Do AI Agents Make Sense at 10-100 Employees?
US Census data shows AI use runs 37% at 250+ employees and under 20% at the smallest firms. The constraint is not interest, it is having someone whose job includes figuring it out.
The adoption gap is a size gap
The US Census Bureau runs a survey of American businesses every two weeks and asks, among other things, whether they use AI. Broken down by headcount, the answer is strikingly consistent:
| Company size | Using AI |
|---|---|
| 250+ employees | 37% |
| 100–249 employees | 32% |
| 4 or fewer employees | under 20% |
This is government data, not a vendor survey, and it says something specific: the constraint is not interest, it is capacity. Large firms are not smarter about AI. They have someone whose job includes figuring it out.
If you are in the 10–100 band you are sitting exactly on the slope. The companies above you are pulling away, and the reason is boring — they have staff to spare and you do not.
Which is the actual argument for buying this as a service
The thing a 40-person company lacks is not the technology. It is the person who owns it after launch. That is the single most common reason these projects die, and it is a staffing problem wearing a technical costume.
A done-for-you arrangement is, at bottom, renting that person. Whether that is worth $2,500/month to you depends entirely on how much repetitive volume you actually have — which is what the audit is for, and why we would rather find out you do not have enough than sell you a build that will not pay back.
When the honest answer is to wait
- Nobody has written the process down, and nobody is going to this quarter. Agents cannot be built on a process that only exists in someone’s head.
- The volume is genuinely low. Fifteen invoices a month does not justify a build, and anyone telling you otherwise is selling.
- You are still changing the process every few weeks. Automating a moving target is how you end up maintaining something that was wrong by the time it shipped.
- Your team has not hit the ceiling yet. If nobody is complaining about copying data between two systems, the spend is premature.
Where these numbers come from. Figures are from the US Census Bureau’s Business Trends and Outlook Survey, reference period 14 December 2025 to 3 May 2026. Two caveats the Census carries itself and we are repeating rather than burying: the published story does not state a sample size, and the question wording changed in November 2025 from AI used “in producing goods or services” to AI used in “any business function” — so any comparison across that date is not valid, including ones you will see elsewhere.