When AI Agents Become the Buyer: What Per-Seat SaaS Didn't See Coming

Per-seat pricing built a €300 billion SaaS market. Workday, SAP SuccessFactors, Oracle HCM, Salesforce — these platforms run the operational backbone of most large enterprises, and for good reason. They solved genuinely hard problems: compliance across jurisdictions, payroll at scale, global sales operations, unified employee data.

The model survived every disruption wave of the last two decades. Cloud migration didn't break it. Mobile didn't. Remote work didn't. Each shift changed the delivery channel, but the fundamental assumption held: a human being would log into a dashboard, navigate an interface, and perform a task.

AI agents challenge that assumption. They are not better users of these platforms; they are a different kind of buyer.

The infrastructure shipped in 2025

This isn't a theoretical future. In 2025, Stripe and OpenAI launched the Agentic Commerce Protocol (ACP) — enabling agents to discover and purchase services programmatically. Google shipped AP2 with 60+ partners including Mastercard, PayPal, and American Express. Visa and Mastercard both launched agent trust and payment frameworks that are live in production.

Now, software can discover, evaluate, and pay for services per request, in milliseconds and at very low cost. The infrastructure for machine-to-machine transactions is in place.

Why HR is the clearest case study

Most people associate "agentic commerce" with shopping. The bigger change is inside enterprises, where transaction costs are high compared to decision complexity.

HR is a clear example. Most employee interactions with platforms like Workday or SuccessFactors are lookups: PTO balances, expense policies, employment history, benefits eligibility. These queries have structured inputs and outputs. They seem complex because of the human-focused interfaces, but the decisions are usually simple.

In an agent-native model, each interaction is a specialized API call. The employee's AI assistant queries an endpoint directly, with structured input and output, and compliance reference included. No logins, navigation, or dashboards are needed.

This shift changes six common HR processes.

1. Employment verification

Verifying a candidate's work history typically means three days of emails and phone calls. In an agent-native model, the hiring agent hits a Verified Employment API with a candidate-provided token and gets a cryptographically signed response — exact dates, title history, rehire eligibility — in 200 ms for roughly €0.05.

2. Salary benchmarking

Compensation teams buy static salary surveys annually — subscriptions that can run €50,000 or more, with data that's months old by the time it's applied. An agent-native service queries live market rates at the moment an offer is being drafted. A query like "Senior DevOps Engineer in Berlin" returns a percentile rank for roughly €0.01.

3. Policy questions

"Can I work remotely from Spain for three weeks?" Answering this means checking residency status, tax rules, remote work policy, and project status. It usually takes 15 minutes of an HR professional's time. A policy-as-code endpoint can check all four and return a yes/no with the policy reference in under a second. No human intervention is needed for this rule-based lookup.

4. Technical screening

Take-home tests are increasingly unreliable, and grading still requires two or more hours of senior engineering time per candidate. A specialized evaluation agent conducts an adaptive technical interview, evaluates reasoning patterns, and returns a confidence score with verification metadata — for approximately €2.00.

5. Benefits enrollment

Open enrollment often takes an hour of clicking through screens employees rarely use, comparing plans with incomplete information. A benefits optimizer agent can query multiple carriers at once, compare coverage-to-price ratios, and present or execute the best option.

6. Retention monitoring

Annual engagement surveys often give outdated insights. An agent-based approach can monitor signals continuously — working patterns, internal mobility, external market demand — and surface specifics like: "High probability this developer leaves within 60 days. Replacement cost: €140,000."

The economic pressure on per-seat pricing

Today, these processes are bundled into platforms that charge per seat. This pricing works because buying each capability separately used to have high transaction costs. Finding vendors, integrating systems, and maintaining connections made it simpler to pay one platform for everything.

This is textbook Ronald Coase. Companies bundle activities because the cost of coordinating through the market exceeds the cost of doing it in-house. The per-seat SaaS model is essentially a firm boundary.

AI agents remove those costs. An MCP server publishes capabilities as structured data. Authentication is programmatic. Compliance can be machine-readable. When integration takes minutes instead of months, the long implementation projects that favored renewing Workday look less necessary.

The incumbents are not standing still

Large platforms are responding quickly. SAP now ships over 40 AI agents. Salesforce's Agentforce has closed thousands of enterprise deals. Oracle has launched an AI Agent Marketplace. Each is trying to become the orchestration layer: the system that coordinates which agent-queryable services are called, under which constraints, and with what budget.

