Agentforce sales agents: what changes when the agent prospects on its own
Most teams build a service agent first, then assume the SDR agent is the same pattern pointed at leads. It isn't. Sales agents act unsupervised and outbound, the standard ones are configured rather than coded, and the failure mode is emailing a prospect something wrong. Here's how Agentforce SDR and Sales Coach actually work, and where a human still has to stand.
Almost every org we talk to builds a service agent first. It’s the obvious starting point: high volume, a customer who initiated the conversation, and a clean success metric in deflection. So when the same team turns to sales, the instinct is to treat the SDR agent as the service pattern with the topics swapped — point it at leads instead of cases and let it run.
That instinct is wrong in a way that matters. A service agent waits to be spoken to. A sales agent reaches out. The moment an agent sends the first message instead of answering one, three things change at once: there’s no human-initiated turn to anchor the interaction, the failure mode moves from “unhelpful answer” to “wrong thing said to a prospect who never asked to be contacted,” and the thing you’re optimizing stops being deflection and becomes pipeline. None of the muscle memory from your service build tells you how to handle that.
The other surprise is architectural. The two headline sales agents Salesforce ships — Agentforce SDR and Agentforce Sales Coach — aren’t the blank-canvas “build a custom agent in Studio” experience you used for service. They’re pre-built, standard agents you configure: enable a feature, assign a licence, set guardrails, ground them in your data, and tune. That’s faster to stand up and narrower in what you can change. Knowing which parts are yours to shape and which are fixed is most of the job.
This post walks the two standard sales agents as they stand in mid-2026: what each actually does, the setup and licensing that trips teams up, the guardrails that keep autonomous outbound from becoming a liability, and where a human still has to stand. If you haven’t built any Agentforce agent yet, start with our service-agent build guide — this assumes you’ve seen Agent Builder before.
Two standard agents, two very different jobs
Salesforce ships sales as two distinct pre-built agents, and the confusion starts with the fact that they were renamed mid-flight. Both went generally available in October 2024 as Einstein SDR Agent and Einstein Sales Coach Agent, then got folded into the Agentforce brand — they’re now Agentforce SDR and Sales Coach. Salesforce’s own help documentation now also files the SDR agent under Agentforce Lead Nurturing, so you’ll see all three names for the same thing. Set that aside and the split is clean:
- Agentforce SDR (Lead Nurturing) is customer-facing and autonomous. It engages inbound leads over email, answers product questions from approved content, qualifies against your criteria, and books meetings — without a rep in the loop for each step.
- Agentforce Sales Coach is employee-facing. It gives a rep deal-specific feedback and, in its most useful mode, plays the customer in a role-play so the rep can rehearse a negotiation before the real call.
One talks to your prospects. One talks to your reps. They share a platform and almost nothing else, and the risk profile could not be more different — which is exactly why teams that treat them as one project get burned.
Why an outbound agent is a different risk class
A service agent’s worst day is an unhelpful answer to a customer who’s already in a support conversation and expecting a bot. Annoying, recoverable. An SDR agent’s worst day is an unprompted, personalized, wrong email landing in a prospect’s inbox with your company’s name on it. There’s no “the customer asked for it” defense, and depending on the prospect’s jurisdiction and consent status, there may be a compliance question sitting underneath the embarrassment.
That’s the frame to hold the whole time you configure a sales agent: every outbound action is a published, attributable statement from your brand, sent to someone who didn’t opt into this exact conversation. It raises the bar on grounding, tone, and volume control far above what a service agent needs. The good news is that Salesforce built the standard SDR agent with that in mind, and most of your setup work is turning those controls on deliberately rather than accepting defaults.
Setup: the licence and the agent user are the gate
Both sales agents run as autonomous users, which means the first real step isn’t in Agent Builder at all — it’s provisioning identity. Each agent needs a dedicated user record assigned the Einstein Agent User License and the Einstein Agent User Profile, plus the relevant permission set (for Sales Coach, that’s the Agentforce Sales Coach permission set). This is the same “agents are users” model we cover in governing an agent fleet, and it’s the model that decides what the agent can see and do — the agent inherits the sharing and field-level security of the user it runs as, nothing more.
