The Price of Autonomy
Hidden costs, forgotten clauses, and the contractual governance agentic AI requires. How Agentforce, MuleSoft, and SAP interact to activate financial liabilities that no budget anticipated.
Tags: Salesforce, MuleSoft, SAP, Agentforce, LATAM, Software Licensing
Hidden costs, forgotten clauses, and the contractual governance agentic AI requires.
Licensing Assurance · Vol. 01, No. 04 · May 2026 · Guillermo Sandí · Principal
Software contracts were written for humans. An autonomous agent does not read shared-use clauses; it activates them.
The emergence of autonomous agents — platforms such as Salesforce Agentforce — represents more than an improvement in automation. It represents a contractual discontinuity. Organizations that deploy these systems without a specific review of their licensing contracts are building operational efficiency on top of financial liabilities that have not yet appeared in any budget.
The problem is not technical. It is structural. An agent that queries an ERP, creates a document in SAP, or processes a reimbursement through MuleSoft executes actions that existing contracts classify — if they classify them at all — under rules designed for human users. The consequences of that gap can reach millions.
§ 01 — Paradigm: From seat licenses to consumption models
Historically, the software budget operated under an understandable metric: number of users multiplied by license price. Predictability was inherent to the model. ✓ In May 2025, Salesforce formalized its transition toward outcome and consumption licensing by introducing Flex Credits: credits consumed by every action executed by the agent, at a base rate of $0.10 per action.
The shift is logical from the vendor’s perspective. An agent can process the workload of multiple employees without breaks, making the “seat” obsolete as a unit of value. What the model does not solve — and where most organizations encounter their first surprise — is the budget volatility it introduces.
The agent’s operating success — more cases resolved autonomously — translates into a linear and potentially uncontrolled increase in billing.
◐ Organizations that do not understand Agentforce expansion mechanics are exposed to a feedback effect where the agent’s own efficiency scales costs. Licensing Assurance acts as a balancing mechanism between agent agility and financial predictability: it analyzes not only Salesforce’s direct cost, but also the shock waves the agent generates across surrounding infrastructure.
Comparison of agentic adoption models — 3-year TCO
| Platform | Billing base | TCO scale (3 years) | Integration dependency |
|---|---|---|---|
| Salesforce Agentforce | Base license + Flex Credits ($0.10/action) | $166K — $3.7M | High — native via MuleSoft |
| Microsoft Copilot | $30 per user / month | Fixed per seat | Medium — Microsoft ecosystem |
| OpenAI Enterprise | Per million tokens ($10 input / $40 output) | Highly variable | Very high — requires custom development |
◐ TCO projections include professional services, training, and Data Cloud usage. Sources: Salesforce, Forrester TEI (MuleSoft), Licensing Assurance operating analysis.
§ 02 — Third parties: The agent as an extension of the team — contractual implications
When an Agentforce Service Agent decides to query inventory in an external ERP, it is acting as a substitute for the human user. This is precisely the operating definition that legacy software providers — SAP in particular — have historically classified as “Indirect Access.” ✓
The critical distinction versus a human user is scale. A human is limited by interface speed. An autonomous agent can orchestrate API-call sequences at volumes business platforms were never designed to handle under current licensing rules.
Every automated workflow Agentforce executes through MuleSoft can be interpreted by the system-of-record provider as an event requiring a digital-access license — or, in the worst case, as a violation of nominal-license sharing policies.
§ 03 — SAP: APIs v4 policy and digital access
✓ SAP hardened its position on autonomous agents through its APIs v4 policy, launched in 2026. This policy prohibits using its interfaces to integrate with (semi-)autonomous AI systems that plan or execute call sequences without explicit and commercially bound authorization. The stated objective: channel that traffic toward the endorsed technology stack — Joule and the Business Data Cloud.
Attention point ◐ — An agent can solve the customer’s problem in seconds. But if that process creates documents in the SAP core — orders, invoices, service orders — without the licensing structure accounting for it, accumulated volume can activate Digital Access fees. The impact is not automatic: it depends on document type, current licensing scheme, and integration architecture. Without cost design before deployment, the agent’s efficiency can erode the budget that was supposed to justify it.
