From SaaS to Agents: How Software Economics Are Quietly Reshaping

The Shift Nobody's Talking About Clearly

Your finance team has always understood software costs in one way: per seat, per month. A Salesforce license costs $X per user. Slack costs $Y per person. The math is simple, predictable, and wrong for what's coming next.

Over the next five years, the software industry is undergoing a fundamental shift in what it sells and how it prices it. This isn't speculation. It's already happening. And if you're evaluating AI tools for your organization, you need to understand it—because the pricing models, risk structures, and economic assumptions you've relied on for decades no longer apply.

According to Goldman Sachs research, the addressable market for software is reshaping dramatically. By 2030, the profit pool is expected to shift significantly from traditional SaaS toward agentic systems—software that doesn't just help people work, but completes work outcomes end-to-end. This shift changes three critical things: what you're actually buying, how you'll pay for it, and who bears the risk when something goes wrong.

From Counting Users to Counting Outcomes

Traditional SaaS pricing starts with a simple unit: the user. You pay per seat because the software helps that person be more productive. One person, one license. Ten people, ten licenses. The math scales linearly with headcount.

But agentic AI breaks this model entirely.

Consider a customer support system. Under the old SaaS model, you might buy a helpdesk platform and pay per agent—say, $100 per user per month. If you have 20 agents, you pay $2,000 per month. If you hire 30 agents, you pay $3,000. The cost scales with headcount.

Now consider an AI agent that resolves support tickets end-to-end. It doesn't need a "seat." It doesn't care how many humans work in your support department. What matters is: how many tickets does it resolve? How many customer issues does it close without human intervention? The pricing unit shifts from people to outcomes.

This isn't a minor accounting change. It's a fundamental reframing of what software does and how its value is measured.

The same pattern appears across industries:

• Document processing: Instead of paying per user of a document management system, you pay per document processed or per page extracted.
• Sales development: Instead of paying per SDR seat, you pay per qualified lead generated or per outbound sequence completed.
• Legal research: Instead of paying per attorney using a legal research tool, you pay per research task completed or per contract reviewed.
• Financial analysis: Instead of paying per analyst, you pay per financial model built or per forecast generated.

In each case, the software is moving from augmenting human work to replacing labor or expertise entirely. And when software replaces labor, the pricing model must change. You can't charge per seat for something that doesn't need a seat.

Why Per-Seat Pricing Fails for Agents

The economic logic is straightforward: per-seat pricing assumes that each user generates roughly the same value. Buy 10 licenses, get 10x the productivity. It's linear.

But agentic systems don't scale linearly with users. They scale with work completed. A single AI agent might resolve 100 support tickets per day. Another might resolve 500. The difference isn't the number of seats—it's the complexity of the work, the quality of the system, and the outcomes it achieves.

When value decouples from headcount, per-seat pricing becomes economically nonsensical for both buyer and vendor.

For buyers: You're paying for seats you don't need. If an AI agent handles 80% of your support tickets, why pay for 20 full-time support agents? You're overpaying for capacity you don't use.

For vendors: They're leaving money on the table. If their agent is resolving 10,000 tickets per month and generating $500,000 in value, but they're only charging per seat, they're capturing a fraction of the value they create.

This is why vendors are shifting to outcome-based pricing. It aligns incentives: the vendor gets paid when they deliver results, and the buyer only pays for what they actually use.

Consider a real-world analogy: shipping and logistics. Decades ago, companies paid logistics providers per employee—per driver, per warehouse worker. Today, nobody does that. You pay per shipment delivered, per ton moved, per mile traveled. The pricing unit is the outcome, not the person. The software and automation industry is moving toward the same model.

The Quiet Reshaping of Software Economics

What's happening isn't that software is becoming more efficient. It's that software is absorbing functions that used to be services businesses.

A services business—a consulting firm, a law firm, an accounting practice—operates on a simple model: you hire experts, they do the work, you bill the client. The vendor bears the risk: if the work is poor, the client doesn't pay. If the expert is slow, the vendor loses money. Revenue is tied to outcomes delivered, not hours billed or people employed.

Agentic AI is bringing this model into software. Instead of paying for a tool that helps your team work faster, you're paying for a system that completes the work. The vendor bears more of the performance risk. If the AI agent fails to resolve a ticket, you don't pay. If it processes documents incorrectly, the vendor absorbs the cost of rework.

This shift expands what software can do—and who can access it.

Previously, if you needed specialized expertise—legal research, financial modeling, technical writing—you either hired expensive experts or you didn't do the work. The cost was prohibitive. Now, agentic AI allows organizations to access capabilities they could never have staffed or afforded. A 10-person startup can now have the equivalent of a specialized research team. A mid-market company can automate work that previously required hiring consultants.

But this expansion comes with a catch: the risk structure is different. When you hire a consultant, you know what you're getting—a person with credentials and experience. When you buy an AI agent, you're buying a system that might fail in ways you don't expect. The vendor's incentive is to deliver outcomes, but the buyer's responsibility is to verify that those outcomes are actually correct.

What Buyers Must Re-Learn

If you've spent the last decade evaluating software based on per-seat pricing, cost per user, and simple ROI calculations, you need to update your mental model.

Here's what changes:

Pricing transparency becomes critical. With per-seat pricing, the math is obvious: 10 users × $100/month = $1,000/month. With outcome-based pricing, you need to understand the cost curve. How much does each additional outcome cost? Does the price per outcome decrease as volume increases? What happens at scale? These questions matter because your total cost is no longer predictable based on headcount.

Performance risk shifts. With traditional SaaS, if the software doesn't work well, you waste time—your team's time. With agentic systems, if the software doesn't work well, you waste money—the vendor's money, if they're pricing based on outcomes. This means you should scrutinize the vendor's performance guarantees, error rates, and what happens when the system fails. "Cheap per seat" can hide expensive failures.

Outcome pricing requires closer scrutiny, not blind trust. Outcome-based pricing can be more economical than per-seat pricing, but only if the outcomes are clearly defined and measured. What counts as a "resolved ticket"? What's the quality threshold? If the vendor has flexibility in how they measure outcomes, they can inflate the numbers. You need contractual clarity on what you're actually paying for.

Capability expansion, not just efficiency, is the real story. Don't evaluate agentic AI solely on whether it's cheaper than hiring people. Evaluate it on whether it lets you do work you couldn't do before. Can you now process 100,000 documents per month instead of 10,000? Can you now research 50 potential customers per week instead of 5? The value isn't just cost reduction—it's capability expansion.

The New Economics of Software

The shift from SaaS to agentic AI isn't about technology getting smarter. It's about economics. When software moves from augmenting work to completing work, the pricing model, risk structure, and buyer evaluation criteria all change.

This doesn't mean per-seat SaaS is going away. It means the profit pool is shifting. The highest-margin, fastest-growing software businesses will increasingly be those that price based on outcomes, not seats. And buyers who understand this shift—who know how to evaluate outcome-based pricing, who understand where the performance risk lies, who can measure whether they're actually getting the capability expansion they're paying for—will make smarter decisions.

The Goldman Sachs data shows this shift is already underway. The question for your organization isn't whether this will happen. It's whether you're ready to evaluate software differently when it does.