Pulling the cord

Most automation business cases carry a quiet assumption about leadership: fewer people on the line means fewer people to lead, which in turn means the leadership task shrinks. A reasonable assumption, albeit often a wrong one. In our experience, automation requires more leadership, better leadership, and, ultimately, a whole other type of leadership.

Whereas a manual line hides its problems in the noise of daily firefighting, an automated line surfaces them – faster, more precisely, and in more places at once. While technology can help identify the issue, every surfaced problem is a demand on someone with the authority and the nerve to act.

This is the part of a concept called Jidoka that most of the industry has moved past too quickly. Jidoka is one of the two pillars of the Toyota Production System, alongside just-in-time, and it means, roughly, automation with a human touch. Sakichi Toyoda's loom, patented in 1924, was the first machine to embody it: it stopped itself automatically the moment a thread broke, rather than continuing to weave a flawed piece of cloth.

But the stop was only half the invention. The other half was the human who came to take a look, diagnose, and improve, and the very leadership system that made showing up and solving the problem the norm rather than the exception.

Companies that treat automation as a technical rollout leave that half of the invention to nobody. The companies that actually scale automation and AI take the human side just as seriously as the mechanical side.

In our experience, four organisational conditions move the needle:

1. Decision rights are settled before the first sensor fires
   A stopped line, a flagged defect, or a data outlier can only be useful if somebody holds the authority to act on them in the moment by stopping production or pulling in maintenance without having to ask permission first. Where that authority is left implicit, the system produces alerts that nobody owns. In that case, operators learn to route around the signals instead of raising them, and the plant ends up with excellent visibility into problems no one is allowed to solve.
   
   
2. One leader owns the cell across functions
   A single automated cell touches production, maintenance, quality, and IT/OT at the same time. Functional leadership does not naturally follow it across those borders. Where no single leader is accountable for how the functions work together around the technology, the pilot stays parked in whichever department built it – good in isolation, invisible to the rest of the plant. This is a large part of what pilot purgatory actually is: not so much a technology gap, as an ownership-leadership gap.
   
   
3. Leadership capacity is treated as a constraint
   Plant managers and line leaders already spend most of their day fighting fires. Sustaining standard work, coaching operators, and holding a new way of working in place requires time that does not exist unless someone above them seriously removes something first. Telling a fully loaded leader to "drive the change" alongside everything else is not a plan, and it certainly does not change any mindsets. In practice, this often leaves the leader accountable for the change without sufficient time to lead it.
   
   
4. Interruption is rewarded
   The cord only works if leaders genuinely welcome interruption. Where stopping the line reads as failure, people learn, quietly but quickly, to ignore the signal rather than raise it. That behaviour is set from the top. Operators quickly notice whether leaders quietly tolerate or genuinely welcome interruptions, regardless of what the formal training says.

We have seen this pattern directly. In engagements with manufacturers, earlier initiatives had already fixed most of what could be fixed at the machine level. The technology was never the constraint. Decision rights blurred between shifts, no line owned the standard across its length, and site leadership was too consumed by daily operations to hold any change in place. The organisation was what needed fixing.

AI as signal amplifier

Think of AI as the cord, one that gets pulled continuously rather than just once a shift. A vision system inspecting every unit, a predictive model watching every bearing, an algorithm re-planning the schedule in real time – each generates the same kind of signal the Andon cord once did, just far more of them and far more often. And where the classic cord asked a leader one question – will you please come and take a look? – AI asks harder ones:

• Who is allowed to overrule the algorithm?
  When the model says run and the operator says stop, or the other way around, somebody must hold that call, and everyone on the floor must know who. Without a clear human decision-maker, an AI recommendation can create uncertainty about who carries the operational risk.
  
  
• Who decides what the system learns?
  The loom stopped, and a human diagnosed it. When an AI model quietly adapts instead of exposing it, the abnormality that Jidoka was designed to make visible becomes invisible again and laundered into the baseline. Keeping problems visible is now an active leadership choice.
  
  
• Who protects people from alert fatigue?
  A system that flags everything teaches people to trust nothing. Deciding which signals deserve a human response, and staffing that response, is a leadership design task. If the response process is not designed deliberately, the organisation may end up collecting more data without improving its ability to act on it.

This is why the leadership question sharpens rather than fades as technology matures. A manual line forgave weak leadership design because humans papered over the gaps. An AI-enabled line executes the design it is given, including those gaps. The more decisions the machines and AI make, the more deliberate leaders must be about the decisions they keep.

Five habits that separate the programmes that scale

Across programmes that survive past the pilot and actually scale, five leadership behaviours recur:

• Assign decision rights before the technology arrives, not after the first incident forces the question. This includes the AI-era version: name, in advance, who may overrule the model.
• Give the cell one accountable owner across functions before the pilot starts, not once it needs to scale.
• Free leadership capacity deliberately. Remove something from a leader's plate before adding automation and AI to it. Creating leadership capacity usually requires removing or reallocating existing responsibilities.
• Show up when the cord is pulled – physically, visibly, and first. Presence at the moment of interruption is the only signal the shop floor believes.
• Reward the interruption, not just the fix. Recognise the person who stopped the line before the person who restarted it. In an AI-enabled plant, extend the same principle to the person who challenged the model and turned out to be right.

None of these requires understanding robotics. But each one of them requires something rarer: a leader willing to be interrupted, in public, by a problem, and to treat that moment as an obligation to improve.

The discipline and the nerve

Do it properly, and be brave. It is the kind of thing parents could tell their children before a match or an exam – and it also happens to summarise a century of production leadership rather well.

Do it properly is the discipline: the standards, the decision rights, the ownership, the freed-up capacity.

Be brave is the rest: walking towards the cord in front of everyone, letting the line stop on your watch, owning the stop instead of chasing down who caused it, overruling an expensive algorithm when your judgement says so, and being visibly wrong sometimes in a culture you have made safe enough to allow it.

The cord was never the hard part. Neither, it turns out, is the AI. A century after Sakichi Toyoda's loom first stopped itself, the scarce resource is unchanged: the leader who runs towards the problem.

The machines have learned to pull the cord. The question for your factory is the same as it was in 1924. Who comes when it is pulled?