Agentic Layer Check
A genuinely good idea for pricing AI by outcome sits inside the most fragmented packaging surface of any major platform vendor, and a three week old partnership just added a construct Salesforce itself doesn't invoice for. Here's how it reads against the nine layers.
What they sell
The Salesforce Customer 360 platform (Sales, Service, Data, and Agentforce) now extending into Claudeforce, a joint interface and infrastructure layer built with Anthropic.
Who they sell to
Enterprise and mid market revenue, service, and IT organisations, sold across a stack of editions and add ons rather than to one buyer holding one number.
How they price
Per user Sales/Service editions, a legacy $2 per conversation Agentforce charge being phased out in favour of Flex Credits ($0.10 per action), three separate per seat AI tiers ($125 to $550+ per user per month), and, since August 2026, a Claude inference contract billed directly by Anthropic, outside all of the above.
One real fix, and Salesforce deserves credit for making it after the criticism, not instead of it.
Flex Credits replaced a flat $2 per conversation charge that drew real criticism for billing the same regardless of what an agent actually did. The fix: $0.005 per credit, applied consistently whether the action is a $0.10 employee onboarding step, a $0.15 voice interaction, or a $0.60 field service scheduling flow. Same rate card, different outcomes priced proportionally. That's a real step toward "one number, one outcome," and Salesforce shipped it in response to buyer feedback rather than defending the original model.
Two findings here, not one, and neither is the smaller story.
A buyer configuring Agentforce today chooses upfront between the legacy per conversation charge and Flex Credits. Salesforce's own documentation states the two "will not be supported in the same org," an architecture decision made once and hard to unwind. On top of whichever they pick sits three separate per seat AI tiers: a $125 to $150 per user per month add on granting unmetered usage, a $5 per user per month metered license requiring credits bought separately, and a $550+ per user per month bundled edition combining both with an annual credit allotment. Five constructs for one product line, and a customer has to understand which stack with which before they can read their own invoice.
Claudeforce, announced August 26, 2026, layers Claude (via Amazon Bedrock, inside Salesforce's own security boundary) across Agentforce, a new "Salesforce in Claude" plugin, and Slack. No pricing has been published for any of it. What is public: customers contract separately with Anthropic for the token consumption Claude generates, a bill that runs alongside, not through, Flex Credits or seat licensing. It's a defensible engineering choice (Anthropic owns Claude's inference economics, not Salesforce), but it means the newest and most hyped piece of the Agentforce story is the one construct Salesforce can't fully price, forecast, or even see the total cost of on its own dashboards.
Neither finding stays contained to where it was announced, and together they compound. Layer 8 (Revenue Architecture) now has to reconcile two vendors' invoices to describe one customer's AI spend. That's a genuinely new problem, since even five internal SKUs were at least all Salesforce's own revenue to model. Layer 1 (Organisation Design) inherits a stranger question neither company is answering yet: if the daily surface a rep actually touches becomes "Salesforce in Claude," what is the $125 to $550 a month Salesforce seat a license to? And it's worth noting who doesn't fill a quota on any of this: a rep comped on seat upgrades (Layer 5) has no incentive to push a $0.10 Flex Credit action or a token bill they never see the invoice for, no matter how well either actually works for the customer. None of these costs show up on a pricing page. They show up on two invoices, in a renewal conversation, and in a comp plan that hasn't caught up.
Part of the Agentic Layer Check series: named companies, read against the nine layers. Want your own read? Request a teardown →
This is a public signal read: Salesforce's own Agentforce pricing page, its May 2025 Flex Credits announcement, and its August 2026 Claudeforce announcement with Anthropic, not a private briefing or company input. It reflects what's publicly documented as of the date above; pricing pages and partnership terms change, and this piece will note it if either does. Salesforce and Anthropic were not consulted on or paid for inclusion; see the disclosure approach this work is run under.
Sources: Salesforce: Agentforce pricing → · Salesforce: Flex Credits announcement, May 2025 → · Salesforce & Anthropic: Claudeforce announcement, August 2026 →