◐ Licensing Assurance manages this risk by evaluating SAP’s nine billable document types and calculating financial impact before deployment.
SAP exposure types by agentic activity
| Exposure type | Activation mechanism | Potential impact |
|---|---|---|
| Indirect Access | Agent querying data without human login | Claim for “Professional User” licenses |
| Digital Access | Creation of invoices or orders via API from Agentforce | Per-document payment — scales with volume |
| Support tax | Increased API traffic requiring additional maintenance | 22% of annual license value ◐ |
§ 04 — MuleSoft: Enabler and risk multiplier
✓ Forrester’s Total Economic Impact (TEI) analysis shows that a composite organization can incur $1.8 million in MuleSoft licensing costs over three years. The platform reduces integration development effort by 60% and enables a 45% reuse rate of assets. That efficiency has a price: by simplifying the connection between Agentforce and critical systems, MuleSoft also simplifies the path toward third-party hidden costs.
MuleSoft is the conduit that transports the transaction volume that activates hidden costs. Its efficiency and its risk are the same thing.
◐ Because Agentforce can execute tasks 24 hours a day, the risk of exceeding API-call quotas and entering penalty pricing bands is structurally high. Mitigation requires decoupling architectures: agents consume data from an intermediate layer — a high-speed database or messaging system — instead of querying the system of record directly on every interaction.
MuleSoft integration ROI — 3-year present value
| Category | Present value | Implication for Licensing Assurance |
|---|---|---|
| Developer productivity | $2.96M | Optimize iFlows to reduce redundant API calls |
| Operating efficiency | $5.75M | Use agents to reduce case volume without saturating the ERP |
| MuleSoft licensing costs | ($1.80M) | Manage quotas to avoid volume overages |
| Maintenance and operations | ($0.59M) | Monitor jobs and handle errors in real time |
✓ Forrester Total Economic Impact™, MuleSoft. ◐ Smaller LATAM organizations should adjust these values to real integration volume and IT team maturity.
§ 05 — Real TCO: The shadow costs of agentic AI
Salesforce costs are only the visible tip of total TCO. Behind every dollar in Agentforce licenses sit layers that rarely appear in the initial analysis: ◐ professional services of $50K to $150K during launch; prompt-engineering training of $15K to $30K; data-governance costs that vary by architecture; and the most insidious risk — “Shadow AI”: unsupervised use of external APIs by agents that administrators can configure without prior compliance analysis.
What most teams miss: Well-managed automation can reduce operating costs by 30%. But that is possible only if duplicate processes are avoided and data is unified into a single source of truth that acts as a buffer between the agent and external systems.
Licensing Assurance classifies agentic deployment costs into four operating categories:
| Category | Main components |
|---|---|
| Direct costs | Salesforce licenses, Agentforce add-ons, Data Cloud storage, API quotas |
| Implementation costs | External consulting, reasoning-model design, MuleSoft flow tuning |
| Personnel costs | Certified administrators, data-security specialists, MuleSoft architects |
| Risk and compliance costs | Third-party audits (SAP), data-residency fines, remediation for agent errors |
Technical debt does not accumulate only in code. In agentic deployments, it accumulates in the contracts nobody reviewed when the agent started working.
§ 06 — DAAP: The negotiation lever in the SAP ecosystem
SAP’s Digital Access Adoption Program (DAAP) offers the clearest contractual mechanism for managing organic transaction growth generated by a successful agentic deployment. ◐ Option A allows licensing 115% of current usage while paying only for the additional 15%.
This buffer is vital. If the agent succeeds, transaction volume will grow organically. Without that insurance, any growth above the license base is charged at full list price, eliminating any efficiency benefit the agent produced in production.
