Agentforce

A Salesforce-native agent, only when the desk is here.

Agentforce inherits sharing, topics, and actions. We implement it when service, sales, or internal queries live in the CRM.

Salesforce consulting partner since 2010 · MVP Hall of Fame · 200+ agents in production · UAE, India and US desks

The runtime question

Agentforce, or something else?

Agentforce fits when

Go custom when

The users already work in Salesforce
The job spans tools Salesforce does not hold
The record the agent changes lives in the org
The system of record is a warehouse or ERP
The agent must respect the same profiles and sharing
You need a model Salesforce will not host
Slack-first is fine when the work lives in Salesforce
Then a custom stack, often Claude, connected through the gateway

Who this is for

Three desks where a native agent fits.

One named job each, with a person on the exceptions.

  • Example

    Service director at a Dubai telecom retailer

    Today: Agents answer the same 'where is my order' question all day.

    After: The agent answers order status from the case and the order record, and hands anything else to a named person.

  • Example

    Inside sales lead at a Riyadh B2B equipment supplier

    Today: Web leads wait a day before a rep reads them.

    After: The agent qualifies the lead against written criteria before a rep opens it, and books the call.

  • Example

    IT service manager at an Atlanta hospital network

    Today: Staff ask the helpdesk for policy answers that already exist in Knowledge.

    After: The agent answers from permissioned articles with a citation, and logs a case when it cannot.

Examples of the work, not case studies. Roles and cities are typical of the desks we see.

Spec, shadow, gate

How the first agent goes live.

One named job. A shadow run before any write.

  1. Week one

    Spec

    The job on one page: inputs, the write, the limit, who approves, and what it may never do.

  2. Before go-live

    Identity and data

    The agent acts as the person it works for, with their access. Knowledge and permissions checked before it takes a job.

  3. Shadow run

    Shadow

    The agent runs beside the person and writes nothing. You compare its calls with theirs.

  4. Go-live

    Gate

    Writes go live behind a human approval. The gate loosens only where the shadow run earned it.

Retrieval

One matching shape, more than one desk.

Graph for relationships. Vectors for similarity. MCP tools behind a gateway. Same shape at more than one desk.

Questions

What we actually say.

Is Agentforce the same as Einstein Copilot?
No. Einstein recommends. Agentforce acts. If a partner conflates them, they have not shipped this.
Do we need Data Cloud first?
You need grounded, permissioned context. Sometimes that is Data Cloud. Sometimes it is clean Knowledge and a sharing model that already works. We will not quote a build before that check.
When do you refuse Agentforce?
When the work spans tools Salesforce does not hold, when the system of record is a warehouse or ERP, or when you need a model Salesforce will not host. Then we assemble a custom stack, often Claude, and connect it.
Can a Slack-first team run Agentforce?
Yes. Agentforce now runs inside Slack, so the team asks in the channel and the agent answers from Salesforce with Salesforce permissions. That fits when the work and permissions live in Salesforce. When the job spans other tools, we put Claude or a custom agent in Slack instead.

The brief · one email

Name the desk. Name the record.

We will tell you Agentforce, a custom runtime, or neither. The assessment is free: a short read of the org and a written note. If you go ahead, the first build week is a fixed AED/SAR 14,900 (about USD 4,050), credited in full when you sign the build within 60 days. The honest no: if the work is a software factory with a named delivery date, we are the wrong partner.