Reshape the Work Before You Reduce the Workforce

Reshape the work before you reduce the workforce. Most boardroom conversations about AI collapse into a single question within minutes. How many people can we let go? It is an understandable instinct because the technology is expensive, the pressure to show a return is real, and headcount is the line on the P and L that leaders feel they can actually move.

But the instinct mistakes the nature of the change that is underway, and it pushes the hardest, least reversible decision to the front of the queue when it belongs at the back. This article argues that AI's first and most consequential effect is not on the size of the workforce but on the shape of the work. It examines where value really moves when AI enters a business, why workforce decisions taken too early tend to destroy capability rather than create it, and what sequence of choices separates the organizations that compound an advantage from those that simply trim a cost.

The reflex that gets the order wrong. The reflex to read AI as a headcount story is not irrational, it is just early. When a capability arrives that can draft, summarize, code, and answer at scale, the simplest financial response is to ask what it lets you stop paying for. That framing treats AI as labor arbitrage, a cheaper way to do the same work with fewer hands.

The trouble is that it assumes the work stays the same while the workers change, and that is almost never what happens. What the evidence shows instead is that the work itself is the thing that moves first. McKinsey's 2025 Global Survey found that organizations seeing real financial impact from AI were nearly three times as likely as their peers to have fundamentally redesigned their workflows rather than layering tools onto existing ones. The lesson sits underneath the number.

The value did not come from the tool, it came from changing how the work was done. Headcount was a consequence of that redesign, not a substitute for it. Work changes shape before it changes size. It helps to separate two ideas that boardrooms routinely merge.

The first is the size of the workforce, the number of people on the payroll. The second is the shape of the work, which tasks are bundled into which roles, where the handoffs sit, who reviews what, and where judgment is required. AI acts on the second long before it forces a decision about the first. When a model absorbs the routine drafting in a marketing team or the first pass triage in a service center, it does not simply remove a slice of labor, it rebundles the remaining tasks, moves the points at which a human needs to check the output, and creates new work in the form of prompting, reviewing, correcting, and governing what the machine produces.

MIT Sloan's workflow research makes the structural point plainly. AI's largest effects show up in how tasks are sequenced and handed off, not in the single task in isolation, because the coordination between human and machine is where time is won or lost. This is why the most useful question is not which tasks can we automate, but what should this work look like once the machine is in it? The answer reshapes roles.

It does not just shorten task lists. And a role that has been reshaped is not the same as a role that has been removed, which is the distinction the layoff reflex misses. Why automate and cut disappoints? A second reason to slow down is that productivity and value are not the same thing, and AI produces the first far more readily than the second.

It is now common to see teams generating output faster, closing tickets quicker, drafting in minutes what used to take hours. Yet speed in one task does not automatically become margin on the P and L, because the time saved has to be captured, redeployed, or converted into something a customer will pay for. The gap is visible in the returns. IBM's research found that only about a quarter of AI initiatives had delivered the returns leaders expected, and far fewer had scaled across the enterprise.

The point is not that AI fails to make work faster, it plainly does. The point is that local productivity leaks away unless the operating model around it changes, which is precisely the redesign a headcount first approach skips. The reshape sequence. If the mistake is sequencing, the remedy is a better sequence.

The pattern that separates the organizations compounding an advantage from those merely trimming a cost can be captured in four ordered moves, which I will call the reshape sequence. The order matters more than any single step, because each move depends on the one before it. The first move is to redesign the workflow. Before deciding who does what, leaders map how value actually flows through the work, where the friction sits, and what the process should look like once a machine carries the routine load.

The second move is to redefine the human role. Once the workflow is clear, the question becomes what humans are uniquely needed for within it. The judgment, the exceptions, the relationships, and the accountability that the machine cannot hold. The third move is to rebuild the surrounding system.

New workflows and roles only stick if the controls, incentives, career paths, and performance measures are rebuilt to match them, because an organization that rewards individual throughput will quietly resist work that now depends on review and collaboration. Only after those three moves does the fourth arrive. Resize the workforce. By then the decision is informed rather than speculative, because leaders can see which capacity has genuinely been released and which has merely been relocated.

Read in order the four moves, redesign, redefine, rebuild, resize. Put the hardest and most irreversible decision last, where the evidence suggests it belongs. Most organizations instinctively run the sequence backwards, starting with a headcount target and reverse engineering the work to fit it. That is the surest way to cut capability you will need and keep work you should have changed.

The bottleneck moves, it does not vanish. One reason resizing first is so risky is that AI rarely removes a constraint outright. More often it moves the constraint somewhere else and the new location is easy to miss. When generation becomes cheap and fast, the scarce resource is no longer the production of work, but the verification of it, checking that the output is correct, safe, compliant, and fit for the decision it will inform.

