Loops — the invention of 2026

Ask people what changed in AI this year and they point at bigger models. The more useful answer, for a leader, is the loop.

A loop is simple: give the system a goal, let it act, measure the result against the goal, and let it try again — improving each round until the outcome is good enough. It is the difference between asking for one draft and asking for a result.

Where loops apply

The test is one question: is there a measurable outcome? If yes, a loop can usually help.

• A report that must hit a readability score
• Outreach that is judged on reply rate
• Code that must pass its tests
• A schedule that must satisfy a set of constraints

In each case the human sets the target and the standard; the loop does the iterating that people rarely have the patience for.

What this changes for you

For an executive, the shift is not "AI writes my emails." It is that your judgement — what "good" means, and when to stop — becomes the scarce, valuable input. The machine supplies effort; you supply the standard.

That is also the risk. A loop optimising the wrong measure will cheerfully get worse in a way that looks like progress. Choosing the right outcome to measure is leadership work, and it does not delegate.