Operational Productivity, Part 4: How Companies and Tools Must Rethink

Walk onto a factory floor at 3am and find the one screen that matters. It is bolted to a pillar near the line, and it shows the same eleven tiles it showed last year. Units per hour. Downtime. A red number nobody has time to read. The operator standing next to it does not look at it. She cannot act on a tile. What she needs to know is that the bearing on station four is running three degrees hot and will seize in about forty minutes unless she swaps it now. The screen knows none of that. It was never built for her. It was built so that someone in an office could see her.

That screen is the whole story of enterprise software, and it is about to be told a second time.

The second chance, named plainly

The first telling is the one this series has been tracing. In Part 1, “Two Kinds of Productivity,” we drew the line that most software strategy still refuses to see: management productivity, the software of seeing and reporting, sitting opposite operational productivity, the software of doing the actual work better. For forty years the money went almost entirely to the first kind. In Part 2, “The Dashboard Arms Race,” we followed that money into its natural end state, a competition to measure everything and improve nothing, where the reward for building a good dashboard is a request for three more. And in Part 3, “The Ideas Only Possible with AI and Agents,” we looked at what changes when the computer can finally act at the point of work instead of only display it back to a manager a week later.

This part is where all of that turns into a decision you make with a budget.

Here is the claim, stated the way you will have to defend it to a board. The companies that win the next decade will move their AI budget, their tools, and their attention from the boardroom to the point of work. The ones that build another dashboard will lose to the ones that build an agent for the person on the line. Not because dashboards are evil. Because the value was always in the work, and for the first time the computer can go stand where the work happens.

“

Aim the computer at the work this time, not at the report about the work.

”

That is the second chance. When the personal computer arrived, we had the same choice, and we aimed it at the report. We built the spreadsheet, the presentation, the status deck, and forty years of software that made it easier to describe work than to do it. AI and agents hand us the choice again. The difference this time is that the machine can carry out the recommendation, not just render it. The mistake to avoid is the one we already made once: pointing the most powerful tool of the era back at the boardroom.

How companies should rethink

The playbook is not a technology purchase. It is five shifts in where a company points its instrumentation, its money, and its org chart. None of them require a model you cannot already buy. All of them require you to stop rewarding the reporting layer.

• Measure the work, not the manager. Most instrumentation budgets flow to whatever is easy to count: logins, clicks, dashboard views, tickets closed. The waste that shows up in a profit-and-loss statement does not live there. It lives in yield, in scrap, in inventory sitting on a shelf, in the first-time-fix rate of a service call. Put your sensors and your return-on-investment math where the physical waste is. If your best telemetry is about how people use your software rather than how the work is going, you have instrumented the wrong layer.

• Deploy intelligence at the edge, at the point of work. The person doing the job should get the analysis and the recommendation in the moment, on the machine, in the cab, at the bedside. Not in a report their manager opens next Tuesday. A recommendation has a half-life, and it is short. The value of “swap that bearing now” decays to zero the instant the bearing seizes. Intelligence that arrives after the decision is made is not intelligence. It is a post-mortem.

• Close the loop. Buy and build tools that act within guardrails, not tools that only show. A recommendation the operator has to re-key into three separate systems is a recommendation that dies on the floor. Let the agent file the work order, reserve the part, adjust the setpoint, reroute the truck, inside limits you set and can audit. The line between a dashboard and a co-worker is exactly this: one shows you the number, the other does something about it and tells you what it did.

• Stop the dashboard arms race. Treat every new dashboard request as a diagnosis, not a feature. When someone asks to watch a metric more closely, they are almost always telling you that a real operational problem is going unfixed and the organization has decided to observe it harder instead of solving it. Redirect that developer-time to the work itself. A team that ships one closed loop on the line will beat a team that ships ten new charts about the line, every time.

• Reorganize around the doer. The operator, the driver, the nurse, the technician becomes the customer of your software. Not the executive. That single reframe changes every design review. The first question stops being “what does the VP see on Monday morning” and becomes “what does the person on the line do differently at 3am.” When the doer is the customer, the reporting falls out of the work as a byproduct. When the executive is the customer, the work bends itself to feed the report.

How tools and technologies must change

If companies move the budget, the tools have to be worth moving it to. The category of software that wins the next decade looks structurally different from the one that won the last one. Six shifts define it.

• From dashboards for managers to agents at the point of work. The default unit of enterprise software stops being a screen a manager reads and becomes an agent that stands next to the doer and helps. The deliverable is not a view. It is an action taken, or a decision made better, at the exact spot where the work is happening.

• From batch reports to real-time deterministic guidance. The old model computes overnight and surfaces the answer the next morning, after the outcome is already decided. The new model puts the answer in front of the person before the decision is locked. Timing is the entire product. Guidance that is correct but late is indistinguishable from guidance that is wrong.

• From the computer as a filing cabinet to the computer as a co-worker. For forty years the machine on the desk was a place to store records and generate reports about them. The new machine is a colleague on the line. It watches the same work you watch, notices what you would notice if you had time, and picks up the parts of the job that are pure friction. It participates in the work rather than filing it.

• From software that records work to software that does work. This is the sharpest line of all. A system of record captures what happened so someone can review it later. A system of action changes what happens next. The winning tools cross from the first to the second. Recording is table stakes and always was; doing is the new product.

• Natural-language and voice interfaces, so the doer needs no training and no wall of tiles. The operator with grease on her hands is not going to navigate a six-tab console. She is going to say what she sees and hear what to do, or glance at one line of plain guidance and act on it. The interface that wins at the point of work is the one that needs zero training and zero screens full of tiles, because the doer’s hands and attention are already committed to the job.

• Deterministic analytics wrapped in judgment. This is how you get real-time action without hallucinating your way into a wrecked machine. The hard numbers stay hard: inventory counts, throughput, tolerances, the physics of the process, all computed exactly by ordinary deterministic code that does not guess. The agent sits on top of that, mediating the interface and choosing the action within guardrails. The math is precise and boring on purpose. The judgment about what to do with it is where the model earns its place.

“

The last era’s flagship product was a system of record. The next era’s flagship product is a co-worker on the line.

”

Notice that none of these six is a request for a smarter model. They are requests for a different target. The frontier is already capable enough to draft the recommendation, read the sensor, and file the work order. What has been missing is the will to aim it at the doer instead of at the doer’s manager. That is a choice about where to point the tool, and it is the same choice we faced with the personal computer.

The mistake we are about to have the chance to repeat

The personal computer could have gone to the point of work. It mostly went to the report. We built four decades of software that made it easier to describe, present, and review work than to do it, and we called that productivity while operational productivity, the kind measured in yield and fix rates and inventory that never piles up, barely moved. Part 2 is the full accounting of how expensive that detour was.

The pull to repeat it is strong, because the boardroom is where the budget is signed and the boardroom likes to be shown things. The easy AI project is the one that generates a prettier summary for an executive. It demos well, it threatens no one, and it improves nothing on the line. The hard project, the one that actually compounds, is the one that hands an operator an agent that does part of her job with her.

Your move this quarter

The argument this series has built pays off in a single decision you can make now. Parts 1 through 3 were the case; this is the action.

That is the whole second chance, made small enough to start on Monday. The computer can finally go stand where the work happens. Forty years ago we sent it to the meeting instead. This time, send it to the line.

That is the move for this quarter. The finale of this series, How Work Should Look for the Next Few Decades, follows that one choice forward, because the shape of work in the 2050s is being decided right now by thousands of small decisions about where to point the next tool.