High Output Management: leverage for managers in the age of AI

High Output Management is Andrew Grove's practitioner's book, explaining the work that runs a company as a production process. Grove led Intel; the English edition first appeared in 1983 and the widely circulated edition is from 1995. It is for anyone who produces results through other people, not only for those with the title of manager.

Grove was born in Hungary in 1936, survived Nazi occupation and a communist regime, and emigrated to the United States in 1956. He studied chemical engineering, became an early Intel employee, and served as its president, CEO and chairman.

As Intel grew, it made a major shift from a memory-centred business to a microprocessor-centred one. Meetings, decisions, training and evaluation had to be treated as repeatable work rather than individual instinct.

The book opens with the metaphor of a breakfast factory. To serve eggs, toast and coffee at the same moment while holding quality, you have to design the steps, the capacity, the inventory, the quality checks and the constraint as one flow.

A manager's output is not the manager's own workload. It is the output produced by the organisation they run plus the neighbouring organisations they influence. Looking busy and increasing results are different things.

This is where leverage matters. One correct decision prevents ten pieces of rework; one training session improves months of work; one process change removes a daily wait. Conversely, a vague instruction is negative leverage.

Just as the breakfast is served at the speed of the slowest step, an organisation's results are set by its constraining step. Rather than making everybody equally busy, concentrate improvement where the whole thing is stalling.

A one-on-one is not a status meeting for the manager's benefit. The report brings the agenda, digs into problems, and obtains a decision or support. Short, regular contact catches issues before they become large ones.

How you engage with a person is not fixed by their personality but varies with their task-relevant maturity. On unfamiliar work, increase direction and checking; on work they know well, widen their discretion. The same person needs different support when the work changes.

Grove treated developing his reports as part of a manager's job. Training takes longer than handling the case in front of you yourself, but it is high leverage because it carries into many later decisions.

When you treat AI as part of the team, do not let chat counts, token counts or seat counts become the result. Measure completed work, rework, exceptions, and the number of cases a human had to judge.

Output and constraint differ by department: for sales, the quality and speed of closing; for accounting, the accuracy of the close and its exceptions; for engineering, production value and incidents. AI produces effect only when placed where it resolves the constraint.

What you can do tomorrow is pick, out of one hour you spend as a manager, the single activity that ripples out to the most people and the most future results. Change the design so it amplifies results rather than adding workload.

Measure a manager's result by the output of their team and the groups they influence rather than by how busy they are, and concentrate time on high-leverage activities and on the constraining step.

The degree to which one decision, one training session or one process change ripples out across many people or a long period. There is also negative leverage, such as a vague instruction that amplifies loss.

Measure completed outcomes, rework, exception handling and human decision time instead of how often the AI was used, and concentrate both AI and human improvement on the constraint that stalls the whole organisation.