Operational Productivity, Part 5: How Work Should Look for the Next Few Decades

Picture the hospital ward from the start of this series, but move the clock forward a couple of decades. The nurse at the end of a long shift has not typed a note. The charting, the medication reconciliation, the coding, the handover summary, all of it happened around her while she worked, done by an agent that watched the same patients she did and acted on the parts that never needed a human. What she spent her shift on was the part only she could do. Reading a face. Catching the quiet deterioration the monitor scored as normal. Sitting with a frightened family. Deciding.

That is a bet about the next few decades, not a report from them. But it is the bet this whole series has been walking toward, and this final part is where I make it out loud and look as far down the road as I can honestly see.

This is the last stop in the argument. In the first part, We Automated the Report, Not the Work, I split productivity in two and showed we spent forty years buying the wrong half. In The Dashboard Arms Race I traced how the competition to see the work better quietly made the work itself more boring. In The Ideas Only Possible with AI and Agents I got concrete about what changes once intelligence can stand at the point of work. In How Companies and Tools Must Rethink I turned to the builders and buyers who have to point the money somewhere new. Everything before this was diagnosis and plan. This part is the payoff, and it comes with a title I chose on purpose. Not how work will look. How work should look, if we get the second chance right.

The shift, named: the machine takes the toil, the human takes the meaning

Here is the thing that is actually happening, in one line you can carry out of this piece.

For forty years the computer sat above the work and asked the human to feed it. You served the machine’s reporting needs. You filled the fields, updated the ticket, wrote the status, chased the number into the dashboard, so that someone above you could see. The person did the toil of feeding the system, and the system did the easy part, which was displaying what it had been fed.

Run the operational-productivity shift forward a few decades and that relationship inverts. The machine does the toil. The human does the meaning.

“

For forty years the human served the machine’s reporting needs. The next few decades should flip it: the machine does the toil, and the human does the meaning.

”

That is the whole prediction, compressed. Not “AI takes the jobs.” The opposite, if we aim it right. AI takes the toil inside the jobs, the coordination overhead and the busywork the computer age piled onto people in the name of visibility, and hands the human back the part that was the reason the job existed in the first place. The deciding. The caring. The craft. The carrying of consequences. I think this is the direction work should go, and I want to argue it as an engineer argues, by walking the mechanism forward one horizon at a time.

Each horizon below is a bet, stated as a bet. I am not telling you the dates are certain. I am telling you what follows if the second chance is taken, and roughly when the pieces come into reach.

Near term (the next 2 to 5 years): the point-of-work agent becomes normal

The bet: within a handful of years, the doer gets an agent that sits where the work happens, senses what is going on, guides the next move, and acts on the parts that do not need a person. Not a chatbot on a laptop back at the desk. An agent in the loop of the actual task.

Walk the examples, because this is close enough to be concrete rather than science fiction. The field technician arrives at the job and the agent has already read the asset’s history, pulled the right part onto the van the night before, and can see through the camera that the fault is not the one dispatch guessed. The warehouse picker’s route re-plans mid-aisle when an order changes, instead of being locked in by a batch job that ran at 4am. The clerk stops re-keying the same figure into three systems because the agent reconciles them in the background and only surfaces the one exception that actually needs a human call. The nurse gets the deterioration flag early, from the pattern across vitals no single alarm was watching.

Notice what all of these have in common. The intelligence moved down to the point of work, into the hands of the person doing it, and it acts in the moment. That is exactly the capability the first four parts said was missing for forty years, the reason the computer age showed up everywhere except in the productivity statistics. The constraint was never that we lacked computers. It was that the computer could only sit up in the management layer and look down. Once it can stand at the workface, the second productivity is finally in play.

Two consequences follow fast, and both are bets I will make firmly.

The dashboard layer starts to shrink. For decades the analysis lived far from the work, up top, after the fact, because that was the only place it could live. When the analysis moves to where the work is and acts there, a large share of the reporting apparatus loses its job. You do not need a screen to tell a manager the line is drifting if the agent at the line already corrected the drift and logged why. The report was always a compensation for intelligence being in the wrong place. Move the intelligence, and much of the report is redundant.

The unit of production becomes the person plus their agent. Not the person. Not the tool. The pair. A technician with a capable agent is a different economic unit than a technician with a clipboard, the same way a developer with a compiler is a different unit than one with a notepad. This is the near-term seed of everything that follows, and it is where this connects to the idea of the company of one and its fleet of agents: the leverage that a solo builder gets from a swarm of agents is the same leverage that is coming to the operator, the driver, the nurse, the clerk. The frontline gets a fleet too.

Mid term (5 to 15 years): the management-productivity industry contracts

The bet: as the point-of-work agent becomes normal, the entire industry built to see and relay the work starts to shrink, and the org chart flattens with it.

Think about what a large slice of middle management actually is. It is a report-relay layer. Information comes up from the floor, gets summarized, gets reconciled against other summaries, gets packaged for the layer above, and decisions come back down the same pipe. It exists in that shape largely because information could not move and reconcile itself. Someone had to carry it. When agents handle the coordination and the reconciliation, when the floor’s real state is legible without a human compressing it into a slide, the relay layer thins. Not because the people were not working. Because the specific work of relaying is the work the machine is best at absorbing.

I want to be careful here, because this is where the argument could tip into something ugly, and I do not mean it that way. The flattening is not a headcount bonfire. It is a change in what the layers between the doer and the decision are for. Fewer layers whose job is to move information. The distance from the person doing the work to the person deciding about it gets shorter, which is good for the work and good for the doer.

