Software agents are quietly rewriting the middle-management job.

For most of the past three years, the assumption inside large organisations has been that generative AI would erode entry level knowledge work first. The evidence from the last twelve months suggests something different is happening.

Interviews with heads of transformation at nineteen enterprises across banking, insurance and professional services point to the same pattern. The tasks most quickly automated are not the ones performed by junior staff. They are the coordination tasks performed by the layer of managers who translate work between them.

Status updates, meeting synthesis, cross team dependency tracking, expectations management with senior leadership. The traditional connective tissue of middle management is precisely the work where a well scoped agent, connected to a company's chat, calendar and ticketing systems, now performs at or above the human baseline.

The organisational response so far has been quiet. Very few of the companies we spoke to have restructured. Most have simply widened spans of control at the manager level, sometimes doubling them, while promising affected managers a redefined role built around coaching and judgement.

Whether that redefinition holds is the open question. Coaching and judgement are legitimate skills. They are not, however, the skills for which most middle managers were originally promoted. They were promoted for coordination, which is the thing the software now does.

The historical analogue is not the loom. It is the introduction of the spreadsheet in the 1980s, which did not eliminate accountants but permanently changed which accountants were valuable. The equivalent transition for middle management is now underway, and it is happening faster than the enterprise HR function is prepared for.