Every piece of enterprise software on the planet was designed with a single assumption baked into it: a human being will evaluate it, purchase it, and use it.
That assumption is about to break.
Gartner predicts that by 2028, 90% of B2B purchases will be intermediated by AI agents, pushing more than $15 trillion in spending through agent-to-agent exchanges. Not influenced by agents. Not recommended by agents. Intermediated by agents – meaning autonomous systems will evaluate vendors, negotiate terms, execute transactions, and manage renewals with minimal human oversight.
This isn’t a theoretical future. Salesforce has already split its pricing into two tracks: per-user products for humans and consumption-based products for agents. ServiceNow launched a separate AI credit system. Procurement platforms like Keelvar deploy autonomous bots that launch RFQs, collect bids, and recommend awards without a human touching the process. The infrastructure for agent-driven commerce is being built right now.
For executives, investors, and practitioners, this shift demands a fundamental rethinking of how software is built, priced, sold, and governed. Here’s what that looks like in practice.
The Agent-as-Employee Mental Model
The most useful way to think about where we’re heading is this: an autonomous agent becomes an employee.
You give it credentials. You give it an identity within your systems. You assign it a budget, a scope, and decision-making authority. Then it operates. (This is exactly the paradigm that Charlie Bell, Microsoft’s EVP of Security, explored on the Talking AI podcast – what happens when agents carry real credentials and real authority inside your enterprise.) It evaluates tools. It provisions resources. It negotiates contracts. It spends money.
And just like you would with any employee, you review what it did. What did it buy? What did it spend? What did it accomplish? The management layer doesn’t disappear – it shifts from approving every transaction to reviewing outcomes and adjusting guardrails.
This mental model matters because it changes the entire enterprise relationship with software vendors. The buyer isn’t browsing your marketing site, reading case studies, or sitting through a demo. The buyer is an API call with a set of evaluation criteria, authorized spending limits, and the ability to switch vendors in seconds.
Why Seat-Based Pricing Is Structurally Broken
For two decades, enterprise software economics have been elegantly simple: more employees equals more licenses equals more revenue. The per-seat model aligned incentives neatly – customers could forecast spend based on hiring plans, vendors could predict revenue based on seat expansion, and investors could model growth based on net dollar retention.
AI agents shatter this alignment in three ways.
First, agents reduce the number of humans needed to perform a given workflow, directly eroding the seat count that drives revenue. When Navan reports that AI agents now handle roughly half of all complex travel resolutions, that’s not an efficiency story – it’s a seat-reduction story. The better AI performs within a product, the fewer seats customers need, creating the perverse dynamic where product improvement erodes revenue.
Second, agents generate wildly variable consumption patterns. A single AI workflow might make thousands of API calls in a day, while another makes none. Usage doesn’t correlate with headcount anymore – it correlates with task complexity, automation depth, and workflow frequency. Per-seat pricing can’t capture this variability.
Third, when an agent is the buyer, it doesn’t care about per-seat pricing at all. An agent evaluating software options will optimize for capability per unit of cost, API accessibility, and consumption flexibility. A rigid seat-based model becomes a competitive disadvantage because it’s illegible to the buyer.
The data reflects this shift. Seat-based pricing dropped from 21% to 15% of SaaS pricing models in a single year, with companies sticking to pure seat-based models experiencing significantly higher churn. By 2025, 85% of software companies had adopted some form of usage-based pricing. The migration is well underway.
The Rise of Consumption-Based and Outcome-Based Models
The replacement model isn’t a mystery. It’s consumption-based pricing – and more specifically, a hybrid approach that combines baseline subscriptions with metered AI usage.
Here’s how it works in practice. The agent has authorization to use a given tool or platform. It starts consuming – compute, storage, API calls, workflow executions. The pricing scales with what the agent actually uses. When the agent hits a budget threshold, it surfaces the decision to a human: “We’ve reached our consumption limit with this vendor. Here’s what we’ve accomplished and what we’d need to continue.”
This is already playing out across the market. Snowflake pioneered pure consumption pricing across storage and compute. OpenAI and Anthropic use token-based billing with distinct input and output rates. Salesforce’s Agentforce is available through multiple pricing models – pay-per-action, Flex Credits, or traditional per-user licensing. Microsoft and Atlassian bundle AI usage credits alongside seats. The market is converging on hybrid models that give customers a predictable baseline with elastic scaling for AI workloads.
McKinsey’s research highlights a critical nuance: the best consumption metrics are those tied to business value generated rather than resources expended. An AI sales development agent priced per qualified lead is more aligned with customer value than one priced per thousand emails sent. The former charges for outcomes; the latter charges for effort. As the market matures, the pricing advantage will go to vendors who can credibly tie their metrics to business results.
