Who’s Using Who? Software Costs and Pricing in the age of Agentic AI

Summary

The rapid deployment of Agentic AI is having game changing impacts across multiple industries. One under-examined aspect is the impact of Agentic use on (a) the broader costs of using technology (might an Agent rack up massive costs on your behalf?) and (b) the pricing models of software vendors – particularly vendors who price software on a per-seat or per-user model. This article considers some of the issues we are seeing, steps users of Agentic solutions should take to manage the impact of such solutions and where the market might go next.

Introduction

As Agentic AI moves rapidly from proof of concept stage to mainstream the conversations around how that usage will be priced are many and varied – will it be output based, based on “value add” or consumption. The pricing models are as many and varied as the use cases.

In all this debate, an area that has been somewhat overlooked is the impact of Agentic AI on traditional enterprise software pricing. In other words, it’s all very well pricing up your use of an Agent but how your Agent then uses your services and systems may in turn have a cost impact.

Total Cost of Ownership?

As an Agent navigates its way around your systems care should be taken to understand the costs that may arise. Of course, this is particularly important if you are paying for the Agent itself on a usage based model, but you should also consider the other costs the Agent may rack up.

If an Agent is using your own computer systems the costs of the compute power consumed by the Agent as it interrogates your system could rise exponentially compared to traditional human users as the number of interactions and queries an Agent can make will be many times higher than traditional users which can (and already does) lead to unexpectedly high charges arising from the Agent’s use of your systems.

Traditional pricing models disrupted?

If you license software or access services, there are a number of pricing models ranging from “all you can eat” enterprise-wide licence models to consumption based models linked to the number of transactions processed or services consumed. In many traditional SaaS and even on-premise software contracts however, the pricing is based on “seats” or “end-users”.

In the consumption-based models, as noted above, care should be taken to consider the total cost impact of deploying an Agent and guardrails on usage should be implemented in order to manage costs. Beyond that, you should review your contracts to see if there are restrictions on Agentic use.

Consider an enterprise-wide licence of a SaaS solution (i.e. a fixed annual price to use a solution as much as you need). As Agents are deployed to use those systems, usage could shoot up resulting in higher compute costs for the SaaS provider and a level of benefit being obtained by the user that was not factored in when the original enterprise-wide price was agreed.

In per-seat pricing models the position is potentially even more fraught. One Agent accessing a system via an API may effectively be treated as a single user notwithstanding it can do the work of ten, twenty or even a hundred traditional users. While on the one hand this could hugely disrupt SaaS vendors’ business models (so called “per-seat leakage”), on the other hand you can expect SaaS vendors to carefully review their terms and monitor usage and seek to charge for Agentic use in a different way to that they have used for traditional human per-seat users.

What rights of use does an Agent have?

A key question when deploying an Agent to use a software tool (beyond matters like security and regulatory compliance of course) is: are you allowed to do so?

• Many software licences already have clear restrictions on machine use of a tool. This can be in the Acceptable Use Policy or a straightforward prohibition.
• Others have clear statements that machine use (in older contracts sometimes called robot process automation or RPA) is allowed but is subject to separate pricing.
• Even licences that do not explicitly ban non-human use either require users to be “named” or have definitions of users that link to “persons” or other terminology that, if tested in court, may be considered to exclude Agentic use.

If Agentic use is explicitly prohibited, then of course a discussion with the vendor will be needed to agree on a right of use (and related pricing). Where Agentic use is not expressly prohibited and the contractual model is enterprise-wide or consumption-based then you can probably proceed with limited risk. Where however the agreement does not prohibit Agentic use and instead pricing is linked to users, care will be needed to assess the exact wording around that usage to assess what is permitted and what is not.

Even if there is no clear prohibition on Agentic use, we anticipate that software vendors will seek comfort via software audits or SaaS system monitoring that Agentic use is not taking place at scale or, if it is, it is being paid for. We also anticipate these debates being opened up when software licences come up for renewal as software vendors seek to fill gaps in their legacy terms to ensure they are covered for future Agentic use.

Unintended consequences

In considering these issues, the case of SAP v Diageo may come back to haunt software users. Very briefly: in that case, Diageo implementing a solution to allow multiple employees to input data directly into a system that previously only a limited number of super-users could access was found to have made Diageo liable to pay the relatively high per-user price payable for the super-users for each and every employee who accessed the system.

While many commentators felt the court erred in its decision making, the case was not appealed and remains the leading decision on the implications of indirect use of software priced on a per-user basis. Where an Agent is accessed by (or fed data by) multiple end-users but software is priced on a per-user basis, one can imagine some software vendors looking to deploy the arguments from SAP v Diageo to increase their revenues.

Conclusion

Clearly, every article about the deployment of AI systems stresses the importance of ensuring that appropriate guardrails are put in place to ensure that use is lawful, safe and delivers the intended benefits. In the context of pricing and cost, we would stress the importance of considering how an Agentic AI deployment may impact your other costs – be they compute costs, energy costs or licence fees – as part of that broader analysis.