Salesforce Spring ‘26: Agent Script Arrives
Salesforce’s Agentforce Spring ‘26 release introduces Agent Script — a feature that gives enterprises “greater control of agent behavior” through scripted behavior paths. Salesforce describes Agentforce agents as operating with “task authority, dynamic decisioning, and multi-step workflow execution within defined safety parameters.”
This sounds like governance. It isn’t.
Script vs. Governance: The Key Distinction
Agent Script controls what agents say and do along predefined routes. Think of it as a conversation flowchart: if the customer says X, the agent does Y, then Z.
Architectural governance controls what agents are permitted to access, use, and execute — across any path, in any department, on any model. It’s the infrastructure layer that ensures agents operate within defined authority boundaries, regardless of their behavior paths.
The difference:
- Script = behavior (what agents do along a path)
- Governance = authority (what agents are allowed to do at all)
Where Agent Script Falls Short
1. CRM-Scoped
Agent Script governs Agentforce agents within Salesforce. It covers 3 departments (Sales, Service, Marketing). What about the other 20 departments — Legal, Finance, Operations, Engineering, HR, Executive?
2. No Cross-Platform Governance
Enterprises use agents from multiple vendors: Salesforce, Microsoft, custom-built. Agent Script governs Salesforce agents only. Who governs the rest?
3. No Quantitative Measurement
Agent Script has no GovernanceScore equivalent. How do you measure whether your agents are adequately governed? JieGou’s GovernanceScore provides an 8-factor metric (0-100) per agent and department.
4. No Regulatory Framework Mapping
Agent Script doesn’t map to EU AI Act articles, NIST AI RMF functions, or ISO 42001 controls. With the EU AI Act now enforceable, regulatory compliance isn’t optional.
What Architectural Governance Looks Like
JieGou’s 10-layer governance architecture provides:
- RBAC — Role-based access control limits agent scope across all applications
- Tool approval gates — Agents can only access pre-approved tools (not just scripted paths)
- GovernanceScore — Quantitative measurement (0-100) across 8 governance factors
- Graduated autonomy — 4 trust levels from supervised to full autonomy
- Approval workflows — Human review gates for high-risk actions
- Audit trails — Every action logged for compliance documentation
- Three-framework compliance — EU AI Act + NIST AI RMF + ISO 42001
This works across 20 departments, any model provider, and any deployment model (cloud, VPC, self-hosted).
They Can Coexist
Many enterprises will use Agentforce for CRM-centric departments and JieGou for the other 20 departments. Agent Script handles Salesforce-specific behavior. JieGou provides the cross-platform governance architecture.
The question isn’t whether Agent Script is useful — it is, within Salesforce. The question is whether scripted behavior paths constitute enterprise-wide agent governance. They don’t.