On Toyota’s assembly lines, any worker can pull an andon cord when they spot a problem. A team leader responds, and if the issue can’t be fixed quickly, the line stops. It’s a way to keep quality or safety problems from compounding, and it gives the people closest to the work real power over the integrity of the process.
I think knowledge work needs its own version.
AI can help employees see across an entire workflow, spot what’s broken, and take on work that once spanned several roles. Yet many still sit in narrow jobs, with layers of approval between noticing a problem and being able to fix it.
That becomes a much bigger issue when AI can act on its own. If an employee sees an agentic process exposing sensitive information or making decisions based on bad assumptions or reasoning, they need to be able to intervene before the problem spreads.
Satya Nadella has called for an "emergency brake" that allows an authorized person to pause or shut down a model mid-task. But a switch is only useful if someone knows when to use it and feels trusted to make that call. We need employees to understand that protecting safety, security, and privacy is part of their job—and that they have the authority to act.
For leaders, that means deciding in advance what people can pause, how concerns get resolved, and who approves a restart. It means backing someone who pulls the cord in good faith.
Building this kind of human control into a business is part of what we call a "hill climbing machine": a learning loop where a company's human capital and token capital compound, and each attempt makes the next one better. Satya’s clear about what drives it: "Without human direction, you have compute running in circles." As we give AI more autonomy, we’ll have to give employees more agency. That’s how we will get the best of AI capability, paired with human judgment.
