Salesforce AI playbook: the credit allowance is the product, not the agent list

The clearest signal about Salesforce's AI strategy is not on a Salesforce page. It is on G2, where the product is now listed as Agentforce Sales, formerly Salesforce Sales Cloud. The largest review platform in the category has renamed a CRM after its AI layer.

That is the context for every decision below. Agentforce is positioned as the product now, and it is priced as an add-on for four of the five editions.

OUR TAKE: Salesforce publishes more about its AI budget than most vendors publish about anything, and the budget is the thing to negotiate. Flex Credits are stated plainly by edition. Agentforce for Sales is an add-on from $125 per user per month, which is roughly two thirds of a Core seat. Size your credit requirement before you compare this with anyone, because the allowance is the ceiling that actually binds.

## What is included, and where

Edition	Flex Credits	Agentforce for Sales
Starter Suite	none	available for purchase
Pro Suite	none	available for purchase
Core $195/user/mo	500 thousand per org per year	available for purchase, from $125/user/mo
Advanced $395/user/mo	1 million per org per year	available for purchase
Max $550/user/mo	2.75 million per org per year	included, with Agentforce Coworker

The suites carry no AI allowance at all. None. The page's own footnote is explicit that AI can be added to Core and above, and that you should contact a sales representative for detailed pricing.

So there are three ways to buy AI here. Pay $550 per user per month for Max and have it included. Pay for a lower edition and add Agentforce for Sales from $125 per user per month on top. Or buy a suite and have no AI layer, which is a legitimate choice nobody will suggest to you.

## Credits are the ceiling, not the feature list

This is the part buyers get wrong, and it is the same mistake we see with every metered AI product.

Flex Credits are consumed by agent work. The allowance is per org per year, which means it is a pool shared across your whole company rather than a per-seat entitlement, and it resets annually rather than monthly. A single enthusiastic team can exhaust a pool that looked generous on paper, and the moment it empties is the moment your AI strategy becomes a purchase order. Credits run out. Budgets do not refill themselves.

Three numbers, 500 thousand, 1 million and 2.75 million, are the real product tiers for anyone buying this for the AI. Note what the ratios tell you: the step from Core to Advanced doubles the allowance for roughly double the seat price, while the step from Advanced to Max multiplies it by 2.75 for roughly 1.4 times the seat price. If credits are your binding constraint, the most expensive edition is the one that gives you the most allowance per dollar.

Ask two questions before signing. How many credits does a typical action in our workflow consume, and what happens when the pool is empty. Neither answer is published, and both determine whether your allowance is a year's supply or a quarter's.

What the AI actually does, by the vendor's own naming

The Max card names Agentforce for Sales and Agentforce Coworker. The comparison table carries rows for Agentforce: Lead Nurture, Agentforce: Pipeline Management, Agentforce: Prospecting and Agentforce: Employee Agent Sales Actions, each marked available for purchase at the tiers where it is not included. Alongside them sit Sales Signals and Call Summaries, both also marked available for purchase, and Lead Scoring and Predictive Forecasting, which are listed the same way.

Two observations from that list rather than from marketing copy.

First, the agent catalogue is organised by sales activity, not by technology: nurturing, pipeline, prospecting, and actions an employee-facing agent can take. That is a sensible way to buy, because you can map each one to a job someone currently does by hand and ask whether the credits are cheaper than the hour.

Second, scoring and forecasting, which are the oldest machine-learning features in CRM, are still sold as purchases rather than included. A buyer who assumes predictive scoring comes with a $195 seat will find it in the available-for-purchase column.

## One contradiction worth resolving before you sign

Momentum appears on the Core card as part of what the edition includes. The Momentum row in the comparison table reads available for purchase at the suite tiers and Coming Soon above them. Both statements are published on the same page. Ask which one binds.

If Momentum is part of why you are buying Core, that is a question for your rep with a written answer, not an inference from a card. The same caution applies to Tableau Next, which the Core card names and the comparison row marks as available for purchase lower down.

The honest cost of the AI story

Set the figures side by side. A Core seat is $195 per user per month. Agentforce for Sales starts at $125 per user per month. The editions comparison table adds $25 per user per month for the Web Services API, which is the interface any custom agent integration would use.

