Idea 03 · Human or AI, a working name

Should a human do this?

Budgets, projections and model choices for AI work, in a tool a junior can read.

40 contracts a month · quality bar: nothing goes out unread

Human£2,600

a senior, 40 hours at £65 an hour

AI£2,890

expensive model, plus a senior rereading every line

Next task Human saves £290 a month

The quality bar eats the saving: a senior rereads every line anyway.

This is the product: one worked example, real-looking numbers. Pick a task; the verdict changes with the work.

AI spend is wrong in both directions.

Nobody can say what a task costs when an agent does it, or when a person should.

Model choice is folklore. Why the expensive model here, the cheap one there?

AI budgets get approved without projections and reviewed without accountability.

How it works.

01

Describe the work

Tasks, volumes and the quality bar, in plain words.

02

See the real comparison

Human cost, AI cost per model mix, and where each wins.

03

Get a budget, a policy, a variance view

A budget, a model policy and a monthly variance view anyone can read.

Why this works.

The uncomfortable truth the tool is built on: humans are often cheaper.

Knowing when is the product.

Model mixing is where savings hide.

Cheap model for retrieval, expensive for synthesis; today that decision is made by vibes.

Pairs with visible evals.

One model does the work, another checks it, quality and cost on the same dashboard.

The business.

Team subscription. Pricing scales with the AI spend under management.

Questions.

Which models do you cover?

Every major provider, plus your own rates for people.

Is this a cost tracker?

It projects before you spend, not just reports after.

Who is it for, finance or engineering?

Both; that is the gap it closes.

Humans are often cheaper. Knowing when is the product.

All twenty-four ideas