Salesforce sells one agent (Agentforce) in eight different ways:
Foundations, free
$2 per conversation
Flex Credits, $500 per 100,000 credits, with every action consuming a different number of them
$5 per user per month, bundled with credits
An add-on at $125 per user per month
Another add-on at $150 per user per month
Agentforce 1 Editions, from $550 per month
$2 per Help Agent resolution
Eight public options on one product, from the company that taught the industry how to price software.
Pricing was never easy. Put an AI inside the product and the old model stops working, and nothing has replaced it yet.
**The question of what to charge for, when the product does the work instead of helping a person do the work, has no settled answer yet. **
👇 Last time I wrote about how much to charge and what belongs in each tier. You can find it here 👇
Today’s issue is about choosing the pricing model itself.
With an agent in the product, one model stops being enough. There are two choices underneath it:
What you charge for: access, activities, outputs or verified outcomesHow the customer pays for it: subscription, pay as you go, prepaid credits, committed spend.
Almost every AI product now combines them.
So most teams guess, and they anchor the guess to their token bill, which is the one method guaranteed to underprice them. On AI products the median target gross margin is now about 50%, against the 75% to 85% SaaS was built on. At those margins the guess is not a rounding error.
📚 By the end of this guide you’ll have a Claude agent that designs the whole thing. It interviews you about what you sell and who buys it and recommends the combination that fits your business.
Inside you will find:
Why One Pricing Model Stopped Being Enough
How to Set Up the Claude Agent
How to Decode What Your Competitors Actually Charge For
How to Cost One Unit of Work Before You Price It
How to Choose the Combination That Fits Your Business
How to Set the Credit Weights and the Price per Credit
How Many Credits to Include and What Happens When They Run Out
Per seat had a good run because for twenty years it was approximately correct. Software helped a person do a job, more people doing the job meant more value delivered.
*An agent breaks the chain in the middle. *The product no longer helps a person do the job, it does some fraction of the job. So founders go looking for a replacement, and this is where the confusion starts, because two different questions get answered with the same word.
The first question is what you charge for.
Inputs: tokens, compute, storage. What the infrastructure providers sell.Access: a seat, a license, a workspace. What SaaS sold for twenty years.Activities: an action run, a document processed, a query executed. What most usage-based AI products sell today.Outputs: a completed piece of work, delivered. A drafted contract, a reconciled invoice, a resolved ticket.Outcomes: a verified business result. A ticket the customer never reopened. A recovered chargeback.
Value alignment goes up as you move down that list, and so does the measurement burden.
The second question is how the customer pays. Subscription, pay as you go, prepaid credits, committed spend, top-up packs. And more importantly:
Credits are a payment mechanism, not a pricing metric.
You do not choose between “outcome pricing” and “credit pricing.” You choose outcomes as the metric and credits as the way they get paid for.
→ Which means most AI products need a stack rather than a single model:
The rest of this guide is how you fill in each row of that table for your specific business, with numbers you can defend.
The rest of the issue is available exclusively to paid subscribers!
Paid subscribers unlock the full library: 200+ decks, 10,000+ investors, AI Agent playbooks, and fresh high-signal content every week.
Not subscribed yet? Try it free for 7 days and unlock this guide, plus our full premium library:
We’re already 80,000+ founders and operators inside!
Let’s begin!
