How to Build an AI Agent for Salesforce Service Cloud (2026): API, Agentforce or Fin

There are three ways to put an AI agent on Salesforce Service Cloud: build it yourself on the Salesforce REST API (weeks of work, full control), buy Salesforce's own Agentforce (Flex Credits, about $0.10 per standard action), or run a third-party agent such as Fin, which Salesforce now owns and which charges $0.99 per outcome. If your tickets live in a different help desk and only your customer data lives in Salesforce, a fourth option applies: an agent platform like Macha that calls Salesforce as a tool.

What does a Salesforce Service Cloud AI agent do?

The loop is the same as any support agent, wired to Service Cloud: a new case arrives, the agent reads it and the contact's history, it calls tools (look up the account, search your knowledge base, check a policy), then it posts a reply as a case comment or feed item, or routes the case to a human queue. The hard part is doing that reliably against real cases. (New to agents generally? Start with AI agents for customer service.)

How do you build one on the Salesforce REST API?

Salesforce has a full REST API, so you can build the agent yourself. The pieces:

• Read cases: the Salesforce REST API sObject and Query (SOQL) endpoints pull the Case, its CaseComment/feed history, and the related Contact and Account.
• Reply or act: create a CaseComment or FeedItem, or PATCH the Case record (status, owner) via the sObject REST API.
• Trigger on new cases: a Platform Event, streaming subscription, or outbound message fires your endpoint when a case is created.

A reply tool against the Salesforce REST API looks like this:

import requests

INSTANCE = "https://yourco.my.salesforce.com"   # from the OAuth token response
API_VERSION = "v60.0"   # GET {INSTANCE}/services/data/ lists the versions your org supports
API = f"{INSTANCE}/services/data/{API_VERSION}"
HEADERS = {"Authorization": f"Bearer {ACCESS_TOKEN}", "Content-Type": "application/json"}

def reply_to_case(case_id: str, body: str):
    r = requests.post(f"{API}/sobjects/CaseComment",
        json={"ParentId": case_id, "CommentBody": body, "IsPublished": True},
        headers=HEADERS, timeout=10)
    r.raise_for_status()          # handle 401 token refresh, API-limit errors
    return {"status": "sent", "id": r.json()["id"]}

Wrap that (and a get_contact, a search_knowledge) as tools in an agent loop on the model of your choice. Our from-scratch walkthrough is how to build an AI agent: from scratch vs. platform.

What does the production part involve?

The API calls are the easy bit. A Service Cloud agent you'd trust also needs: an event endpoint (Platform Events or outbound messages, with verification and idempotency so a retried event doesn't double-reply), API-limit handling (Salesforce enforces per-org daily request limits), grounding in your Knowledge base or help center (embed the articles, keep them fresh), guardrails (escalate refunds, don't leak PII), OAuth token refresh and 24/7 hosting, observability, and an eval harness over real historical cases. That's weeks of work plus upkeep.

What do Agentforce and Fin cost?

If you'd rather buy than build, Salesforce sells two agents for Service Cloud. Agentforce pricing (checked 24 September 2026; as of October 2026 its $2-per-conversation card is for existing Conversations customers only) offers new buyers Flex Credits at $500 per 100,000 credits with a standard action costing 20 credits (about $0.10), or a Service add-on at $125 per user per month with unmetered usage. Salesforce also completed its acquisition of Fin on 10 September 2026; Fin runs on Salesforce among other help desks, costs $0.99 per outcome, and carries a 50-outcome monthly minimum when it runs outside Intercom. Salesforce cites a 76% average resolution rate for Fin.

Where does Macha fit with Salesforce?

Macha is an AI agent layer that runs on top of the help desk you already use: Zendesk, Freshdesk, Gorgias, Front, HubSpot or Intercom. Salesforce Service Cloud is not one of them, so if your cases live in Service Cloud, the build route, Agentforce or Fin are the direct options. Macha fits the common split where tickets live in one of those help desks and the customer record lives in Salesforce:

• Salesforce as a tool: Custom Tools turn the Salesforce REST API into agent capabilities by describing them, so the agent can pull the account, opportunity and owner into the ticket or log the ticket back to the CRM timeline. Custom tools are included in every plan.
• Built for you: the Macha team builds and tests the Salesforce tools and the agent, then runs it in safe mode, with drafts as internal notes in your help desk, until the drafts are right. That's included on the trial and every plan at about $0.40 a ticket, along with a dedicated success manager who handles your changes.
• Ground on your help center: add your knowledge base or public help center as a Source so replies quote real articles.
• Runs and grades: Agent Analytics show every run, and simulations grade the agent against real historical tickets before it goes live.

Grounding matters as much as the connection: pointing the agent at your real Knowledge articles as a Source is what keeps replies accurate instead of invented.

Which route should you pick?

Build on the API if you need a fully custom runtime, have strict data-residency needs, or the agent is your product. Buy Agentforce or Fin if your cases live in Service Cloud and you want the shortest path. If your support runs in another help desk and Salesforce is your CRM, you can start free on Macha with $50 of free usage and no credit card; the Macha team builds the Salesforce tools and the agent and tests it on your own tickets. After the trial it's about $0.40 a ticket, setup and a dedicated success manager included. If you're weighing a Salesforce partner against Salesforce's own services team, our Service Cloud AI implementation guide compares weeks, published costs and partner tiers.

FAQ

Is a Salesforce Service Cloud AI agent the same as Salesforce's own Agentforce? Not necessarily. Agentforce is Salesforce's built-in agent, priced in Flex Credits at about $0.10 per standard action. You can also build a model-agnostic agent on the Service Cloud API yourself, or run Fin, which Salesforce acquired on 10 September 2026.

Does Macha run on Salesforce Service Cloud? Not as a help desk. Macha runs on Zendesk, Freshdesk, Gorgias, Front, HubSpot and Intercom, and reaches Salesforce through Custom Tools that call its REST API, so it suits teams whose tickets live in one of those desks and whose customer data lives in Salesforce. The Macha team builds those Custom Tools and the agent for you, included on every plan.

Can the agent take actions, not just answer? Yes. Via the Salesforce API it can update case status, reassign an owner or add comments, and your own systems' APIs (order lookups, refunds) can be tools too, with guardrails on what it does unattended.

How do I make sure it's accurate before it replies to customers? Ground it in your Knowledge base and grade it against real historical cases first, start in a draft-and-approve mode, and widen autonomy per case type as it earns trust.