, €5 per published asset). The flat fee anchors retention; the variable fee captures upside.

Operations / RPA-style agents. Per-task. These agents do high-volume, low-margin work — invoice reconciliation, data entry, inventory updates. €0.20 per invoice processed scales beautifully and is impossible for a customer to dispute when your agent is reading the actual document.

How to Set the Price Without Guessing

The mistake most indie builders make is picking the model first and the number second. Reverse it. Pick the number first.

Find the human baseline. Whoever you're replacing — a VA at €25/hour, a support agent at €18/hour, an analyst at €60/hour — that's your ceiling. You're not allowed to price above what the human costs, because the buyer will simply hire the human.

Aim for one-third of the baseline. Anchor at 30–40% of the human cost. This is the magic zone where the buyer feels savings (loss aversion) but you don't leave money on the table. If a human SDR enriches 500 leads/month for €2,500, your enrichment agent at €600–€900 looks like a no-brainer.

Then back into the unit. Once you have the monthly target, divide by the natural unit of work. 500 leads at €700/month = €1.40 per enriched lead. Now you have your per-task price and your monthly anchor, and you can present either depending on how the prospect is buying.

Always include a floor. Even per-resolution models need a minimum monthly commitment. €99/month minimum protects you from the customer who runs your agent twice and ghosts.

The Pricing Decision That Compounds

Operators who get pricing wrong don't just lose margin on the current deal — they create a ceiling on the business. A per-seat agent priced at €49 caps your account-level revenue at the customer's headcount. A per-resolution agent at €1.50 has no ceiling: as the customer grows, you grow with them. As they automate more, you earn more. Same product. Different gravity.

The agents that will compound into real businesses over the next twenty-four months are the ones priced like the volume they enable, not like the humans they replace. Pick the model that scales with your customer's success, then ship the smallest possible agent that proves you can deliver on it. The seat is over.