The prerequisites underneath are the usual Agentforce stack: Agentforce (formerly Einstein Copilot) enabled, Einstein Generative AI turned on to power the prompt templates, Agentforce Studio / Agent Builder to configure topics and actions, and Data 360 (Data Cloud) for grounding. If you’re still weighing whether Data 360 is mandatory, we mapped that dependency in detail in does Agentforce need Data Cloud — for a sales agent that’s writing personalized outbound, treat good grounding as non-optional, not a nice-to-have.
The convenience — and the thing to watch — is that enabling the SDR agent cascades. When you turn on Lead Nurturing, Salesforce automatically enables the supporting features it needs for outreach, including Sales Engagement, Automated Actions, Einstein Activity Capture, and Email Productivity. That’s helpful, but it also means a single toggle can switch on activity capture and engagement tracking across your org. Know that before you flip it in a production org, not after a rep asks why their sent mail suddenly shows up in a capture feed.
Inside the SDR agent: topics, actions, and the email meter
Structurally the SDR agent is the same object model as any Agentforce agent — topics scope what it’s allowed to handle, and actions are the concrete steps it can take within a topic. For a sales agent those actions are things like querying lead and account records, drafting and sending an email, updating a field, booking a meeting, or triggering a Flow. The reasoning that picks which topic and which action fires is the Atlas engine, and if you want to understand why an agent chooses one path over another, our Atlas walkthrough is the mental model to bring.
Three configuration surfaces do the heavy lifting on the outbound side:
Content and tone. The agent answers prospect questions from approved content, not from open generation, and you set the language and tone it uses. This is your first line of defense against an agent inventing a product claim — it’s the sales equivalent of the grounding discipline in Prompt Builder. If it isn’t in approved content, the agent shouldn’t be asserting it.
Engagement rules. You control how often the agent reaches out, on which channels, and when it engages an inbound prospect versus holding off. These aren’t cosmetic — cadence and channel rules are how you keep an autonomous agent from over-contacting a lead into a complaint.
The email ceiling. This is the operational number most teams miss. Per Salesforce’s Lead Nurturing considerations, each Lead Nurturing agent can send up to 1,800 emails a day on a Gmail account and 9,800 a day on a Microsoft Exchange account. If your inbound volume can exceed that, the agent silently hits a wall — which is a very different failure than a rejected API call, because it looks like the agent simply went quiet.
On qualification, the agent interprets intent signals and can update BANT (Budget, Authority, Need, Timeline) fields in real time as the conversation develops, then hand a qualified lead to a human AE with the context attached. That handoff is the whole point: the agent does the patient, repetitive nurturing at 2 a.m., and a person takes the meeting.
Inside Sales Coach: the agent that plays the customer
Sales Coach inverts the direction — it’s the agent working for your rep, and it’s genuinely one of the more clever standard agents Salesforce ships. Beyond generating deal-specific feedback on a rep’s messaging, its standout capability is role-play: for opportunities in the Negotiation/Review and Proposal/Price Quote stages, the agent takes the role of the customer and lets the rep rehearse, then gives structured feedback on how the exchange went.
Because it’s grounded in the actual opportunity — the stage, the products, the CRM context — the practice isn’t generic. A rep can pressure-test a discounting conversation against an agent that “knows” the deal before they have it live with a buyer who only gives them one shot. The setup mirrors the SDR agent: an autonomous agent user, the Einstein Generative AI + Agentforce + Data 360 prerequisites, and topic/action configuration in Agent Builder, with role-play topics you can customize to your stages and playbook.
The reason Sales Coach is a good second project rather than a first: it’s employee-facing, so a bad turn coaches a rep poorly instead of emailing a prospect wrongly. Lower blast radius, faster iteration, and a great way to build organizational trust in agents before you point one at the outside world.
