AI Productivity Should Not Start With Layoffs

Every productivity wave eventually becomes a people question.

At first, the conversation sounds technical. A new tool arrives. A workflow gets faster. A team discovers that AI can summarize documents, draft code, answer support questions, generate SQL, review contracts, or automate parts of a reporting process. The first demos are exciting because they show the same work happening with less effort.

Then the financial question arrives: if the work takes less time, do we still need the same number of people?

That question is not new. Companies have always had to decide what to do when technology changes the amount of labor required. What makes the current moment harder is the speed and visibility of AI. Leaders can see individual tasks getting faster before they understand which capabilities are actually being preserved, moved, weakened, or created. The temptation is to convert every apparent efficiency gain into headcount reduction.

I think that is often the wrong starting point.

This does not mean layoffs are never necessary. Organizations sometimes face real financial pressure. Some work disappears. Some roles become structurally less useful. A leadership team cannot pretend cost does not matter. But starting with layoffs can hide the deeper question: what capability does the organization need to keep, and what is the best way to fund it?

AI can reduce costs. It can also create new operating risks, quality problems, security questions, support burdens, and evaluation work. If leaders cut people before they understand that full system, they may save salary expense while quietly destroying the judgment that makes the AI useful.

Productivity Is Not the Same as Capacity

A common mistake is to treat productivity as if it automatically creates removable capacity.

If an analyst uses AI to prepare a first draft faster, that does not prove the analyst is now unnecessary. It may mean the analyst can spend more time checking assumptions, talking to stakeholders, validating data, improving the model, or translating the work into decisions. If a developer uses an AI coding assistant to generate boilerplate, that does not remove the need for architecture, testing, code review, security thinking, and maintenance. If a support team uses an AI assistant to retrieve policy answers, that does not eliminate escalation handling, customer empathy, process improvement, or accountability for bad answers.

The work changes shape.

Sometimes the change really does reduce the need for people doing a narrow task. But sometimes AI only removes the visible part of the work and pushes more responsibility into review, orchestration, and exception handling. This is especially true in data, software, and AI systems because the cost of a bad answer is rarely limited to the answer itself. A wrong SQL query can mislead a business review. A generated code change can introduce a security flaw. A hallucinated policy answer can damage trust. A poorly governed agent can take an action the business never meant to authorize.

That is why the useful question is not, “How many hours did AI save?”

The better question is, “Which human responsibilities remain, which ones become more important, and which new ones did the technology create?”

Microsoft’s 2026 Work Trend Index makes this point in a practical way. It describes AI and agents taking on more execution while placing a higher premium on judgment, quality control, intent, and work design. That is the important leadership lesson. If AI does more of the execution, humans do not disappear from the system. Their role moves toward deciding what should happen, evaluating whether it happened well, and taking responsibility for the outcome.

That kind of work is easy to underestimate because it is less visible than production volume.

The Layoff Decision Can Distort the AI Decision

Once AI is framed primarily as a headcount reduction tool, teams start optimizing for the wrong evidence.

They look for demos that prove work can be done with fewer people. They count tasks that can be automated. They talk about “efficiency” before they have defined quality. They reward teams for reducing visible labor even when the hidden work grows somewhere else. In that environment, people also become less honest. Employees avoid sharing what the tool cannot do. Managers overstate savings to satisfy the budget conversation. Technical teams rush deployment because the business case depends on cost reduction.

The organization may get the worst of both worlds: weaker trust and unreliable automation.

This matters because most companies are still early in the AI operating model. McKinsey’s 2025 State of AI survey found that 88 percent of respondents reported regular AI use in at least one business function, but only about one-third said their companies had begun scaling AI programs across the enterprise. The same survey reported that many organizations are experimenting with agents, but broad enterprise impact is still uneven.

That is not a mature replacement story. It is an adoption story.

When leaders turn an adoption story into a layoff story too early, they risk cutting the very people needed to learn what the system should become. The people closest to the workflow often know where the data is messy, where customers ask ambiguous questions, where exceptions appear, and where the current process survives only because someone quietly checks the result.

AI does not automatically expose that knowledge. In many cases, it depends on it.

The leadership mistake is not using AI to reduce cost. The mistake is deciding that cost reduction must come before understanding the work.

Cutting People Can Also Cut Organizational Memory

In technical work, the value of a person is rarely limited to the tasks listed in their job description.

Someone knows why a dashboard uses a strange filter. Someone remembers why an API endpoint behaves differently for one customer segment. Someone knows which database table looks reliable but is not. Someone knows which vendor promise failed during the last implementation. Someone knows which stakeholder will reject an automated workflow unless the review step is designed carefully.

This knowledge does not always appear in project plans. It may not be documented well. It may not look like productivity in a spreadsheet. But it is part of the organization’s ability to move without breaking things.

