Hey AI Agent Developer – yes, you. What is the value you create? Let’s find out.

The Skeptic’s Take: Useful vs. Intelligent

On the spectrum of “Is AI intelligent and able to reason?”, I fall on the skeptical side. But is it useful? Absolutely. We have built a few AI workflows, and they have unlocked superior automation. Evidently, the world is currently witnessing the transformative potential of AI agents in customer support and coding, to some degree.

Take customer support, for instance. If AI agents help each support employee handle 30% more tickets, that’s like adding 30 new hires to a 100-person team, without the cost. For a $100M company, this efficiency gain could translate to $20-30M in additional enterprise value (EV), under sweeping assumptions. The difference in EV between pre- and post-AI agents is the Value you create. Alternatively, one could also use operating income for a flow perspective.

Two Critical Questions

‘Digital worker’ is, in fact, a fitting neologism for the AI agent. I take the view here that human workers, together with digital workers, can do more, creating abundance. As AI agent developers, we should be asking:

• What is our share in the value we create, and how should we price AI agents?
• Why do legacy billing systems prove inadequate?

Outcomes Matter. Internals Don’t.

There are two parts to our first question: First, what an AI agent or a workflow does internally, for instance, the number of reasoning steps involved, is irrelevant. However, measuring outcomes such as the number of support tickets resolved or the number of successful hires is directly proportional to the value we want to measure. Second, what share do we take? Simply going by the 10x rule of SaaS pricing guidelines, we can claim one-tenth of what we measured. Thus, measuring outcomes becomes imperative. We have built Valmi to ingest outcomes as the first step. The added complexity that Valmi addresses is where legacy billing systems sputter.

Why Legacy Billing Systems Don’t Cut It

Let’s start with the cost argument. Building AI systems is expensive. There are ineffective paths in workflows and autonomous agents that simply fail. LLM costs hit COGS linearly, whereas the marginal cost in traditional SaaS systems is negligible. SaaS generates ~70% gross margins. While cost is an important element in the case of AI agents, legacy systems, such as Stripe and Zuora, are inadequate to capture it.

Secondly, the pricing model. The seat-based model works against selling AI agents, since the seat-based model is under attack from two forces. One is the decrease in the number of human workers required. By pricing against seats, you are setting your AI agents up for decline, not for growth. The other is worse: It does not capture the human-equivalent value, as discussed above. Even the usage-based models that legacy systems support do not distinguish between activity and outcomes.

Do It Yourself? No Need.

To effectively price AI agents, we need to measure outcomes, and observe and allocate costs. To evaluate margin contraction, we also need to understand where outcomes and costs diverge, including in aggregate, for which agents and for which customer instances. Therefore, the top AI agent developers who use outcome-billing, such as Harvey, Sierra and Usepropane, put together two sources: one to track cost information and the other to measure value. And they unified these two sources of information in a single system. Valmi helps you do that very easily. You can bring up the whole billing infrastructure on your premises. We have made it available under a permissive license. No need to do it yourself.

The Unreliability Problem (And How to Sell Despite It)

Outcome measurement serves a greater purpose. AI is unreliable and not deterministic. We simply do not understand when it succeeds and when it fails. Consequently, it is incumbent upon the buyer of your AI agents to demand proof of value. It is simply easier for you to convince and onboard buyers if you switch to outcome-billing. Why? You assume the risk of failure of your AI agents. But, the buyer simply pays for agent’s performance and pays nothing for its failure. To prove the value and help you convince your buyers, Valmi supports customer dashboards that show outcomes such as ticket resolution percentage. Your buyers can view and embed these dashboards in their workflows.

The Bottom Line

To sum it up, we have built Valmi, and made available open-source SDKs and free to deploy packages, to address the imperative for your AI agents. It solves outcome-billing that provides proof of value and cost tracking that exposes margin contraction. It lets you simulate and set prices for your AI agents and quickly onboard your customers. We understand hybrid models, combining seat-based, activity and outcomes, will be required in the transitory phase of billing models. Valmi supports all of these as well. Try it out for free or deploy it within your premises.