Ruben, excellent analysis of how background autonomous AI agents are dismantling standard seat licensing workflows. It’s a compelling look at where traditional enterprise software is moving.
However, when you apply this thesis to high-compliance, real-time clinical medical fields—like aesthetic medicine—the paradigm flips completely.
In our industry, an "AI agent" can never sit across from a human patient, analyze dynamic three-dimensional facial architecture, evaluate emotional hesitation, or shoulder clinical medical liability. The software is an active, synchronous utility that must be operated by a licensed human provider right at the point of care.
Furthermore, the alternative models you noted—like usage, outcome, or micro-metered "hybrid" pricing—actually introduce massive structural flaws into a clinical setting. Practice managers and providers look at software as fixed Operational Overhead (OpEx). If you switch them to a per-consultation or outcome-metered model, it feels identically to the highly hated "consumable click-fees" of the laser device industry. It taxes their patient volume and penalizes clinic efficiency.
If a provider converts a single high-ticket treatment plan using software, the infrastructure cost for the entire year is paid for. In high-stakes human environments, a flat, unrestricted premium subscription that eliminates "click-hesitation" and guarantees budgeting predictability will always beat out a micro-metered tax.
Curious to hear your thoughts on how you see these compliance-heavy, 1:1 human workflows resisting the background automation shift!
I think the issue is that good inference isn't cheap (yet). We used to be able to provide software for free because a request would be so light that it wouldn't hurt to let people use it for free. Now, everyone expects apps to have AI integrations or provide some AI features for free, and that really raises the cost per request. And not to mention that agents are thrashing our servers because there are now theoretically more "users" in the world, and they work faster than before.
A software request used to mean a few cheap checks. Now, with AI inference, it can mean running one of the most power-intensive operations in computing.
Interesting seeing this play out outside software too, Tutor Intelligence just raised $75m to rent warehouse robots by the hour or by the pick which is basically outcome pricing for physical work. The catch is the vendor carries all the hardware cost so it only works if the robot earns more per hour than it costs to build, feels like the same question every usage model ends up at
Interesting seeing this play out outside software too, Tutor Intelligence just raised $75m to rent warehouse robots by the hour or by the pick which is basically outcome pricing for physical work. The catch is the vendor carries all the hardware cost so it only works if the robot earns more per hour than it costs to build, feels like the same question every usage model ends up at
The founder question is a super solid point. Name the one thing your product does that someone would pay for on its own. It looks like a pricing question, but if it's hard to answer, the problem usually sits further back.
When you can't name the unit, it's often because the reason the customer should care hasn't been worked out yet. That's a positioning problem. Pricing is often where it shows up first, because it pressure-tests the value prop.
It's why I'd be a bit wary of usage-based as the fallback. It gets you a price, but you still don't know what you want to be picked for, and that gap tends to show up on the homepage and in the sales deck too. By the time a buyer gets to pricing, you should already be the obvious choice.
The point about a vendor earning less precisely because its product gets more efficient really clarified the problem for me. If a seat only measures who logs in, it stops measuring the work once agents do more of it. Of the three alternatives, resolution pricing feels the clearest to explain, because both sides can see the same finished job — though defining what counts as resolved, and knowing the cost of delivering it, seems essential before promising a fixed price. The hybrid route also makes sense as a bridge: a base fee for predictability, with a metered layer for the work actually done.
Ruben, excellent analysis of how background autonomous AI agents are dismantling standard seat licensing workflows. It’s a compelling look at where traditional enterprise software is moving.
However, when you apply this thesis to high-compliance, real-time clinical medical fields—like aesthetic medicine—the paradigm flips completely.
In our industry, an "AI agent" can never sit across from a human patient, analyze dynamic three-dimensional facial architecture, evaluate emotional hesitation, or shoulder clinical medical liability. The software is an active, synchronous utility that must be operated by a licensed human provider right at the point of care.
Furthermore, the alternative models you noted—like usage, outcome, or micro-metered "hybrid" pricing—actually introduce massive structural flaws into a clinical setting. Practice managers and providers look at software as fixed Operational Overhead (OpEx). If you switch them to a per-consultation or outcome-metered model, it feels identically to the highly hated "consumable click-fees" of the laser device industry. It taxes their patient volume and penalizes clinic efficiency.
If a provider converts a single high-ticket treatment plan using software, the infrastructure cost for the entire year is paid for. In high-stakes human environments, a flat, unrestricted premium subscription that eliminates "click-hesitation" and guarantees budgeting predictability will always beat out a micro-metered tax.
Curious to hear your thoughts on how you see these compliance-heavy, 1:1 human workflows resisting the background automation shift!
I think the issue is that good inference isn't cheap (yet). We used to be able to provide software for free because a request would be so light that it wouldn't hurt to let people use it for free. Now, everyone expects apps to have AI integrations or provide some AI features for free, and that really raises the cost per request. And not to mention that agents are thrashing our servers because there are now theoretically more "users" in the world, and they work faster than before.
A software request used to mean a few cheap checks. Now, with AI inference, it can mean running one of the most power-intensive operations in computing.
Interesting seeing this play out outside software too, Tutor Intelligence just raised $75m to rent warehouse robots by the hour or by the pick which is basically outcome pricing for physical work. The catch is the vendor carries all the hardware cost so it only works if the robot earns more per hour than it costs to build, feels like the same question every usage model ends up at
Interesting seeing this play out outside software too, Tutor Intelligence just raised $75m to rent warehouse robots by the hour or by the pick which is basically outcome pricing for physical work. The catch is the vendor carries all the hardware cost so it only works if the robot earns more per hour than it costs to build, feels like the same question every usage model ends up at
The founder question is a super solid point. Name the one thing your product does that someone would pay for on its own. It looks like a pricing question, but if it's hard to answer, the problem usually sits further back.
When you can't name the unit, it's often because the reason the customer should care hasn't been worked out yet. That's a positioning problem. Pricing is often where it shows up first, because it pressure-tests the value prop.
It's why I'd be a bit wary of usage-based as the fallback. It gets you a price, but you still don't know what you want to be picked for, and that gap tends to show up on the homepage and in the sales deck too. By the time a buyer gets to pricing, you should already be the obvious choice.
The point about a vendor earning less precisely because its product gets more efficient really clarified the problem for me. If a seat only measures who logs in, it stops measuring the work once agents do more of it. Of the three alternatives, resolution pricing feels the clearest to explain, because both sides can see the same finished job — though defining what counts as resolved, and knowing the cost of delivering it, seems essential before promising a fixed price. The hybrid route also makes sense as a bridge: a base fee for predictability, with a metered layer for the work actually done.