AI can make this more important, not less.

Modern AI systems often depend on context: policies, historical decisions, data definitions, business rules, edge cases, evaluation examples, prompt versions, retrieval quality, human approval paths, and escalation procedures. If the people who understand those details are removed too quickly, the organization may still have the tool but lose the ability to operate it intelligently.

This is one reason I am skeptical of simple “AI replaces X percent of work” thinking. A task may be automatable in isolation and still unsafe to remove from the surrounding workflow. The output might need review. The data might need cleanup. The model might fail on rare cases. The system might require monitoring for drift, latency, cost, and security. The business might need an accountable person who can explain why a decision was made.

Those are not small details. They are the difference between a demo and an operating capability.

This is also why I like to connect workforce planning with practical AI skills. In How to build practical AI skills for today’s tech job market, I wrote that knowing AI vocabulary is not the same as building reliable systems. The same principle applies to leadership. Saying “AI will make us more efficient” is not the same as designing a workflow where AI, people, data, controls, and accountability fit together.

When Cost Pressure Is Real, Do the Capability Audit First

There are moments when an organization truly has to reduce labor cost. Revenue falls. Funding changes. A product line shrinks. A public agency faces a budget constraint. A company has overhired. A workflow has become less important. Leaders then have to choose among painful options.

The AI era does not remove that reality. It does change what a responsible decision should examine.

Before treating layoffs as the default answer, I would want leaders to do a capability audit. Not a sentimental exercise. A practical one.

Ask these questions:

• Which work is genuinely no longer needed?
• Which work is still needed but can be redesigned with AI?
• Which work looks automatable but still requires human judgment, review, or relationship management?
• Which people hold critical context about data, customers, systems, or risk?
• Which AI workflows need evaluation, monitoring, security, and human approval?
• Which roles should be retrained because the work is changing rather than disappearing?
• Which teams are already understaffed for the governance burden AI creates?

This audit should include technical and nontechnical work. AI adoption often creates dependencies across product, engineering, data, security, legal, operations, customer support, finance, and HR. A narrow view of productivity can miss those connections.

The U.S. labor market data also reminds us to be precise. The Bureau of Labor Statistics JOLTS release for May 2026 reported 1.708 million layoffs and discharges in the month, with large differences by industry. That kind of data is useful because it keeps the discussion grounded. Workforce reduction is not an abstract management lever. It is a recurring economic event with industry-specific causes, timing, and consequences.

For a technical leader, the point is not to make layoffs sound better or worse in the abstract. The point is to understand what is actually happening in the work.

Salary Cuts Are Not a Simple Moral Alternative

When leaders want to avoid layoffs, one tempting answer is to reduce pay across the board. It feels fairer because more people keep their jobs. In some situations, it may be the least harmful temporary option, especially if the financial shock is short, the team trusts leadership, and the alternative would damage a community or destroy a hard-to-rebuild capability.

But it is not automatically better.

Across-the-board pay cuts can create their own damage. The people with the strongest external options may leave first. Trust can fall. Employees may conclude that leadership wants loyalty without a credible plan. People may reduce effort, delay hard work, or start searching quietly. The organization may keep headcount on paper while losing energy, expertise, and commitment.

In AI-era work, this risk can be sharper because strong people often have options. Someone who understands data engineering, AI evaluation, security, product judgment, and workflow redesign is valuable in a market where many companies are still trying to turn pilots into production systems. If that person leaves, the organization does not merely lose hours. It loses learning capacity.

So the decision is not simply layoffs versus salary cuts. The better decision framework is:

• Is the pressure temporary or structural?
• Does the organization need to preserve specialized knowledge?
• Is the team likely to trust a shared sacrifice?
• Are leaders also participating in the sacrifice?
• Is there a concrete plan for restoring pay, changing work, or rebuilding growth?
• Would a smaller focused team outperform a larger demoralized one?

There is no universal answer. But there is a universal warning: any cost decision that ignores trust will be more expensive than it looks.

AI Savings Should Be Reinvested Before They Are Harvested

One of the healthiest uses of AI productivity is reinvestment.

If a team saves time with AI, the first instinct should not always be to remove the time from the organization. Sometimes the better move is to spend that time on work the team previously neglected: better documentation, more testing, cleaner data, stronger evaluation sets, improved onboarding, customer research, security review, workflow redesign, or technical debt reduction.

These are not glamorous activities, but they are often where durable productivity comes from.

For example, suppose a data team uses AI to speed up first drafts of analysis. A weak organization treats the saved time as proof that fewer analysts are needed. A stronger organization asks whether the team can now improve metric definitions, build better data quality checks, document assumptions, reduce recurring manual requests, or create reusable decision dashboards.

Suppose an engineering team uses AI coding tools. A weak organization assumes code generation means fewer engineers. A stronger organization asks whether engineers can now write more tests, improve architecture, reduce incident-prone code, strengthen security reviews, and make systems easier to maintain.

