Off-the-Shelf AI Agents Just Got Real, and Small Businesses Are Already Outspending Big Ones
Two things happened this month that do not usually get put in the same sentence, but should be.
On September 11, Salesforce introduced seven named AI agents (Casey, Paige, Carter, Hunter, Marshall, Piper, and Fin) each built for a specific job: sales, service, commerce, IT and HR, supply chain, customer experience. They sit on top of a company’s existing Customer 360 data and run inside whatever permissions and business rules already exist. You do not design the agent. You configure it and point it at your data.
Around the same time, Ramp published its Summer 2026 Business Spending Report, drawn from more than 70,000 businesses. One line in it undercuts a story most of us have been telling for two years: small businesses, the ones with one to 24 employees, spend $23 per employee per month on AI. Companies with 100 or more employees spend $3. Small firms are outspending large ones on AI, per head, by roughly seven times.
The “SMEs are behind” story was already wrong
We have written before about how the size gap in AI adoption closed faster than almost any business technology in history. The Ramp numbers go a step further: on spending intensity, small businesses are not catching up, they are ahead. That makes sense once you think about who is actually swiping the card. A twelve-person company does not have a procurement committee. The owner or the ops lead sees a subscription that removes an afternoon of work and buys it that week. A 500-person company has the same idea trapped in a budget review three quarters out.
The catch, which Ramp’s own data shows too, is that adoption is wildly uneven even within the small-business bracket. The top 1 percent of spenders are putting in over $7,000 per employee a month. The median adopter spends $12. Being small removes the bureaucratic drag. It does not remove the need to actually decide what you are buying and why.
What the Agentforce launch changes for that decision
This is where the Salesforce news matters beyond Salesforce customers. For two years, “AI agent” mostly meant a custom build: pick a use case, wire up a model, connect it to your systems, test it, babysit it. That is expensive and slow, and most small businesses correctly avoided it. Named, packaged agents that plug into data you already have are a different proposition. You are not commissioning software. You are turning on a feature and watching what it does with your real invoices, real tickets, real orders.
That does not make the packaged agent the right answer for every job, and it is not free of risk just because someone else built it. It still touches your data, your customers, and your existing rules, and it still needs a human checking its output until you trust it. But it changes the first question a small business should ask from “should we build an agent for this” to “does a packaged one already exist for this workflow, and what does it actually do with our permissions.”
The practical move
Before adding another line to that per-employee spend figure, look at the systems you already run: your ERP, your CRM, your accounting package. Ask each vendor directly whether they now ship a packaged agent for the workflow that costs you the most admin time, and ask exactly what data it touches and who can turn it off. Often the agent you need is a setting away, not a project.
If you want a second pair of eyes on which of your existing systems already has an agent worth turning on, our free assessment is built for exactly that conversation.