Pricing power in a 25-agent company
What running on agents does to margins, and how a buyer should think about valuing a company built that way.
By Ali Asif,
I run about two dozen agent roles across Northlane and my other work. They're written roles with memory files, run through Claude Code, and a person approves every price and every outbound send. The operating model took roughly 90% of the manual operations work off my plate. That figure is an operating estimate, not an audited one, and I keep the tilde on purpose.
What does that do to the economics of a business? I've thought about this as an operator and as someone who used to look at other people's numbers for a living.
Where the margin comes from
Services and bid businesses carry a lot of cost in people who read, extract, chase and format. A tender pack arrives, someone reads 99 pages, someone builds a bill of quantities, someone emails 20 suppliers, someone tracks replies. Each step is skilled but repetitive.
When agents do the reading and chasing, that cost moves from payroll to usage fees plus a small amount of supervision. Payroll scales with the number of bids. Usage fees scale too, but at a far lower rate per bid. So the cost of handling one more tender drops, and the people left do judgment work: the price, the risk call, the relationship.
I've seen the effect on speed first. Supplier quote turnaround went from days to hours. Speed shows up in revenue before it shows up in the cost line, because a fast reply wins work a slow one loses.
Pricing power is a separate question
Cheaper production doesn't automatically mean you can charge more. It gives you a choice.
You can hold your price and take the margin. Or you can cut your price and win volume. In a market where buyers compare quotes line by line, I lean toward the second in the short run. Where your edge is speed and accuracy, the first becomes possible, because the buyer is paying for certainty. The right answer depends on the segment, and I'd test it bid by bid instead of picking a policy.
What does change is the floor. A bid that was marginal at my old cost base becomes viable. That widens the set of tenders worth pursuing, which is a growth lever and a margin lever at once.
What a buyer should check
If someone offers you a company that says it runs on agents, I'd ask for these before looking at the multiple.
Hours per unit of work, with logs. A claim like 90% is a story until it's backed by time per bid, before and after. I haven't published mine because I don't yet have clean logs, and I'd mark down any seller who offered the number without them.
Who signs. Where does a human approve? If the answer is nowhere, the company has a liability, not an efficiency.
Model dependence. What happens if a provider changes pricing or retires a model? Good operators write roles in plain language so they can move.
Where the knowledge lives. In my setup it lives in Markdown memory files I own. That is portable. Knowledge trapped in one vendor's tool is not.
Error rates. Agents misread things. On one scanned tender a single pass agreed on only 30% of part numbers, and it took voting across independent readers to reach usable data. A company that has found and fixed failures like this is worth more than one that says it has none.
How I'd value it
The old logic for a services company was revenue per head. For an agent-run company I'd look at revenue per supervised hour, then discount for model dependence and key-person risk. A company with two people and strong logs can be worth more than a company with twenty and weak ones. It can also be worth less if everything depends on one founder's judgment, and in my case that's a fair criticism.
I don't think agent-run companies deserve a premium by default. They deserve a closer look at the evidence, and a buyer who asks for it will usually find out quickly which kind of company is on offer.