I'm puzzled as to how these organizations are racking up giant bills without realizing it until it's too late. We pay $20/month/user for Standard "team" seats, which have built-in usage limits. Completely predictable and controllable spend. If someone keeps hitting their usage limits, then we either educate them on token efficiency or upgrade them to a $100/month Premium seat. Still completely predictable and controllable. For larger organizations or different use cases, the Enterprise and API plans have spend-limit settings.
Are there really organizations out there who are just giving Anthropic their credit card number and then looking away?
There are different philosophies here, and both are valid depending on what is the goal.
The $20/seat model optimizes for cost control. That's the right tool when AI is a productivity add-on. But when someone is using AI to build something genuinely new (a workflow, an agent, a system) capping tokens is like capping the number of experiments a researcher can run. I've seen people burn 2-3x their salary in API costs, then 3-4 months later ship something that saves the company millions. The tricky part is: during the building phase, output is genuinely hard to measure. You can check some things, but you're also betting on the person.
The risk of the unlimited model isn't overspend, it's junior people burning tokens without a clear direction. So the real question isn't seat-based vs. usage-based. It's: do you trust the person enough to give them the runway?
I'm puzzled as to how these organizations are racking up giant bills without realizing it until it's too late. We pay $20/month/user for Standard "team" seats, which have built-in usage limits. Completely predictable and controllable spend. If someone keeps hitting their usage limits, then we either educate them on token efficiency or upgrade them to a $100/month Premium seat. Still completely predictable and controllable. For larger organizations or different use cases, the Enterprise and API plans have spend-limit settings.
Are there really organizations out there who are just giving Anthropic their credit card number and then looking away?
There are different philosophies here, and both are valid depending on what is the goal.
The $20/seat model optimizes for cost control. That's the right tool when AI is a productivity add-on. But when someone is using AI to build something genuinely new (a workflow, an agent, a system) capping tokens is like capping the number of experiments a researcher can run. I've seen people burn 2-3x their salary in API costs, then 3-4 months later ship something that saves the company millions. The tricky part is: during the building phase, output is genuinely hard to measure. You can check some things, but you're also betting on the person.
The risk of the unlimited model isn't overspend, it's junior people burning tokens without a clear direction. So the real question isn't seat-based vs. usage-based. It's: do you trust the person enough to give them the runway?