At Cirrius Solutions, we implement both Custom AI Agents and Salesforce Agentforce. One of the most common questions we get is how the cost of a Custom Agent compares to Agentforce.

The reality is that custom agents offer greater architectural flexibility, but require significantly more technical expertise to design, build, and operate. Agentforce, on the other hand, comes out of the box with the core building blocks you need, such as authorization, security, workflow, data, and routing, which makes it easier and faster to stand up. When you look at the total cost of ownership, the numbers usually end up surprisingly close.

The real decision isn’t “Which is cheaper?” but rather: Do you want the simplicity and consolidation of Agentforce, or the flexibility and control of a custom agent? Below is a summary of the considerations for your review.  See our high-level summary below.

Cost of Custom Agent vs Salesforce Agent

What It Takes to Build a Custom AI Agent

When you build your own agent, you’re essentially building a mini AI platform around your business. The main workstreams look like this:

A. Architecture & Stack Selection

  • Pick your LLM provider(s) (OpenAI, Anthropic, etc.).
  • Choose an agent framework (LangChain, Pydantic AI, custom orchestration).
  • Choose a vector database/search engine for RAG.
  • Decide where to host everything (AWS/Azure/GCP/Render/etc.).
  • Decide how you’ll do observability (traces, logs, metrics).

All of this is your responsibility to design and maintain.

B. Data & FAQ Grounding

  • Build or configure a crawler to ingest your FAQ/help center.
  • Normalize and chunk content, create embeddings, index it.
  • Implement a retrieval layer and wire it into the agent.
  • Handle staleness, updates, and re-crawls on a schedule.

You own the full RAG lifecycle.

C. Channels & UX

  • Integrate with your chat widget (Intercom, custom web chat, in-app SDKs).
  • Map chat events → your agent API → responses back to the user.
  • Build the handoff logic to human agents and your existing support tools.

You’re doing custom integration work per channel.

D. Authentication & Business Actions

  • Implement login/verification (OTP, SSO, etc.).
  • Bind authenticated chat sessions to user identities.
    • Implement tools that call your subscription/order APIs:
  • Pause/skip shipments, change flavors, start returns, etc.
  • Enforce authorization, masking, and audit trails at your own code layer.

You design and own the security and policy enforcement.

E. Safety, Governance & Testing

  • Design guardrails (no medical advice, no PII leaks, etc.).
  • Red-team for prompt injection, auth bypass, and data leakage.
  • Build monitoring dashboards for success metrics and errors.

You are the platform owner from a risk and compliance perspective.

What It Takes to Build a Salesforce Agentforce Agent

With Agentforce, you’re still building a serious solution, but more of the “platform-y” pieces are handled inside Salesforce. The main workstreams shift from building to configuring.

A. Stand Up the Core Service Stack

  • Ensure you have Service Cloud for Cases, Service Console, Omni-Channel.
  • Add Digital Engagement for Messaging (web widget + in-app messaging).

This gives you the core service and routing fabric out of the box.

B. Grounding & FAQ via Data Cloud

  • Configure Data Cloud Web Content connections (Sitemap/Crawler) to your help center.
  • Model content as Unstructured Data and set up a Search Index with vector/hybrid search.
  • Connect that index to your agent via a Data Library/Retriever.

Salesforce handles the crawling, indexing, and vector search as platform features.

C. Configure the Agentforce Agent

  • Create an Agentforce Service Agent for customer-facing conversations.
  • Define topics/intents, tone, and guardrails (e.g., escalation triggers).
  • Link the agent to:
    • Your Data Library (for FAQs)
    • Your Flows / Invocable Actions (for account and transaction work)

You focus on behavior and policies, not low-level orchestration code.

D. Authentication & Actions in Salesforce

  • Use Experience Cloud + External Identity / Experience licenses for login.
  • Turn on User Verification in Enhanced Web Chat to bind sessions to users.
  • Build Salesforce Flows to:
    • Pause/skip shipments
    • Change items
    • Start returns
    • Update contact data
  • Let the agent invoke those Flows with Salesforce’s FLS/sharing rules enforced.

Authorization and auditability live inside the CRM you already use.

E. Routing, Handoff & Observability

  • Use Omni-Channel / Omni-Channel Flow for routing:

Tier 1 → Agentforce

    • Escalations → human queues
    • Surface the transcript, summary, and context right inside the Service Console.
  • Use standard Salesforce dashboards and logs for performance and safety monitoring.

You’re using existing Salesforce operations patterns instead of stitching together multiple tools.

The Key Message: Effort & Cost Are Roughly Comparable

When you zoom out, both paths require:

  • Thoughtful discovery and design
  • Clear intents, policies, and guardrails
  • Solid authentication and security
  • Thorough UAT and live monitoring

The difference is where the platform responsibilities live:

  • With a Custom Agent, you build and own the AI platform: stack selection, crawling, vector search, hosting, security layers, observability, and multi-vendor management.
  • With Salesforce Agentforce, you assemble those same capabilities using Salesforce building blocks: Data Cloud, Agentforce configuration, Flows, Omni-Channel, and Experience Cloud.

Because of that, Salesforce Agentforce is not inherently more expensive in general—especially if:

  • You’re already Salesforce-based for service, and
  • You’d rather configure on a mature platform than stand up and run your own AI stack.

So the real decision tends to be less “Salesforce is too pricey” and more:

  • Do we want platform consolidation and procurement simplicity (Agentforce)?
  • vs.
  • Do we want maximum architectural flexibility and control (Custom Agent)?

We’ve done this analysis for many customers and are happy to share real-world findings (with customer details appropriately redacted), including actual cost breakdowns. If you’re working through this decision and want an outside perspective, reach out — we’d be glad to offer our two cents.