Agentic AI: The SaaS Model Facing an Economic Revolution

A Revolution in Progress

The rise of artificial intelligence agents, exemplified by advancements like Gemini 3 and GPT-5.2, is not only pushing technological boundaries but also challenging a decades-old economic model: SaaS (Software as a Service), and more broadly, traditional software. For over 20 years, SaaS has dominated the digital landscape, but agentic AI, capable of autonomously handling up to 60% of customer service requests, is disrupting this dynamic. This shift from software as a tool to software as an autonomous worker signals the end of the subscription-based era.

AI, the Margin Predator

Historically, software required high initial investments for development, but its distribution was nearly cost-free. The advent of generative AI and now agentic AI reverses this economic equation. Inference costs, which refer to the resources needed to run these AIs, heavily burden profit margins. “Inference whales,” those customers who consume vast amounts of computing power without their subscriptions covering these costs, turn publishers into unwitting patrons. Analysts have revealed that OpenAI's inference costs exceeded its revenues during the first half of 2025.

To counter this trend, companies like Anthropic and OpenAI have begun imposing usage limits, whether in terms of hours or number of requests. These measures are necessary because the gross margins of agentic AI are 20 to 30 points lower than those of traditional SaaS.

The Mirage of “Pay-for-Performance”

In the face of these challenges, the sector is exploring the possibility of outcome-based pricing. Companies like Zendesk have already started charging for their AI agents based on the results achieved. OpenAI is even considering collecting royalties on discoveries made through its AI, a sort of new intellectual property. However, this model remains marginal, accounting for less than 10% of the market.

Buyers are hesitant to invest in such an unpredictable monetization system, and the operational complexities it entails hinder publishers. Thus, for the majority, this model remains a theoretical dead end, at least for now.

Towards Fragmentation of Offerings

To survive, the market is diversifying. The concept of “software salary” is emerging, where the agent is viewed as a digital collaborator, as proposed by Nullify. However, charging a flat fee for an agent amounts to reinventing the existing model without considering the computational effort or added value. Publishers risk selling at a loss.

A potential solution lies in activity-based monetization. Companies like Hippocratic AI charge by the hour of agentic care, while Artisan charges per business opportunity generated. This model allows for synchronizing technological costs with operational realities, ushering in the era of Service-as-a-Software.

Embrace Hybridization or Risk Obsolescence

In this climate of uncertainty, one thing is clear: the future belongs to hybrid models. Some customer service companies charge their AI per conversation for certain segments and per resolution for others. Others combine a fixed base for stability with a variable share for added value.

This pricing complexity is now essential to meet the diverse needs of clients. Companies must develop multiple models tailored to their clients' engagement capacity. This strategy must be clear and simple for the end customer.

Time is of the essence. According to the firm RGP, 66% of CFOs expect a measurable impact from AI within two years, but only 14% currently see a significant return on investment. With the European Data Act allowing for a cloud provider switch in 30 days, the grace period is over. Ultimately, AI demonstrates that the best customer retention strategy relies on a cleverly orchestrated value-sharing approach.