The number of AI agents an organisation can safely run is set by how many its accountable humans can govern. Two routes raise that number without deploying first.
When operating units build production software with agentic coding tools and no IT oversight, IT’s order-taking operating model no longer describes how build decisions are made. Where build authority sits has become an operating model design decision.
When autonomous execution runs, gaps in management accountability become AI governance risk. The management tier now governs agents on a mandate written before they existed.
Culture programs stall when accountability, promotion and pay still reward the old behaviour, however well the program is designed. Changing those structures is a decision only the executive can make.
A central AI team recreates Plan-Build-Run for AI agents, turning same-day fixes to agent instructions into queue items and moving accountability away from the evidence.
The governance that approved each AI investment had no way to stop one. Projects meeting their own milestones kept their funding, and no one owned the decision to redirect it.
Enterprise agent platforms now govern agent identity and access. The board’s AI agent governance question is about the operating model decisions no vendor can make for you.
ANZ leaders and employees measured the same organisations and arrived 14.9 points apart on workforce capability. If your AI investment case draws its capability baseline from leaders’ numbers, that gap shapes your delivery outcome.
Naming a successor and evidencing that they are ready are separate governance obligations, and a succession plan that lists names meets only the first.
The length of task frontier AI can complete on its own has doubled roughly every seven months since 2019, a shorter interval than most business case and delivery cycles. Plan-Build-Run disbands project teams at closure, losing what they learned; persistent teams funded against outcomes keep it.
© 2019 All rights reserved
