Every major software vendor has announced or released an Agentic AI product in the past quarter. They promise to be enterprise-grade alternatives to open-source frameworks and early-stage startups. Salesforce has been among the first vendors to publish a pricing model for their Agentforce product ($2 per conversation). However, a common pricing model or metric has not been established in this nascent market yet. Some vendors have even deferred the pricing details to early in the new year.
But just like major software vendors who are “figuring out” the optimal pricing model for Agentic AI, product managers, IT leaders, and founders embracing this new technology face the same questions. Yet, pricing is a highly complex domain—especially Agentic AI. So much so that I am building an online course around “Pricing Agentic AI Products.” But more on that in a bit…
Putting a Price Tag on Work
AI agents are primed to create higher levels of productivity through higher levels of autonomy and automation. But instead of supporting only parts of a business process, agents enable these efficiency gains at the task level—of any task. In other words, they can take over more complex tasks, reason over the best course of action, and complete them (somewhat) independently without requiring step-by-step instructions.
Yet, technologists are putting a price tag on work—not HR. One might rightfully say, “Technologists always have.” From perpetual licenses to subscriptions, software has always been used to complete tasks and workflows. But agents are different.
Agentic capabilities are more akin to those of human users than to rigid “if-this-then-that”-driven automation. This means that the types of tasks which agents can take over will be more valuable and agents will be more versatile to complete them. Draft an email? An agent can do it. Propose strategies to reduce the volume of open invoices (and implement them)? An agent can do it. Review supplier proposals for alignment with your business requirements? An agent can do it.
Over time, human workers will delegate more and more tasks to AI agents (if the technology becomes robust enough that companies are comfortable rolling it out at scale, and if it makes economic sense). Although the basic assumption is that these digital workers will be able to complete the same task much cheaper than their human counterparts, business, and technology leaders are asking: “How cheap—or how expensive?” And that’s where the dilemma starts.
Because enterprise leaders are looking for predictable costs (operational expenses or OPEX) to better forecast the required budget and track any deviations. Vendors are looking for predictable revenue enabled by predictable delivery costs.
But how do you price agentic AI work?—At least for human labor, there are job families and descriptions, proficiency levels, regional differences, etc…
Thank you for reading The AI MEMO. Feel free to share it with your network.
CREATING A NEW COURSE: “Pricing Agentic AI Products”
I’m working on a new online course and would love to get your feedback. Agentic AI is the big topic in 2025. Hopes are high—and the pressure to make money with it is even higher. But AI agents can create a lot more value for your customers and stakeholders than the cost of the technology itself. So, how should you price your Agentic AI product then? Which components do you need to factor in?
Join me for this course and learn how you can price your Agentic AI products!
At this time, there is no common pricing model that has been adopted across the software industry. What will eventually become the prevailing pricing model remains to be seen. If you are evaluating your options to price your application (for internal or external users), your pricing model will depend on the product capability and where in the tech stack you operate:
Software-as-a-Service (SaaS) vendors typically price based on value. In the case of Agentic AI, this could mean “cheaper than a human but more expensive than just tokens + compute.”
Platform-as-a-Service (PaaS) vendors rather price their products per transaction or resource consumption. This will be a volume play when the value created by an individual agent varies greatly. For example, every transaction or every token costs a fraction of a cent, independent of the business scenario that it enables (and the value it creates).
A variation of this by company size:
Startups can quantify the value of an agent in a smaller niche and will have fewer complexities than large players. Startups are best positioned to offer agents in an outcome-based model given their narrow scope and dedicated focus on one problem or industry. It’s easier to establish or experiment with a new business model. There’s a big potential for doing what Big Tech cannot do.
Big Tech will struggle with pricing agents. The broader the portfolio, the harder and the more complex it will be—for the vendor and the customer. What’s the value of a marketing agent? Do you charge for it by task, by outcome, by transaction, by user, varied by region, …?
AI agents are different from AI features, such as generating job descriptions or drafting email responses in customer service.
Imagine the following example for simplicity's sake: Eventually, your department, yes, even you(!), will have a virtual team of agents that supports you in your work. These agents can be experts in specific domains or on individual tasks. They will have a discussion and deliberation among themselves–about the goal, the subgoals, the best way to reach them, optimizing their response, and presenting the result to you.
