The Agent Economy -Part IV: Economics — Is SaaS under siege?

Personal opinion and thoughts based on my observations and experience/ will continue to evolve/update this

Part I: https://medium.com/@gopivikranth/the-agent-economy-6a16cecb707d || Part II: https://medium.com/@gopivikranth/the-agent-economy-part-ii-is-it-time-for-enterprises-to-own-their-intelligence-layer-with-a7a6d2ddf590 || Pat III: https://gopivikranth.medium.com/the-agent-economy-part-2-a43b64f5003c

AI agents are likely to disrupt traditional, feature-bloated SaaS (Software-as-a-Service) applications. This shift is due to massive underutiliztion of expensive SaaS seats, the inefficiencies inherent in navigating complicated software stacks, and AI agents matuirng to be able to execute end-to-end tasks.

In such a situation what would be the right pricing mechanism for AI agents? Will it be traditional SaaS or will it be possible to align the cost with actual value delivered based on outcomes and usage?

Since the early 2000s, the SaaS model has dominated the enterprise software market, eventually offering on-demand applications accessible through the cloud. This initially reduced on-premises IT complexity, lower maintenance costs, and improved accessibility.

However, over the past decade, as more vendors entered the space, competition led to rampant feature expansion. SaaS applications frequently became everything to everyone, layering specialized capabilities to differentiate themselves, but often at the expense of user experience and cost-efficiency.

Enterprises now find themselves burdened with sprawling SaaS portfolios with hundreds of applications — many of which are rarely or never used. Per-seat SaaS software has spread in the enterprise, but only a fraction of this is actually utilized. There are estimates that close to 35–40% of enterprise IT spend is going to SaaS with significant under utilization.

A new paradigm is emerging: AI agents. They can understand natural language instructions, execute complex sequences of tasks across different tools, and present results in a user-friendly manner. Instead of logging into half a dozen SaaS dashboards, employees can simply query an AI agent that orchestrates these underlying services invisibly. As business logic moves into AI agents which can connect to multiple systems, tasks can be done by AI agents at a fraction of the cost of traditional SaaS.

A few factors that are in favor of Agentic AI:

1. SaaS Feature Bloat: The first major friction point in the current enterprise software model is feature bloat. Vendors have historically responded to competitive pressures by adding capabilities to justify premium tiers and retain customers. While this created a rich feature landscape, it also led to complexity. The utility of many features is marginal — rarely used except by niche power users.
2. Underutilization of Seats and Licenses: Enterprises often pay annual subscriptions for large volumes of seats. A marketing automation suite might provide 500 seats, of which only 150 are ever actively used. This discrepancy emerges for multiple reasons: over-optimistic initial purchase decisions, staff turnover, and the difficulty of tracking software utilization across distributed teams. The end result is wasted spend and difficulty rationalizing renewal at budget time. According to multiple industry analysts, 30–40% of SaaS seats remain dormant.
3. Data and Integration Silos: SaaS has resulted in creating operational, data and integration silos. They have done the job of delivering software as a service for specific issues but in the process made the problem of silos in enterprise much worse. As agentic AI cuts across multiple service lines, it can solve the silo problem as well.
4. Operational Complexity and Costs: Beyond direct subscription fees, SaaS introduces significant operational overhead. IT teams must manage credentials, vendor risk assessments, integrations, training and compliance. This complexity has reached a tipping point, inviting a simpler, more flexible solution.

AI Agents are different:

1. Contextual Understanding: AI agents leverage LLMs to interpret ambiguous or high-level instructions. Instead of learning complex UIs, the user simply states a goal, and the agent figures out the “how.”

2. On-Demand Access to Functionality: Rather than purchasing static feature sets, AI agents adaptively pull the capabilities needed to accomplish a given task. This reduces both cost and cognitive load for end-users.

3. Reduced Training and Onboarding Costs: By abstracting away complex UIs and integrating multiple tools, AI agents streamline workflows. New employees can be productive without extensive training on multiple SaaS tools.

4. Scalability and Extensibility: As an agent’s capabilities are defined by its underlying integrations and intelligence, adding new functionality becomes a matter of plugging in new APIs. This allows agents to evolve in step with changing business requirements at better cost without overwhelming users with cluttered interfaces.

Pricing and Monetization Models for AI Agents, aligning cost with value: A central advantage of AI agents is the potential to align cost directly with delivered value. Traditional SaaS models charge fixed fees for capacity — such as user seats — irrespective of how much value is realized. This mismatch leads to overpayment and wasted resources.

AI agents can have usage-based or output-based pricing, becoming the opposite of how SaaS used to operate. This is an anathema to existing SaaS models as companies which create AI agents from first principles will have diametrically opposite pricing models that supplant existing SaaS revenues (and agents are cheaper to make and maintain than traditional SaaS).

For example, a call center AI agent can operate at 8–10 dollars an hour, and Devin, a software engineer, although a bit overhyped in its abilities comes at a very affordable $500 per month with a few agent compute units and for additional ACUs you can pay by usage. Hence agent pricing can be consumption based, a few options are:

1. Usage-Based (Consumption) Pricing: Companies pay per token, transaction, or data volume processed by the agent. This ensures cost scales with demand. For instance, OpenAI and many others charge per API call, aligning fees with actual usage.

2. Subscription-Based (small base plans): Some vendors may offer fixed monthly or annual fees for baseline access and rest is consumption based. While this provides predictability, it risks recreating some of the old SaaS pitfalls if its not well-structured.

3. Outcome-Based (Performance Pricing): Linking fees to tangible business outcomes, such as leads generated or artifacts created or tasks completed, can ensure customers pay only for quantifiable value.

In the short term, AI agents will coexist with traditional SaaS applications, augmenting them and helping users perform tasks more efficiently. As trust in AI agents grows and their capabilities improve, enterprises will begin questioning large, static SaaS contracts.

Ultimately, as AI agents become more capable and solve more tasks, they will continuously take over existing SaaS tasks done by applications, eroding the legacy model. This shift mirrors earlier transformations: from on-premises software to SaaS, and now from SaaS to agent-driven orchestration.

AI agents represent not just a new category of software tool but a fundamental reimagining of how enterprises consume software capabilities. Just as the cloud allowed companies to pay for only the infrastructure they need, AI agents let them pay for only the functionality and outcomes they actually consume. The result is a leaner, more efficient, and more intuitive enterprise software environment.