Agent Reliability Index

You signed for a year. You found out in 90 days.

Vendors quote accuracy per step. You buy completion end to end. Those two numbers are not close, and the gap is why half of agent buyers walk away within a quarter.

85.0% per step

Tool calls, handoffs and decisions in one full run.

19.7%

of runs complete end to end

4,016

failed runs / month

1

max steps for 80% success

Compounding is the whole story

At 85% per step, a ten-step workflow fails four times out of five. Nothing about that is hidden. It is just never the number on the slide.

Between half and two thirds of teams who buy an AI SDR abandon it within three months, roughly ten times normal software churn. Buyers now refuse to commit past twelve months, and they are right to.

There is no way to know whether an agent works on your workload until you have already paid for it and lost a quarter.

The index

Collection in progress. We run end-to-end trajectories against commercial agents and publish measured completion rates, not vendor-reported per-step numbers.

No score gets published until it is measured. Leave your email below for the first release.

Run this on your own workload

$900 per evaluation

Two weeks, your real workflows, a measured end-to-end completion rate before you sign anything. Roughly 1.5% of a typical annual contract.