The work just got cheap. The aim just got expensive.

OpenAI just started selling finished work. Not answers. Work.

On July 9 they launched ChatGPT Work — an agent. Give it a goal, connect your apps, and it comes back with finished work: the deck, the forecast, the account plan. It runs for hours without you.

By OpenAI’s own account, nearly every team inside the company uses it. Sales turned a discovery call into a working proof of concept in 24 hours. Finance closed the month in hours, not days.

The obvious read: AI just got dramatically better at doing work. That read is right. It’s also incomplete.

If the work is nearly free, what’s still expensive?

Here’s the question worth sitting with: if finished work is about to be nearly free, what’s still expensive?

The aim.

Yes, you’ll still check the agent’s work. But can you check work against a target nobody agreed on?

So, before you hand an agent your workflows, ask around your own table:

If I asked three of your executives to name your ideal customer, would I get one answer — or three?

Who owns growth in your company — not marketing, not sales — growth? Can you say their names?

When a dollar goes out the door today, can you follow it to an outcome? And if not, what changes when the spender is an agent?

An agent inherits your answers

An agent inherits your answers. Wrong customer in, wrong campaign out — finished, at scale, on a schedule.

Worth noticing in the same release: OpenAI shipped spend controls for the agent. Even the company selling finished work is asking whether busy is the same as paid off.

Foundation first. Then deploy.

Who in your company decides what the agent works on? Honest answers only.