Skip to content
Articles

The three Ts · 3 min read

The AI works. The decision does not.

AI use is high. Organisational performance is not. The gap sits between the answer and the decision.

By Amit Srivastava · First published on LinkedIn, 8 September 2026

A folded paper path that splits in two, one branch in ember

At nine in the morning, someone sends you an AI-created document. By 9:10, you have read it twice and replied, “Looks good.”

AI is very good at filling things: drafts, decks, replies and annexes. The page comes back looking complete. That is the water.

Then you hesitate. “Looks good” and “Go ahead” are not the same decision. You are no longer sure which one you made. That is the Air.

The page is full. The decision is not.


Half full

“Half full or half empty” is an old English test of temperament. The optimist sees the water. The pessimist sees the air. Here, both are looking at the wrong thing. One half has been filled. The other still requires a decision.

For AI-assisted work, the water is everything the machine has produced: the words, the structure, the analysis and the polished presentation. The air is the work that remains: checking what the answer assumes, testing whether we want to believe it and considering what the prompt left out.

Leaders have to see both before deciding whether the document is ready to leave the room.


Half Empty

Glean's Work AI Index, published in June 2026, asked six thousand office workers in the United States, the United Kingdom and Australia.

87 per cent use AI at work.
They report saving 11 hours each week.
13 per cent say their organisation is performing significantly better.

Work AI Index · June 2026

AI use is high. Organisational performance is not.

01 · Use

87%

use AI at work

02 · Time

11 hours

saved each week, as reported by workers

03 · Performance

13%

say their organisation is performing significantly better

Three separate findings: use, reported time saved and reported performance. They are not parts of one total. Source: Glean, The Work AI Index 2026, June 2026. Survey of 6,000 digital workers in the US, UK and Australia.

These are three separate findings. Together, however, they show the problem clearly:

Widespread use and reported time savings do not automatically become better organisational performance.

The hours saved may be real. But time saved is an employee-level benefit. The organisation is judged on performance. Filling creates output. Leadership must turn that output into a decision and a result.

Do not argue with the water

The empty half is not a mood, a whim or a point of view. It contains the work no AI model can do for you: What still needs human judgement? or What needs to be checked before anyone acts? or What could make this recommendation fail? or Who owns the next step?

AI can help you investigate these questions. But it does not carry the consequences of the decision. You do.

That is why a finished document can still contain an unfinished decision. Filling is pleasant. Deciding is not. So we do the pleasant one and call it work.

I am not asking anyone to use AI less. Faster drafting can be valuable. The failure begins when fluency is mistaken for readiness.


Today, take one AI-assisted document and ask:

Which half is still air?

If you can do that, use the speed AI gives you. If you cannot, the document may be finished. The decision is not.


Source: Glean, The Work AI Index 2026, June 2026. Survey of 6,000 digital workers in the US, UK and Australia.

More articles