The Capability
ISVs embed Agentforce 360 inside their own products. The ISV’s app runs on the Salesforce platform; the ISV ships AI features powered by Agentforce under their own brand. Embedding spans the full stack: Atlas Reasoning, Topics and Actions, the Trust Layer, Data 360 grounding, Prompt Builder templates, the Testing Center for evals, and Command Center for observability. ISVs receive a partner-tier API surface that exposes Agentforce metadata operations programmatically — provision agents, deploy Topics, run evals, fetch traces — all from the ISV’s own DevOps pipeline.
Why Embed
Agentforce provides infrastructure ISVs would otherwise build from scratch: Trust Layer with PII redaction and toxicity filtering, evaluation framework with LLM-as-judge scoring, Data 360 vector grounding, Command Center observability with OpenTelemetry tracing, and per-call billing rails. Embedding reduces ISV engineering effort by 6–12 months on the platform layer and frees the team to focus on vertical depth — eval design, domain prompts, integrations with non-Salesforce systems.
ISV-build vs Agentforce-embed effort estimate:
DIY Embedded
Trust + redaction 4 mo 0 mo
Eval framework 3 mo 0 mo (use Testing Center)
Vector retrieval 2 mo 0 mo (Data 360)
Observability stack 3 mo 0 mo (Command Center)
Per-call billing 2 mo 0 mo (AgentExchange)
Compliance (BAA, DPA) 6 mo inherited
-----------------------------------------------------
Time to first ship ~9 mo ~2 mo
Revenue Implications
The ISV charges customers for the AI feature; underlying Agentforce platform costs flow through ISV pricing. Margin depends on volume and pricing model. High-volume AI features make sense embedded because the platform fee amortizes well. Low-volume premium features may make sense as direct LLM integration via Anthropic or OpenAI APIs without the platform layer. Run the unit economics: Agentforce platform fee plus model token cost plus Data 360 consumption versus a DIY stack on AWS Bedrock or similar — the embed wins on time-to-market and operational burden, not necessarily on raw cost per call.
Brand Positioning
ISVs can white-label (no mention of Agentforce) or co-brand (“Powered by Salesforce Agentforce”). White-label preserves the ISV’s differentiation and is strongly preferred in verticals where customers don’t perceive Salesforce as a category leader (legal, healthcare, manufacturing). Co-brand leverages Salesforce trust and shortens enterprise InfoSec review cycles meaningfully — common in horizontal sales/service categories. Strategic choice per ISV positioning; the partner agreement supports both.
Migration Path from DIY to Embed
Three-phase migration for an ISV already running on AWS Bedrock or similar: (1) keep your business logic in Apex Invocable Methods or Heroku microservices; (2) port prompt templates into Prompt Builder and register tools as Agentforce Actions; (3) cut over evaluation and observability to Salesforce-managed services last, after enough traffic has accumulated to validate parity. Total migration: 3–6 months for a mid-complexity product.
Common Failure Modes
- Embedding before validating product-market fit — the platform fee adds friction to a product still searching for adopters.
- Hardcoding to Salesforce xGen and losing the option to switch models when economics shift.
- Skipping the partner-tier API and trying to provision agents through the customer-facing UI; CI/CD breaks down.
What to Do This Week
If you’re building an AI feature inside an AppExchange app, model the unit economics for embedded Agentforce versus a direct LLM integration. Pick the path that matches your volume and time-to-market.