Comparative analysis of DAAP options for Agentforce
| Dimension | Option A — Growth buffer | Option B — Deep discount |
|---|---|---|
| Discount off list | ~85% | ~90% |
| Included buffer | 15% additional volume at no cost | 0% |
| Ideal for | Aggressive Agentforce deployments | Stable systems with limited agency |
| Critical risk | True-up after exceeding 115% is expensive | A moderate volume increase triggers massive costs |
◐ Discount percentages and conditions subject to negotiation. Not applicable in all LATAM regions.
§ 07 — Architecture: The integration layer as financial defense
Resolving hidden costs technically requires MuleSoft to operate not only as a data transporter, but as an abstraction and consolidation layer. By moving SAP data into a lower-cost environment — a cloud-native database or Salesforce Data Cloud through Zero Copy — the agent operates on that governed copy without activating SAP licensing events at every verification. ◐
Recommended architecture practices:
- OAuth 2.0 and SAML: Avoid system credentials that make traceability difficult and could be interpreted as one user multiplexing access for thousands of agents.
- Monitoring from design: An agent stuck in an API error loop can consume months of Flex Credits in hours. Response-time and retry-count alerts are not optional.
- Logic decoupling: Business rules outside the MuleSoft mapping layer allow fast provider changes without rewriting the backend integration.
§ 08 — SAM: Software asset management in the agentic era
✓ Deloitte has identified that implementing software asset management (SAM) practices can generate average savings of 23% in annual maintenance spend. In the agentic context, SAM takes on three specific functions:
- Full visibility: knowing which agents have access to which external systems through MuleSoft. Without this map, contractual governance operates blind.
- Inventory control: no new agent should be deployed without a prior review of its third-party license impact.
- License harvesting: reassign licenses from human users fully replaced by agents in specific tasks. Unused software funds agentic expansion.
§ 09 — Human factor: The invisible cost
Frequently absent from traditional financial analysis: the skills gap. Deploying Agentforce requires capabilities few organizations have on staff — Salesforce administrators who understand LLMs, MuleSoft experts who understand agent orchestration. ◐
A system operating without that training produces more errors, which in turn produces more API calls and higher costs from failed actions. The cost of change resistance does not appear on any invoice line. It appears on all of them.
The methodology recommends a ticket-taxonomy audit before deployment to ensure the agent is applied to cases where return is clearest and the risk of unexpected cost is lowest. Efficiency first; scale later.
Four strategic recommendations
Licensing Assurance is not a purchasing or audit function. It is the discipline that makes it possible to scale autonomous agents without eroding financial margins in the process.
Redefine contractual governance. Move toward contracts that explicitly contemplate autonomous-agent usage. Negotiate with SAP under the Digital Access model with growth buffers suited to the expected adoption pace.
Architect for decoupling. Use MuleSoft not only to connect systems, but to isolate the business core from agent volatility. Intermediate data layers are not an architectural luxury; they are a financial hedge.
Seek flat-billing models where possible. Consumption-credit variability needs predictable infrastructure counterweights. The right mix depends on each organization’s risk profile.
Implement agent-centered SAM. Every action must generate business value greater than its execution cost. Real-time visibility is the only way to verify that equation holds.
An agent’s success is not measured only by resolution speed. It is measured by the ability to scale that resolution without making the vendor contract untenable.
Epistemic posture
✓ Verified: SAP APIs v4 policy (2026) · Salesforce Flex Credits (May 2025) · Forrester TEI MuleSoft · Deloitte findings on SAM.
◐ Inferred: TCO ranges by organization size · specific DAAP conditions · SAP support tax percentage (22%) · shadow cost estimates ($50K-$150K professional services).
◯ Undetermined: Post-DAAP true-up costs in specific LATAM contexts · adoption velocity of Joule as SAP’s endorsed channel in the region.
Deep Sight Consulting has no commercial partnerships with Salesforce, SAP, MuleSoft, or any of the actors mentioned in this analysis. It does not sell, resell, or receive commission on software licenses from any manufacturer. DS · LA-26.04 · SF-SAP · deepsightconsulting.com