An agent can run every day, without a single error, and still cost more than it returns.
Running is not the same as paying off. Almost everything measured today measures the first one.
The platform it runs on is paid by the run. The model provider behind it is paid by the run. Neither is paid to ask whether the run was worth it.
Agentell is paid by the run too. The difference is what we are paid to do with it: tell you which runs earn their cost, and which agent should be switched off.
That difference is only worth something if it holds when it costs us. So we are writing it down.
What we will never do
-
We will never sell the thing we measure.
No AI agents, no agent builders, no agent templates — not built, not resold, not licensed, not white-labeled. We use AI models inside Agentell to do the measuring. That is a tool we buy, never a product we sell you.
-
We will never call an agent successful because it ran without errors.
Uptime, completion rate and clean logs describe execution. A result is a ticket resolved, a meeting booked, a case closed — work the business would otherwise have paid a person to do.
-
We will never hold back a recommendation to shut an agent down, including when we are paid for its runs.
Switch an agent off and the platform loses everything that agent was spending. So does the model provider. We lose the fee for watching it. That loss is real, and it is small enough that we can afford to be the ones to say it.
-
We will never measure one platform differently from another.
One set of definitions, applied the same way to Make, to n8n, to an agent you built in-house, and to any platform we partner with. Where the agent runs does not change the number.
-
We will never accept payment to influence a number, a ranking or a recommendation.
No paid placement, no sponsored benchmarks, no revenue share from a vendor tied to what we report about it. Our customers pay us. Nobody else does.
What this means in practice
Every number we report carries the definition behind it: what counted as a result, what did not, and over what period. You can disagree with a definition. You will never have to guess at one.
We run our own agents on Agentell, under this doctrine. When one of them costs more than it returns, we shut it down.
When an agent is not worth what it costs, we say so plainly — whatever it runs on, and whoever we partner with.
Why this is permanent
A promise that can be quietly withdrawn is not a promise.
This page is dated and versioned. If it ever changes, the change and its date are recorded below and the version it replaced stays public. Nothing here gets edited away.
Hold us to it: [email protected].