McKinsey's case study of the software firm Sonar shows the shift cleanly. Once AI accelerated the writing of code, the binding constraint moved from generation to verification, and the value of reviewing, testing, and assuring quality rose accordingly. The team did not need fewer people so much as a different distribution of effort, weighted toward the judgment that now governed whether the faster output could be trusted. This is the deeper reason judgment becomes more valuable, not less as AI spreads.

The more an organization can produce, the more it depends on the humans who decide what is good enough to ship, what to escalate and what to override. Cut those people in the name of efficiency, and you remove the very capacity that makes the new volume safe to use. The new work that appears around the machine. The same dynamic explains why new work tends to appear around AI faster than old work disappears.

Someone has to design the prompts, supervise the agents, evaluate the outputs, own the quality, and govern the risk. And these are not the jobs that existed a year ago. They are roles created by the technology rather than destroyed by it. The hiring data already reflects this.

IBM found that more than half of chief executives were recruiting for AI-related roles that did not exist twelve months earlier, from agent supervision to evaluation and governance. A workforce plan that counts only the roles AI might shrink while ignoring the roles it is busy creating will arrive at a misleadingly negative number. Why human contact becomes more valuable, not less. There is a particular category of work where the layoff reflex is most tempting and most mistaken, the human contact at the front of the business.

As AI absorbs routine queries and standard communication, the obvious move is to thin out the people who used to handle them, but the more machine-generated communication a customer encounters, the more they value the moments when a capable human steps in. Trust is the reason. Salesforce's customer research found that only around four in ten customers trust businesses to use AI ethically, and that buyers still prefer human expertise for the complex, high-stakes decisions that matter most to them.

The implication is not that human service survives by accident. It is that human contact becomes a deliberate, scarce, and increasingly valuable point of reassurance, best reserved for the situations that are emotional, ambiguous, or consequential. Best Buy's redesign of its customer support, built with Accenture in Google, shows the shape of this in practice. Routine search and self-service moved to a generative assistant, while human agents were equipped with real-time guidance and freed to spend their time understanding and reassuring customers.

The headcount question was secondary to the design question, which was where human attention created the most value. The quiet danger of cutting the bottom rung. A further hazard hides in the part of the organization that looks easiest to cut. Junior roles are often the most exposed to automation because they contain the most routine work, which makes them the obvious target when a leader is hunting for savings.

The danger is that these roles are also where judgment is formed and future leaders are made. PWC's analysis found that entry-level roles in the most AI exposed sectors have become more demanding, not less, with the share of seniorized junior roles growing by more than a third since 2019. Strip out the bottom rung to save cost, and you do not just lose cheap labor, you break the ladder that produces the experienced people whose judgment the rest of this argument depends on. The apprenticeship that turns a graduate into a trusted decision maker does not happen if the graduate role no longer exists.

What happens when you resize first? Put these threads together and the cost of running the sequence backwards comes into focus. Resize first, and you risk cutting the verification capacity that makes AI safe, the human contact that protects customer trust, and the junior pipeline that produces tomorrow's judgment. You also commit to an irreversible decision while the workflow is still in flux, which means you are optimizing a structure you have not yet designed.

The labor market evidence offers some reassurance that there is time to get this right. A detailed study of rapid AI chatbot adoption across firms in Denmark found no measurable effect on earnings or hours after two years, even though the work underneath was clearly being restructured. The honest reading is not that AI will never affect workforces, some reduction is coming, but that the immediate and strategically decisive question is how the work is being reshaped and who gets redeployed, retrained, and elevated as a result.

What this means for leaders. The temptation is to treat AI as a faster way to do less with fewer. The opportunity is to treat it as a reason to redesign how the most important work gets done, and to let workforce decisions follow from that redesign rather than precede it. The organizations that do this well will not simply have fewer people, they will have better shaped work, with humans concentrated where judgment, trust, and accountability actually matter.

In practical terms, that means running the sequence in order, map the workflows that create the most value and friction before touching any task list. Define the points where human judgment is genuinely required and build the controls, incentives, and career paths that protect them. Decide where release capacity will be reinvested in growth, customer experience, quality or innovation, rather than assuming it simply falls off the payroll. The discipline is to keep the hardest decision last.

A board that starts with a headcount target will reverse engineer the work to justify it and call the result a strategy. A board that starts with the work and reaches the workforce question only once the redesign is clear will make a smaller number of better decisions. Technology creates the possibility here. It is management that turns it into value.

This concludes the article. You can also read this article on my LinkedIn page where I share regular insights on AI, strategy, and emerging technologies.