The pricing model is the weak point. You can add agents to a per-seat platform, but if most interactions move from humans using dashboards to agents querying endpoints, per-seat pricing becomes harder to justify.

What this means for CTOs and engineering leaders

The key question is not if this transition will happen. The infrastructure is live and the economic logic is clear. The real question is what to do now, before your next contract renewal.

Timeline: when does this become real?

This shift will not happen all at once. It will follow the complexity curve:

• Now: Low-complexity, high-frequency lookups — policy questions, PTO checks, benefits eligibility — are ready for agent-native delivery. The protocols exist, the economics work, and the risk is low.

• 12–18 months: Mid-complexity processes like salary benchmarking and employment verification. Specialized services exist, but enterprise-grade trust infrastructure — audit trails, compliance verification, data residency — is still developing.

• Longer term: High-stakes processes — performance management, succession planning, workforce strategy — will remain human-led for the foreseeable future. This is where the current platforms retain their strongest position.

The practical step is not to wait for a full replacement. Identify which of your current platform interactions are simple lookups; those are the best candidates for early unbundling.

Integration: the real hurdle isn't technical

The agent commerce protocols (ACP, A2A, MCP) have solved the technical integration problem — discovery, authentication, and payment can happen in a single HTTP roundtrip. The harder challenge is organizational.

Most enterprises have years of workflows, approval chains, and reporting structures built into their HR platforms. Unbundling a policy query endpoint from Workday is not just a technical task. It also means rethinking data ownership, business rule maintenance, and how outputs connect to downstream processes.

A practical first step: map your current HR platform usage by interaction type. Separate lookups (structured input and output) from workflows (multi-step, require judgment). Agent-native alternatives will outperform on lookups first; workflows are where your current platform still adds value.

Security and compliance: agent identity is the new frontier

This is the concern I hear most from enterprise leaders, and it's legitimate. When a human submits an expense report, there's an identity chain: SSO, role-based access, an audit log tied to a person. When an agent does it on behalf of an employee, you need equivalent guarantees.

The good news: the infrastructure is being built specifically for this.

• Visa's Trusted Agent Protocol uses cryptographic signatures to verify agent identity.
• Mastercard's Agentic Tokens are scoped, time-limited, and revocable.
• Google's AP2 uses digitally signed "mandates" that capture user intent and create non-repudiable records.

These aren't theoretical specs — they're in production.

The main gap is in regulated industries where compliance frameworks are still catching up. If your organization is under GDPR, SOX, or industry-specific rules, ask your legal team now: "What is our policy on AI agents acting for employees in automated transactions?" Most companies do not have an answer yet. Having one early is a competitive advantage.

Vendor strategy: what to do at the next renewal

This is where you can take action. At your next contract negotiation with Workday, SAP, or any enterprise SaaS platform, consider these steps:

1. Ask about API-level access to the data and business logic you're paying for. Not a reporting API — the actual policy engine, the eligibility rules, the compliance checks. If the platform's value is in the rules, you should be able to query those rules programmatically. If they resist, that tells you something about where they think their moat is.

2. Push for usage-based pricing. Even if the base contract is still per-seat, negotiate for consumption-based tiers on high-volume, low-complexity interactions. This protects you if agent-native alternatives mature faster than expected, so you are not locked into paying full per-seat rates for interactions that no longer need a seat.

3. Evaluate your vendor's agent strategy. SAP, Salesforce, and Oracle are all shipping agents, but are they building an open orchestration layer or a closed system? Vendors that expose their capabilities via MCP servers and A2A Agent Cards are preparing for the agent-native future. Those that only offer agents within their own dashboard are sticking to the old model.

4. Start small. Choose one low-risk process — policy lookups, employment verification, or salary benchmarking — and pilot an agent-native alternative alongside your current platform. This comparison will give you data for your next renewal and valuable organizational learning.

The trillion-dollar question

The platforms will not disappear. Workday, SAP, and Salesforce have strong positions in enterprise relationships, data, and compliance. But per-seat pricing is under pressure for the first time in two decades, not from a better platform, but from a different delivery model.

When most interactions move from humans using dashboards to agents querying endpoints, the meaning of "per seat" changes.

Enterprise software has not answered this yet. Companies that start asking now, on both sides of the table, will help shape what comes next.