But how do you price a task?—And multiple tasks of different complexities, involving LLMs, vector data bases, temporary and long-term memory, etc.?
Between Predictable and Transactional Costs
Human knowledge workers typically receive a monthly salary. The salary is tied to achieving pre-defined goals and objectives in a given time period. But whether an employee is highly productive one day or idle another does not impact pay (or cost to the company). Tasks such as researching information and synthesizing it take more time. Some tasks require more expertise than others, such as preparing a marketing strategy.
Although organizations prefer predictable costs (OPEX), this approach is risky for all involved parties. In Agentic AI, “predictable costs” is the equivalent of a salary: whether or not an agent is idle (and how much of the time) the company will need to pay. If you build Agentic AI products, there are transactional costs involved—for the LLM, vector store, etc. That’s why you will want to limit the risk of overconsumption and your customers creating more costs than they pay you in revenue or via cross-charging (between departments).
On the other hand, Agentic AI is an opportunity for software vendors to create a new revenue stream at the same time as the number of “per-user” licenses will decrease due to AI taking over more tasks. An important risk is cannibalizing the existing business by introducing Agentic AI in parallel to subscription-based SaaS products. Hence, it’s important to find a vehicle to compensate for that future revenue drop early on.
Lastly, a purely transaction-based pricing model will be hard to sell to a business buyer who is looking for outcomes when the transaction volume might fluctuate during the year. Imagine a sales rep asking your prospect: “How many discussions between your (multi-)agents do you think there will be?” (Yeah, good luck with that!)
What should your pricing model look like then?—And will it be set in stone?
Initial Pricing and Evolution
Whatever your initial pricing model will look like, it will most likely change over time. You will need to create one along with the first release of your agentic products and platforms. This pricing model won’t be perfect. You will need to get more feedback from customers to dial in the right model, and neither vendor nor customer will initially know how much customers will consume (even with beta and pilot programs). It takes time.
Eventually, the industry might gravitate towards a freemium model in which a number of agentic transactions are included in a subscription price. When an organization exceeds this limit, they need to purchase additional ‘credits’ to continue using the agent that billing period. This will also be the point when vendors increase their subscription prices or introduce new tiers.
Changing pricing models will also create a challenge for vendors who close deals on the initial pricing model and who need to honor that pricing for the duration of the contract. However, vendors won’t change their pricing model every quarter. It’s a huge effort and it takes education among sales teams and customers alike. So, expect one, maybe two pivots.
Summary
Agentic AI is another basic technology that allows organizations to automation tasks with a higher degree of autonomy and without defining the exact steps for reaching a given objective. The types of tasks can vary greatly in duration, complexity, and business value. This creates highly variable costs for vendors and IT organizations who develop and provide these new Agentic AI product.
While HR defines job descriptions, skill levels, requirements, and salary in the real world, technologists define the price tag of Agentic AI capabilities. Yet, between PaaS- and SaaS-type offerings, the price points and metrics typically vary as well.
Agentic AI promises to unlock new revenue streams for companies—especially, as transacation- and value-based pricing evolves toward outcome-based pricing. Whatever pricing model you start with, it will likely evolve over time with additional customer feedback and real-world consumption data.
Interested in joining my upcoming course to learn how you can price Agentic AI products? “Pricing Agentic AI Products” is ideal for product managers, IT leaders, and founders.
Join my bi-weekly live stream and podcast for leaders and hands-on practitioners. Each episode features a different guest who shares their AI journey and actionable insights. Learn from your peers how you can lead artificial intelligence, generative AI & automation in business with confidence.
When companies start to see the mounting costs, they might feel the pinch, and things might readjust. For example, during the mass migration to Cloud services, companies sometimes experienced sticker shock and needed to know how to scale and scope their solutions. I am seeing a similar bubble here.
When companies start to see the mounting costs, they might feel the pinch, and things might readjust. For example, during the mass migration to Cloud services, companies sometimes experienced sticker shock and needed to know how to scale and scope their solutions. I am seeing a similar bubble here.