Each of those is published, and we do not add them together here, because the sum depends on which of them your quote actually includes. What we can say is that the AI layer is priced as a product in its own right at roughly two thirds of the CRM underneath it, and that the thing being renamed on review platforms is also the thing that is not in the base price.

Compare that with the alternative structure in this category, where HubSpot includes a small credit allowance from its cheapest paid tier upward, 500 at Starter and 3,000 at Professional, and gates which agents exist by tier. The units are different and we will not invent a conversion. The structures are comparable: one vendor includes a little AI early, the other includes none until the top edition and sells it separately below that.

## The one AI feature the cheap editions do include

It would be wrong to say the suites have no AI at all, so here is the precise position. The Starter Suite card names automatic syncing of emails, events and contacts as an AI feature, and that is included at $25 per user per month.

What the suites do not have is a Flex Credit allowance or any of the named agents. So the distinction is not between AI and no AI. It is between AI that runs quietly in the background on your activity data, which you get at the entry price, and agentic work that consumes a metered pool, which starts at Core and is mostly a purchase until Max.

That distinction matters when a rep says the product has AI at every tier. It does, in the first sense. Ask which sense is in your quote.

What the renaming on G2 actually tells you

G2 lists this product as Agentforce Sales, formerly Salesforce Sales Cloud, across a pool of roughly 25,900 reviews. Capterra still calls it Salesforce Sales Cloud, with 18,811 reviews, and both rate it 4.4 out of 5.

Two useful readings of that. The vendor has pushed the Agentforce name far enough into the market that the largest review platform in the category restructured its listing around it, which tells you how the product will be sold to you. And the reviews under that new name were overwhelmingly written about the old product, which tells you not to read the 4.4 as a verdict on the AI layer. If you are buying Agentforce specifically, the aggregate ratings are evidence about the CRM it sits on, not about the agents.

## A governance checklist for procurement

If your company has an AI review process, these are the items that were not answerable from any page we read in October 2026, which means they belong in your questionnaire rather than in your assumptions.

Model provenance: which models run the agents, and whether that can change without notice. Training: whether your CRM data is used to train anything, and with what controls. Retention: how long prompts, outputs and intermediate data persist, and where. Opt-out: whether specific objects, fields or records can be excluded, and who can set that. Region: where inference happens, which matters if you have data-residency obligations. Audit: what log of agent actions exists, since an agent that writes to records is making changes somebody will eventually need to reconstruct.

The vendor publishes a dedicated security and trust hub with sections for multi-factor authentication, security advisories and compliance, which is the right starting point. The compliance subdomain returned an empty page to our US render, so we read no certification list and publish none.

The questions that change the price

Four questions, in the order they affect your bill.

Which edition, and is the AI add-on in the quote or outside it. How many Flex Credits are included, and what the overage mechanism is. Whether the $25 per user per month Web Services API surcharge applies to your users, since any custom agent integration would run through it. And which named agents are included rather than available for purchase, given that Lead Scoring and Predictive Forecasting both appear in the purchase column.

A rep who answers all four in writing is giving you a quote you can compare. A rep who answers them verbally is giving you a demo.

A plan for the first ninety days of AI

Start with one job, not one agent. Pick the task your team does most often that is mostly reading and summarising, since that is where metered AI pays back fastest and where a bad output costs the least.

Measure consumption before you expand. Your credit pool is annual and shared, so the useful metric in month one is credits per completed task, not output quality. Quality you can judge by reading. Consumption you can only learn by running it.

Then decide on edition rather than on add-on. If your measured consumption suggests you will use a large share of your allowance, the arithmetic of the ladder favours moving up an edition over buying credits piecemeal, because the allowance grows faster than the seat price between Advanced and Max.

And keep one question in front of you throughout: would this task be cheaper done by a person at your actual labour rates. Metered AI is not automatically cheaper than the hour it replaces, and the vendor's pricing page cannot answer that for your company.

## What we could not establish

How many credits any specific action consumes. The vendor publishes allowances, not rates, and the footnote directs you to a sales representative for detailed AI pricing. The provenance of the models, whether customer data is used in training, and what the retention and opt-out terms are did not appear on any page we read^[5]. AppExchange, where agent extensions would be found, returned an empty render from the United States on 6 October 2026.

Those are gaps in what is published, and we would rather name them than fill them. If AI governance is a procurement requirement for you, the security and trust hub is the place to start, and the answers will come from your rep rather than from the pricing page.