Suppose a customer support team uses an AI assistant. A weak organization removes agents as soon as deflection improves. A stronger organization studies which questions remain difficult, which answers need approval, which policy gaps customers keep exposing, and how the assistant should be evaluated before expansion.

The point is not to protect every old role forever. The point is to avoid harvesting savings before the organization has built the capability that makes those savings repeatable.

AI productivity that depends on fragile prompts, undocumented workflows, unreviewed outputs, and exhausted employees is not real productivity. It is borrowed time.

Leaders Need a New Workforce Metric

Traditional workforce planning often counts people, cost, utilization, and output. Those metrics still matter, but AI adds another layer: human leverage.

A person may now be able to supervise more work than before, coordinate agentic workflows, review higher volumes of output, or translate ambiguous business problems into structured AI-supported processes. But that leverage depends on skill, judgment, tooling, and organizational support.

Anthropic’s Economic Index research found that Claude usage was concentrated in software development and writing tasks, with both augmentation and automation patterns present. Microsoft also reports strong growth in agent use inside Microsoft 365. These signals point in the same direction: AI is not affecting all work evenly. It changes task composition, and the impact depends on how people and systems are arranged.

So leaders should stop asking only, “How many people do we need?”

They should also ask:

• How much reliable work can a skilled person now direct with AI?
• Where does AI increase output but also increase review responsibility?
• Which roles should shift from production to supervision, evaluation, and workflow ownership?
• Which teams need fewer task executors but more system designers?
• Which skills are becoming bottlenecks because AI makes poor judgment more scalable?

This is a better conversation than headcount math alone. It recognizes that some roles will shrink, some will grow, and many will change. It also makes reskilling concrete. Training should not be a vague promise offered after cuts are announced. It should be tied to real work: evaluation, RAG quality, workflow mapping, prompt and model regression testing, tool governance, cost monitoring, data stewardship, and human approval design.

If a company cannot name the new responsibilities, it probably is not ready to claim the old responsibilities have disappeared.

Be Honest About the Tradeoff

There is a sentence leaders should be careful with: “AI lets us do more with less.”

Sometimes it is true. Sometimes it hides a tradeoff nobody wants to name.

Doing more with less may mean fewer defects because the workflow is better designed. It may also mean fewer people reviewing more machine-generated output under more pressure. It may mean faster customer response. It may also mean more unsupported answers reaching customers. It may mean lower cost. It may also mean the organization has less ability to recover when the system fails.

The honest version is more specific:

“AI lets us automate this repeatable step, but we still need human review for high-risk cases.”

“AI lets analysts draft faster, so we are reallocating time to data quality and stakeholder work.”

“AI lets support handle simple questions at scale, but escalation and policy feedback are now more important.”

“AI reduces manual reporting work, so we are retraining part of the team toward metric governance and automation ownership.”

“AI has made this old process structurally smaller, and we need to redesign roles rather than pretend nothing changed.”

Specificity creates trust. It also makes the decision testable. If leaders say AI will reduce cost without reducing quality, what metrics will prove it? If they say a team can support the same workload with fewer people, how will they monitor burnout, defects, latency, customer satisfaction, security exceptions, and rework? If they say reskilling is possible, what work will people move into, and when?

Vague productivity claims are easy. Responsible workforce decisions require proof.

The Better Goal Is a More Capable Organization

The durable lesson is not that layoffs are always wrong or that pay cuts are always noble. The lesson is that labor cost decisions are also capability decisions.

AI makes this more important because it can create a false sense of completeness. A model can produce a polished answer before the organization knows whether the answer is correct. An agent can complete a workflow before the organization knows whether the workflow should have been completed. A dashboard can show saved time before anyone measures the cost of errors, rework, supervision, or lost trust.

Leaders need to slow down enough to see the full system.

If the work has truly disappeared, be honest. If the work has changed, redesign roles. If the pressure is temporary, consider temporary measures with clear terms. If the organization needs new AI skills, invest before cutting. If AI creates savings, reinvest some of them into evaluation, documentation, data quality, security, and workflow ownership before removing capacity. If layoffs are unavoidable, do not pretend they are proof of AI maturity.

The goal should not be the smallest possible team. It should be the most capable sustainable organization for the work that matters now.

AI will keep changing how technical and knowledge work is done. Some teams will need fewer people in old roles. Some will need more people in new roles. Many will need people who can combine domain knowledge, technical judgment, communication, and responsibility for AI-supported outcomes.

That is the practical leadership challenge. Not to protect every existing job from change, and not to treat every productivity gain as permission to cut. The better challenge is to understand the work deeply enough to know what should be automated, what should be redesigned, what should be preserved, and what human judgment is now too important to lose.