The Gemba Was Always There. We Just Couldn't See It.

I once spent the better part of a Friday building a PowerPoint presentation about a problem I had already solved. The solution was clear, the data was solid, and I could have explained the whole thing over lunch in about twenty minutes. Instead I spent six hours translating it into a format that a room full of executives, who hadn't been close to the work, could digest in thirty. Nobody asked me to do this. It was just understood. It was, in the parlance of my calendar, "the job."

Lean practitioners have a word for that Friday: waste. Not laziness, not bad intentions, just pure non-value-added activity that consumed time and energy without moving anything closer to the customer. On a factory floor, you'd have circled it on a value stream map before lunch. In an office, we dressed it up in a nice template and called it a deliverable.

Lean's Dirty Secret

The Toyota Production System changed manufacturing because waste on the factory floor was visible. You could watch a worker walk four hundred feet to retrieve a part. You could count the pallets of work-in-process stacking up between stations. You could draw the spaghetti. Taiichi Ohno's famous insistence on going to the gemba, on seeing the actual work in the actual place, worked because the work was there to be seen. Waste revealed itself if you were willing to look.

Knowledge work broke that model. Companies tried to apply lean in offices for thirty years, with mixed results at best. The value streams were too abstract, the handoffs too cognitive, the waste too thoroughly disguised as legitimate activity. That deck I built on Friday wasn't obviously waste; it was "communication" and "alignment" and "stakeholder management." The spec document I wrote so a team in another building could execute a decision I'd already made was "documentation" and "process." We had entire job categories built around what lean would recognize, without hesitation, as the coordination overhead of humans trying to work together despite the fundamental limitation that we cannot share a brain.

A recent piece by Nate Jones put a number on this, arguing that the majority of knowledge work hours exist not to create value but to manage the friction of human collaboration: writing specs so someone not in the room can act, syncing state in meetings, building decks so an executive who couldn't read the primary source can make a call. His specific percentage may be debatable (and I'm skeptical of any precise figure applied universally), but the underlying phenomenon is real. Call it thirty percent or call it sixty; the point is that a substantial chunk of what we've been categorizing as "the role" is actually coordination infrastructure. It's muda in a white-collar wardrobe.

Flow Kaizen, Not Point Kaizen

What makes AI genuinely interesting here isn't that it can automate tasks within this structure. It's that it threatens to dissolve the structure that made those tasks necessary. That's a meaningful distinction. Lean distinguishes between point kaizen, improving a specific step in a process, and flow kaizen, redesigning the flow itself so wasteful steps don't need to exist. Most technology applied to knowledge work over the past few decades has been point kaizen: faster email, better project management software, more collaborative documents. The coordination overhead remained; we just executed it more efficiently. AI agents capable of maintaining context, synthesizing information across sources, and acting on decisions without a human translation layer are pointing at something closer to flow kaizen. The handoff doesn't get faster. The reason for the handoff disappears.

This is where the gemba question gets genuinely interesting. Ohno insisted on direct observation because that's where reality lived, unmediated by reports or summaries or secondhand accounts. Knowledge work has never had a real gemba. The work happened in people's heads, in email threads, in the accumulated institutional memory that walked out the door when someone quit. You couldn't stand in the corner of a conference room and see the value stream. AI systems operating on actual workflows, with visibility into where decisions stall, where information gets re-explained, where the same context gets reconstructed from scratch in meeting after meeting, may be the first technology capable of mapping that hidden flow. The gemba was always there. We just didn't have a way to see it.

The Hard Implication

There's an implication here that's worth sitting with honestly. Organizational structures aren't arbitrary; they're shaped by the coordination limits of the humans inside them. Layers of management, approval chains, cross-functional committees, they exist in part because information has to move through people, and people have limited bandwidth, limited context, and perfectly reasonable incentives to protect their piece of the translation layer. If AI meaningfully compresses that coordination overhead, the organizations built around it will need to change shape. Some of what feels like leadership infrastructure is actually just the tax we pay for human cognitive limits. That's a hard thing to say in a room full of people whose calendars are ninety percent coordination.

The forward-looking version of this isn't pessimistic, though. What remains when you strip the coordination overhead is the work that was always most interesting: the judgment calls, the novel problems, the synthesis of ambiguous information into a decision that nobody else can make for you. Lean at its best didn't just eliminate waste; it forced the remaining work to be meaningful. That's the promise here too, if organizations are willing to look at the gemba they've never been able to see.

The question worth asking this week is a simple one. Pull up your calendar. Look at last week honestly. How much of it was value creation, and how much was translation?