For CFOs, this shift introduces real complexity. Software spend now behaves more like a utility bill than a predictable subscription renewal. Invoices fluctuate with experimentation cycles, model retraining, and automation adoption. Finance teams accustomed to forecasting software costs based on headcount must build new muscles around usage monitoring, threshold management, and dynamic budgeting.
What Changes When You Sell to Agents, Not People
This is the question that should keep every product leader, CRO, and CMO awake at night: how does your entire go-to-market change when the buyer is a machine?
Product architecture shifts from UI-first to API-first
Agents don’t care about your dashboard. They don’t appreciate clever onboarding flows or thoughtful empty states. They need structured APIs, clean documentation, machine-readable product catalogs, and composable microservices they can integrate into automated workflows. Products built with composable, API-first, cloud-native, headless architectures will have a structural competitive advantage because they’re legible to agent buyers. Products that depend on human UX for differentiation will become invisible to the fastest-growing buying channel.
Marketing shifts from brand storytelling to data trust
In an agent-mediated buying environment, verifiable operational data becomes the currency. Agents will evaluate vendors based on structured performance data, uptime metrics, SLA compliance, and integration compatibility – not on brand perception or content marketing. Digital trust frameworks and data provenance become prerequisites for even being considered. The marketing function doesn’t disappear, but its primary audience shifts from human decision-makers to the systems that inform and execute decisions on their behalf.
Sales cycles compress and transform
When an autonomous sourcing agent can evaluate multiple vendors simultaneously, run comparisons against structured criteria, and execute a purchase decision in hours rather than quarters, the traditional enterprise sales cycle – with its discovery calls, POCs, executive sponsorships, and procurement reviews – gets radically compressed. Sales teams will need to focus on the much smaller number of genuinely strategic decisions that still require human judgment, while ensuring their products are positioned to win in agent-driven evaluations for everything else.
Contract structures evolve
Static three-to-five year enterprise license agreements don’t fit a world where consumption is variable and switching costs are low. Enterprise agreements should be reevaluated every six to twelve months to adjust based on actual usage. “True forward” mechanisms – where committed spend adjusts based on prior-period consumption – are emerging as a middle ground between predictability and flexibility.
The Governance Challenge: Managing Agents Like Employees
If agents are spending company money, companies need governance frameworks that treat them like employees – with clear authorities, spending limits, audit trails, and performance reviews.
This is more than a theoretical concern. When an autonomous procurement agent can negotiate contracts, execute purchases, and commit organizational resources, the governance implications are significant. Consider the questions every enterprise will need to answer:
What spending authority does each agent have, and who sets those limits? How do you audit agent decisions to ensure they align with organizational strategy, not just optimize for the narrowest cost metric? What happens when an agent’s decision conflicts with a supplier relationship that matters for reasons an algorithm can’t easily quantify – like a strategic partnership or a diversity commitment? How do you prevent rogue agent behavior, where an agent optimizes for its defined objective in ways that produce unintended consequences? (For a deeper dive into the security dimensions of this challenge, including Zero Trust frameworks applied to AI agents, see our breakdown of AI agent security and the “double agent” problem.)
Gartner’s prediction that 22% of monetary transactions will be programmable by 2030 – with embedded terms, conditions, and authorization logic – suggests the financial infrastructure for agent governance is already being built. But the organizational policies and management practices around agent governance are still in their infancy.
The companies that build robust agent governance frameworks early will have a significant advantage. This isn’t just about risk mitigation – it’s about being able to deploy agents more aggressively because you trust the guardrails.
The Competitive Moat Shifts from Features to Automation Depth
Here’s the strategic implication that matters most for investors and executives: when agents are the buyer, the competitive moat shifts.
In a human-buyer world, moats were built on brand, UX, integration ecosystem, and switching costs. In an agent-buyer world, moats are built on automation depth, data quality, API composability, and outcome measurability. Software vendors will no longer be valued primarily on feature breadth or seat penetration. They’ll be valued on how deeply their automation integrates into customer workflows and how effectively they capture a share of the productivity gains they generate.
This has direct implications for portfolio strategy. If you’re an investor, the question for every software company in your portfolio becomes: in the agent-mediated future, is this product something agents will consume, or something agents will route around? Is the data structured and accessible? Is the pricing model consumption-friendly? Is the architecture API-first?
If the answer to those questions is no, you’re holding a company that’s optimized for a buying motion that’s about to become obsolete.
What to Do Now: A Practical Framework
The shift to agent-mediated commerce won’t happen overnight, but the leaders are already moving. Here’s a practical framework for how to prepare.
Audit your product for agent readiness. Can an autonomous system discover, evaluate, provision, and consume your product entirely through APIs? If any step in your customer journey requires a human to navigate a UI, fill out a form, or sit through a demo, that step becomes a bottleneck in an agent-driven world. Map your customer journey from discovery to renewal and identify every point where human interaction is currently required.
