Quick question: how many seats does an agent need?
Zero. That's the whole problem.
Per-seat pricing was never really about value — it was a proxy for value. More people using the tool usually meant more value being created, so charging per head was close enough. Nobody loved it, but it worked, because for twenty years "using software" and "a human sitting at a screen" were basically the same sentence.
Agents break that sentence in half.
The math nobody's running
Say you sell a support tool at $79/seat/month. A team of 12 pays you roughly $950 a month. Now you ship an agent feature that resolves 60% of those tickets without a human touching them. Congratulations, you just built a product that quietly justifies cutting your own invoice.
The customer notices before you do. They don't need 12 seats anymore — they need 5 people to handle the exceptions and an agent to handle the rest. Your ARPU doesn't grow with the value you're creating. It shrinks with the headcount you're helping them cut.
Every seat you sell an agent to replace is a seat you'll have to un-sell later. You're financing your own downgrade.
Why usage-based isn't the fix either
The obvious next move is "charge per API call" or "charge per task run." Better, but still not it — usage and value aren't the same thing. An agent that runs 10,000 cheap, low-stakes tasks isn't obviously worth more than one that runs 50 tasks that each saved a client six figures. Pure usage pricing rewards chattiness, not outcomes.
What actually survives contact with an agent
Outcome-based pricing — charging for the result, not the access or the compute — is the only model that scales in the same direction as the value the agent creates. More resolved tickets, more qualified leads, more closed loops: more revenue for you, proportional to what the customer actually got.
It's harder to build. You need a real definition of "outcome," a way to measure it without the customer arguing every invoice, and pricing logic that doesn't collapse the first time an edge case shows up. Worth it anyway.
Same old SaaS pricing, same old tricks, won't survive agents that don't need seats to begin with.