And new roles appear, because they always do. Three that I would bet on specifically:

Outcome owners. People accountable for a whole result end to end, from intent to shipped reality, with a fleet of agents under them doing the execution. This is the operational version of the company-of-one model, applied inside the firm. The value is in owning the outcome, not in performing the tasks.

Agent orchestrators. People whose craft is directing fleets of agents well: specifying clearly, decomposing work, wiring agents to the real systems, knowing which model does which job. The conductor, not the player.

Verification and judgment specialists. People whose entire value is catching what the agents got wrong before it ships, and making the calls the agents are not allowed to make. As production stops being scarce, checking becomes the bottleneck, and the person who verifies well becomes more valuable than the person who produces fast.

Here is the part that surprises people, and the part I feel most strongly about. Frontline and operational work gets revalued upward, not down. A frontline worker with a fleet of agents is not a cheaper version of today’s frontline worker. They are high-leverage. One skilled operator directing a fleet can hold an outcome that used to need a department. When the leverage lands in the hands of the doer, the doer’s judgment becomes the scarce input to a much larger output. That is a raise in what the role is, even before it is a raise in what the role pays.

“

For forty years the org chart got taller to move information. The next stretch should make it shorter, because the information moves itself, and the frontline worker with a fleet becomes the high-leverage unit, not the low one.

”

Long term (15 to 30 years): the default relationship flips

The bet, and the biggest one: over a couple of decades, the default relationship between a human and a computer at work inverts completely, and work reorganizes around what only humans should do.

Today the default is still the old one. The human serves the machine’s reporting needs. You spend a real fraction of your working life feeding systems so they can display your work back to someone. The long-term bet is that this becomes as strange to our successors as hand-cranking an engine is to us. The machine does the toil. The human does the meaning. That becomes the resting state of work, not the aspiration.

If that flip happens, work reorganizes around a short list of things that were always the point and that we let the reporting era bury:

Deciding what is worth doing. An agent optimizes for what you asked. It cannot tell you what deserves to be asked. Setting the goal, choosing the tradeoff, knowing when the obvious move is a trap, that stays human because it is the part that holds the goal above the task.

Caring for other people. The nurse sitting with the family. The teacher who sees which kid is about to give up. The colleague who notices you are drowning. Care is not a workflow the machine absorbs. It is closer to the reason the workflow exists.

Taste and craft. Knowing what good looks like before it exists, and recognizing its absence when it comes back. When production is no longer scarce, taste becomes the bottleneck, because generating ten options is free and choosing the right one is everything.

Carrying the consequences. Someone has to own the result, stand behind the call, be accountable when it goes wrong. An agent cannot carry a consequence. Accountability is a human thing, and in a world of infinite cheap production it becomes more important, not less.

This is not a soft, hand-wavy future. It is a specific claim: strip the toil out and what remains is denser with the things people actually want from work, judgment and care and craft and ownership, and thinner in the parts nobody ever wanted. That is what I mean when I say work should re-humanize. Not that it gets easier. That it gets more human.

The failure mode: we aim it at the boardroom again

Now the warning, because the “should” in the title is doing real work. None of this is automatic. There is a way to waste the second chance, and it is the most likely way, because it is exactly the mistake we already made once.

We point AI at management productivity again.

Picture it, because it is easy to picture. The agents get pointed at the boardroom, not the workface. They write the deck nobody reads, faster. They generate the dashboard nobody acts on, in real time, with a chat interface. They summarize the summaries and reconcile the reports and produce a beautiful, animated, AI-narrated view of a business that is still working exactly the way it worked in 2005. We automate the report again, one more time, with a better engine. And in 2040 some economist writes the update to Solow’s line: you can see the AI age everywhere but in the productivity statistics.

That is the failure mode, and it is not doom. It is worse than doom in a way, because it is boring. It is another lost productivity decade where the tools got smarter and the work stayed dumb, where the frontline job still resembles the frontline job of twenty years earlier, and the whole gain was captured as a nicer view from the top floor.

There is a sharper version of the risk hiding inside it. When you point powerful automation at producing output nobody understands, you build up what the software-factory writing calls comprehension debt. Work gets shipped, decks get generated, systems get changed, and no human on the team actually understands what was done or why. Aim agents at the management layer and you scale that debt. You produce more representations of the work, faster, that fewer people comprehend. That is not progress. It is the old mistake with a bigger engine bolted on.

The future I am arguing for

I titled this “should” instead of “will” because I do not think the outcome is decided, and I do not want to pretend otherwise. So let me state the position flat, as opinion, which is what it is.

The future of work I am arguing for is this. Give the toil to the machines. Give the judgment and the care to the people. Put the doer at the center, not the report. Let the intelligence stand where the work is, so the person doing it spends their hours on the part that was always the reason the work mattered, and let the machine carry the busywork we spent forty years mistaking for the job.

That future re-humanizes work. It makes the frontline high-leverage instead of low. It flattens the distance between doing and deciding. It hands people back the deciding, the caring, the craft, and the weight of the outcome, which are the parts that were never the machine’s to take.

But it depends entirely on where we point the tools, and that is not decided in a keynote. It is decided in ten thousand small, boring choices about which problem the next agent gets aimed at. The person choosing is often you. The next time you specify an agent, buy a tool, or scope a project, you will point it somewhere. Point it at the work.