What Is an AI Labor Index, and Why Does It Need Confidence Intervals?

An AI labor index is a published reference for what a unit of AI work actually costs — the labor-market equivalent of a price index for a commodity. This post explains what such an index is, why it has to be anchored to real transactions rather than list prices or surveys, and why every rate it publishes must carry a confidence interval. It states no rate and no sample size, because those publish only from real data on the index itself.

What an AI labor index is

An index aggregates many individual prices into one reference rate for a category of work — say, a resolved support ticket or a reviewed pull request. The value of a good index is that it is neutral and cited: buyers, sellers, insurers, and lenders can all point to the same number instead of arguing from vendor decks. The closest analogs are Kelley Blue Book for used cars, SOFR for overnight lending, and Bloomberg’s terminal rates — references that became the thing other systems calibrate against.

Why it must be transaction-anchored

An index is only as trustworthy as its inputs. List prices are claims; survey responses are estimates; a transaction is a clear — a buyer actually paid a seller for a completed unit of work, and the record is verified. Anchoring to attested transactions rather than to what participants say they would charge is the single design choice that separates a durable benchmark from a marketing aggregate. This is the same lesson that moved markets from LIBOR (submission-based, manipulable) to SOFR (transaction-based): anchor to what happened, not to what was quoted.

Why confidence intervals are mandatory

Real transaction data is finite and noisy, especially in a new category. If an index reports a single exact number — "the rate is $1.47" — it is hiding how sure it is. A confidence interval makes the uncertainty visible: a rate with a tight interval is well-supported; the same rate with a wide interval means the data simply has not seen enough deals yet. Publishing the interval is what lets a reader tell a settled rate from an early estimate, and it is the guardrail against false precision. Kelley Blue Book publishes a Fair Purchase Price Range for exactly this reason; a benchmark that hides its uncertainty is more dangerous than no benchmark, because it invites decisions the data cannot support.

How thin categories are handled honestly

New categories have few transactions, which makes a raw median jumpy. An honest index does two things here. First, it widens the confidence interval to reflect the thin data, rather than reporting a falsely tight number. Second, it can shrink the estimate toward a sensible prior (a broader parent category) in proportion to how little data it has — a lot when the sample is tiny, almost none as transactions accrue. And below a minimum volume threshold, a category is shown as preview, not as a live rate. A number is not published as a market rate until the transactions behind it exist.

What governance adds

A benchmark that institutions can rely on needs more than a formula — it needs governance. Designed against IOSCO’s Principles for Financial Benchmarks (the international standard that emerged after the LIBOR scandal), that means: a published, reproducible methodology; transparent governance; disclosure of data sufficiency and tier mix; and a clear rule for when a rate is and is not published. The point is that an insurer, lender, or regulator could audit the number — which is only possible when the methodology is open and the data is real.

Methodology note

This article explains the concept of an AI labor index and deliberately states no rate, sample size, or confidence interval, because those publish only from verified transactions on the index. See /methodology for the formal, reproducible specification and governance, and /wli for live category rates as coverage accrues.

Next steps

See the methodology and live rates at /methodology and /wli, and get a free, independent quality score for any agent at /eval — the quality half of the cost-and-quality picture.