With the growth of interest in AI agents and the increasing number of practical use cases emerging for agentic workflows, there is a great deal of innovation occurring around the technology. One area that has perhaps taken a backseat so far, however, is how those providing AI agents will monetize agentic services, and also the pricing models customers may come to prefer or expect.
By now, it should be clear that agents are set to turn how organizations sell software on its head. SaaS offerings exist in a clear transactional paradigm in which the service provider, or managed service provider, manages licensing and the delivery of the technology. When it comes to agents, this paradigm shifts away from selling or reselling the software as a whole, to AI consuming exactly what it needs, exactly when it needs to.
Essentially, this means that the service is broken up into smaller units of consumption. Much like we can control the resource consumption of GPUs that are activated or shared by certain workloads, agents can dynamically identify and consume what they need with a given workload.
This shift could be seen as a move away from selling software to selling work, or units of work. If we extrapolate this paradigm to a monetized service, there are a number of options that providers and customers may consider—and it’s not obvious at this stage which model will win out.
End of the Line for SaaS?
If we look at how software development and delivery have evolved over the last couple of decades, the direction of travel makes perfect sense. The age of one-time products that require one-time payments for infinite usage is all but over, having been replaced by SaaS solutions that are maintained, upgraded and updated following a licensing model.
The benefit for customers is clear with this posture. Pay a yearly fee, choose your level of sophistication and the extent of the services your licenses cover, and benefit from largely handing off monitoring, security, maintenance, and updates—and receive assurances around uptime through SLAs, and disaster recovery. Tiered pricing extends this model, enabling services to be scaled for different customer cohorts, such as SMB and enterprise customers, with different feature sets and price points.
Of course, the SaaS model works well for technology vendors, too, as it ensures regular streams of income. However, they must also remain competitive through delivering the aforementioned assurances, as well as improving customer experience to retain business and remain competitive.
Potential Pricing Models for Agents
With the explosion of AI, and subsequently agents, and the meaningful integration of these technologies across systems and workflows in all industries, we are quickly moving towards the arrival of fully agentic systems. To demonstrate how fast things are moving, experts are already predicting that agentic commerce will be a significant part of the e-commerce landscape by 2030.
The shift from selling software to selling work performed by agents opens the door to several possibilities. In my opinion, the following six models are the most likely to emerge:
- Per-Task Completion: This is where customers pay for each successfully completed task. To use a financial services example, a mortgage lender would only be paid once an AI lending agent successfully processes a mortgage.
- Outcome-Based Pricing: For this model, payments are tied to specific business outcomes, such as paying a certain amount for uncovering a sales-qualified lead (SQL) and then more once a sale is made.
- Performance-Based Fees: This is where costs vary depending on quality and efficiency metrics. Think of travel and hospitality scenarios where agents are tasked with delivering bespoke vacation packages against specific criteria. The AI agent—and therefore the agent provider—could have a rate or bonus structure built in that is based on customer satisfaction scores.
- Agent Hours: This model would entail billing for active agent work time, which would likely involve more complex workloads, such as detailed analyses of complex databases or materials.
- Processing Time: Costs are based on computational resources consumed. This model could emerge as the preferred method for businesses that have specific commitments around sustainability goals that are attached to their usage of AI.
- Credit Systems: Prepaid credits are consumed based on the level of complexity of the work.
Regardless of whether one of these pricing models ends up being the de facto standard or we become accustomed to an array of models, pricing is an element that will also need to benefit from innovation in the coming years to support agentic AI systems.
Many vendors are already experimenting with pricing models, but some may be more applicable to certain industries and contexts. For example, complex data analysis would most likely use models 4 or 5, or a hybrid of the two, while a credit system might work well for use cases that have fluctuating requirements over long periods of time.
Working out which model works for customers and adapting to their unique requirements will be the key to succeeding as an agentic AI vendor. Likely, we will also see specialist agentic managed services providers and vendors emerging in the coming years, with some further specializing in specific industries. This is a positive evolution for AI, as we will see the emergence of an agentic AI economy that opens up new job opportunities and avenues of innovation.
