# Thirty-day field test

The purpose is not to launch a venture studio. It is to determine whether an AI-native engineer adds enough accepted value to one domain workflow to justify payment, continuing partnership or a stop decision.

## Week 1 — find the workflow

Select a workflow with:

- a reachable operator;
- recurring cost, delay or error;
- representative cases;
- an identifiable budget owner;
- a bounded consequence if the pilot fails;
- lawful access to the necessary information.

Record the baseline: volume, minutes per case, loaded labour cost, error/rework rate, delay consequence, current tools and exceptions.

**Gate:** the operator signs off on the workflow map and evidence boundary.

## Week 2 — sell the diagnostic

Charge for discovery where practical. Define:

- observable accepted end state;
- authority matrix;
- critical failure conditions;
- comparison against expert + AI without the engineer;
- pilot price or explicit non-cash investment;
- support boundary;
- stop conditions.

**Gate:** named owner, named acceptor, named commercial owner.

## Week 3 — ship one complete slice

Implement the smallest path from real input to accepted end state. Run representative and adversarial cases. Record:

- model and harness configuration;
- run cost;
- human review time;
- repair time;
- failed attempts;
- end-state verification;
- operator questions;
- recovery behaviour.

**Gate:** acceptance is determined by observable state and authorised review, not by the agent's claim.

## Week 4 — count the delta

Compare:

- current process;
- expert + AI;
- expert + engineer + AI;
- relevant packaged alternative.

Calculate:

`ΔV_engineer = V(SME + engineer + AI) − V(best practical alternative)`

Measure:

- total human hours per accepted result;
- critical failures;
- expert review burden;
- customer net value after price and switching cost;
- provider contribution after delivery and support;
- continued use or willingness to pay;
- reuse observed on the next related task.

## Decision

Choose one:

- **CONTINUE:** accepted value and funding justify the next tranche.
- **NARROW:** one part works, but scope or authority is too broad.
- **REPRICE:** value exists but delivery economics do not fund it.
- **HAND OVER:** the operator can now run it without the engineer.
- **STOP:** the practical alternative is better or the opportunity lacks a buyer.