Redesign your pricing for variable consumption. If you’re still purely seat-based, start experimenting with hybrid models now. Identify the usage metric that most closely correlates with the value your product delivers – not the cost of serving the customer. Build the metering infrastructure to track that metric at the granularity agents will demand.
Invest in data trust and provenance. Make your operational data – uptime, performance benchmarks, SLA compliance, integration specifications – structured, verifiable, and machine-readable. In an agent-mediated marketplace, this data is your marketing. If agents can’t find it or verify it, you don’t exist.
Build agent governance into your offering. Don’t just build products agents can use – build products that help enterprises govern their agents. Budget controls, audit trails, spending dashboards, anomaly detection, and policy enforcement are all product features that become critically valuable when your customer’s “employee” is an autonomous system.
Start treating AI agents as a customer segment. This is the most fundamental shift. Just as companies eventually built mobile-first experiences when mobile became a primary channel, companies will need to build agent-first experiences. That means dedicated agent onboarding, agent-specific documentation, agent-optimized pricing tiers, and agent-aware support workflows.
The Transition Is the Opportunity
We’re in the early innings of a shift that will reshape how trillions of dollars in enterprise software is bought and sold. The companies, investors, and practitioners who recognize this early – who start building for agent-mediated commerce now, rather than waiting for it to become obvious – will capture disproportionate value.
The good news is that this transition creates enormous opportunity. The enterprises that deploy agentic AI for procurement, vendor management, and operational decision-making will gain compounding efficiency advantages. The software vendors that make their products agent-ready will win in a buying channel that’s about to become the dominant one. The service providers that help enterprises navigate this transition will be building the infrastructure for the next decade of enterprise commerce.
The question isn’t whether this future is coming. It’s whether you’re positioned for it.
Frequently Asked Questions
What does it mean for AI agents to buy software?
AI agents buying software means autonomous systems – not humans – evaluate vendors, compare pricing, negotiate terms, and execute purchases on behalf of an organization. The agent operates with assigned credentials, a budget, and decision-making authority, much like an employee would. Humans shift from approving every transaction to reviewing outcomes and setting guardrails. Gartner projects that by 2028, 90% of B2B purchases will be intermediated by AI agents, representing over $15 trillion in spend.
How does AI agent procurement differ from traditional software buying?
Traditional software buying involves humans researching vendors, attending demos, running pilots, and navigating procurement committees over weeks or months. AI agent procurement compresses this into hours or days. Agents evaluate vendors against structured criteria simultaneously, compare consumption costs in real time, and can switch providers with minimal friction. This shifts the competitive advantage from brand perception and sales relationships to API accessibility, data trust, and consumption-friendly pricing.
Why is seat-based SaaS pricing breaking down?
Seat-based pricing assumed a direct correlation between employee headcount and software usage. AI agents break this in three ways: they reduce the number of humans needed for workflows (fewer seats needed), they create highly variable consumption patterns that don’t map to user counts, and when agents are the buyer, they optimize for capability per unit of cost rather than per-seat licensing. By 2025, 85% of software companies had already adopted some form of usage-based pricing in response.
What is consumption-based pricing and why does it matter for AI?
Consumption-based pricing charges customers based on actual usage – tokens processed, API calls made, workflows executed, or compute consumed – rather than a flat per-user fee. It matters for AI because agent-driven workloads are inherently variable and don’t correlate with headcount. This model aligns cost with value delivered and gives AI agents the flexibility to scale usage up or down based on need. Companies like Snowflake, OpenAI, Anthropic, and Salesforce’s Agentforce already use variations of this approach.
How should enterprises govern AI agents that make purchasing decisions?
Enterprises should treat AI agents like employees with defined authorities. This means setting explicit spending limits per agent, establishing audit trails for every transaction, building anomaly detection to flag unusual behavior, and conducting regular reviews of agent decisions against organizational strategy. The governance challenge extends beyond cost control – agents may optimize for narrow metrics that conflict with strategic relationships, diversity commitments, or long-term partnerships that require human judgment.
What does “agent-ready” mean for a software product?
A product is agent-ready when an autonomous system can discover, evaluate, provision, consume, and renew it entirely through APIs without requiring a human to navigate a UI, fill out a form, or attend a demo. This means having structured APIs, machine-readable product catalogs, composable microservices, verifiable performance data, and consumption-based pricing. Products that depend on human UX for differentiation risk becoming invisible to the fastest-growing B2B buying channel.
How will AI agents change B2B sales and marketing?
Marketing shifts from brand storytelling aimed at humans to data trust aimed at machines. Agents evaluate vendors based on structured performance data, SLA compliance, and integration compatibility – not content marketing or brand perception. Sales cycles compress dramatically as agents run multi-vendor evaluations in parallel. Sales teams will focus on the smaller number of strategic decisions that still require human judgment while optimizing for agent-driven evaluations everywhere else.
