Human Attention Is Capital

One of the most important lessons I learned while increasing my own use of AI is that human attention has to be treated as capital.

That sounds obvious. In practice, it is violated constantly. A team will spend hours saving a few dollars of software cost. A manager will perform repetitive technical steps because they are “quick”. An automation will be considered successful even though someone must check it every morning.

A five-minute manual action is not necessarily “only five minutes”. If it breaks a focused block of work, requires remembering context, logging into another system and checking whether something succeeded, the real cost is larger.

This became particularly clear as my own workflows moved from manual AI use toward more agentic automation. The early pattern was familiar: AI produced something, then I copied it elsewhere. Another tool needed a command, so I issued it. A process stopped at a technical checkpoint, so I told it to continue. None of those actions was individually difficult. Together, they made me the bottleneck.

Reserve human attention for decisions where judgment changes the outcome

That leads to a useful design test for every AI workflow: does this step require a human because the decision is genuinely human, or because the system has not been designed to proceed safely without one?

• Good human gates: strategic decisions, material financial commitments, privacy and security choices, legal or reputational risks, irreversible releases and genuinely ambiguous evidence.
• Bad human gates: copying text between systems, confirming routine technical success, repeatedly re-explaining context, manually checking a deterministic condition, or pressing “continue” after an expected intermediate step.

The same principle applies beyond automation. If a new AI tool saves an employee 30 minutes but requires a manager to spend 20 minutes reviewing the output, the value may be far smaller than expected.

Ask two questions, not one

The strongest AI systems are not those that create the most impressive demonstrations. They are the ones that quietly stop asking people to do low-value things.

That is the kind of leverage worth measuring.