The credit meter is a design constraint, not a footnote
Sales agents consume like any other Agentforce agent, and outbound sales is exactly the shape of workload that can surprise a budget. On the Flex Credits meter, a standard action runs 20 credits — about $0.10 at list — and an SDR agent nurturing thousands of leads with multi-touch email sequences executes a lot of actions. We laid out the full model in Agentforce pricing explained, but the sales-specific takeaway is this: an agent that sends a three-email sequence, checks engagement, and updates BANT on every lead is doing five-plus metered actions per prospect, and your bill scales with lead volume, not headcount.
That makes topic and action design a cost lever, not just a quality one. Before you scale, model the actions-per-lead on real traffic and run the numbers with our Agentforce ROI calculator — a sales agent that pays for itself at 500 leads a month can look very different at 50,000.
Where a human still has to stand
Autonomy is the selling point and the risk. A few guardrails we’d treat as non-negotiable before an SDR agent touches a real prospect:
Ground it or don’t ship it. An agent that answers product questions from open generation instead of approved content will eventually assert a price, a feature, or a commitment you can’t honor — in writing, to a prospect.
- Cap the cadence and channel deliberately. Set the engagement frequency low and widen it only when transcripts prove the agent isn’t over-contacting. The email daily limits are a ceiling, not a target.
- Keep qualification human-reviewable early. Let the agent update BANT, but have reps spot-check the first weeks of qualifications before you trust the agent to route unattended. Data quality decides this — a sloppy lead record produces a confident-but-wrong qualification, which is why AI-ready data matters even more for sales than service.
- Test the outbound before it’s live. Run the agent against a batch of your hardest real leads in a sandbox and read every email it drafts. Agentforce’s Testing Center is built for exactly this, and outbound is the last place you want to learn a lesson in production.
- Watch it after launch. Autonomous outbound needs the same observability as any agent fleet — reply rates, bounce rates, escalation quality, and the transcripts themselves. If you can’t see what it’s sending, you don’t have a sales agent, you have a liability that emails people.
Takeaways
Agentforce sales agents are not the service pattern pointed at leads. They’re standard, configured agents — fast to stand up, narrower to shape — and the SDR agent in particular changes the risk model the instant it sends the first outbound message. Provision the agent user and Einstein Agent licence first; expect the enablement toggle to cascade supporting features across your org; ground the agent in approved content so it can’t invent claims; and set cadence, channel, and volume guardrails deliberately rather than accepting defaults. Start with Sales Coach if you want a low-risk way to build trust, because coaching a rep badly is recoverable and emailing a prospect badly is not. And model the credit consumption against real lead volume before you scale, because outbound sales is precisely the workload where the meter runs fast.
Understanding the basics
What is the difference between Agentforce SDR and Agentforce Sales Coach?
Agentforce SDR (also called Lead Nurturing) is a customer-facing, autonomous agent that engages inbound leads over email, answers questions from approved content, qualifies them, and books meetings without a rep in the loop. Agentforce Sales Coach is an employee-facing agent that gives reps deal-specific feedback and role-plays the customer so they can rehearse negotiations. One talks to prospects; one talks to your sales team, and they carry very different risk profiles.
Do you build Agentforce sales agents from scratch like a service agent?
No. Unlike the blank-canvas custom agent you might build in Studio for service, the SDR and Sales Coach agents are pre-built standard agents you configure — enable the feature, provision the agent user with the Einstein Agent User License, ground it in Data 360, set guardrails, and tune topics and actions. That’s faster to deploy but narrower in what you can change, so the work is knowing which parts are yours to shape.
How many emails can the Agentforce SDR agent send per day?
Per Salesforce’s Lead Nurturing considerations, each agent can send up to 1,800 emails a day on a Gmail account and 9,800 a day on a Microsoft Exchange account. If your inbound lead volume can exceed that ceiling, plan for it — the agent hitting the limit looks like it went quiet rather than throwing a visible error. Always confirm current limits in the official documentation, since Salesforce revises them across releases.
Weighing whether a sales agent earns its place in your org — or which one to build first? Talk to us about our Agentforce practice. We scope agents to move real pipeline, not to demo.
