Will AI create more jobs than it eliminates? According to Gartner’s forecast for 2025–2030, yes. But a positive global employment balance does not guarantee security for any individual employee, nor does it give an enterprise an operating plan. The immediate challenge for leaders is to redesign roles, retrain people and turn automation into profitability without losing institutional knowledge or professional accountability.

The forecast projects 10.6 million jobs created against 8.9 million lost. That is a net gain, but it should be treated as a forecast, not a guaranteed outcome.

  • 10.6 million new jobs projected by 2030.
  • 8.9 million jobs projected to be lost.
  • 1.7 million net additional jobs projected.

The gap between creation and elimination does not tell the whole story. New jobs will not necessarily emerge in the same places, in the same occupations or for the same people. A business can grow while some of its employees struggle to find a place in its new operating structure.

The bigger number is how much work will change

According to Gartner, more than 20% of the workforce, approximately 435.5 million workers within the forecast’s scope, will need to adapt to the technology’s effects. The scale of change in tasks and how they are performed is therefore much larger than the projected number of jobs lost.

That distinction matters for management. A role is not a single unit that can simply be retained or removed. It is a collection of activities: gathering information, reviewing documents, analyzing, communicating, making decisions and handling exceptions. AI affects each differently.

A collections employee, for example, might spend less time assembling correspondence histories and drafting outreach, and more time negotiating and managing sensitive customer situations. The job title may stay the same while the skills, output expectations and responsibilities change.

The right unit of planning for AI adoption is the task and the process, not just the headcount.

Start with a headcount reduction target, and it is easy to remove the very knowledge that helps people spot mistakes. Start by mapping the work, and you can decide what to automate, where to support employees and where human judgment remains essential.

Higher productivity does not automatically create jobs

Gartner’s John Lovelock describes a progression from hiring freezes through workforce reductions to a stage when net job growth may emerge. It is a useful explanation for the tension in today’s labor market, but not a path every organization will necessarily follow.

The outcome depends on what the business does with the capacity it frees up. A professional services firm that reduces the time needed to prepare deliverables can serve more clients, improve service or reduce working hours. Without additional demand, sales capacity or a change in pricing, higher productivity will not turn into hiring on its own.

An executive discussion about AI therefore needs to connect three things: the cost of delivering a service, demand for that service and the ability to charge for it. Time saved is a starting point, not proof of profit.

The distinction between a narrow cost-cutting approach and a capability-building approach shows up in several management choices:

  • Implementation goal: A narrow approach targets fewer positions; a capability-building approach targets better cost and output performance.
  • Unit of analysis: A narrow approach examines entire roles; a capability-building approach examines tasks and processes.
  • Employee training: A narrow approach teaches a tool; a capability-building approach develops applied professional skills.
  • Success measures: A narrow approach counts hours saved; a capability-building approach measures business outcomes and quality.
  • Accountability: A narrow approach leaves responsibility with the technology vendor; a capability-building approach assigns it to an internal process owner.

Both approaches may involve workforce efficiencies. The difference is whether the organization ends up with a stronger operating capability or simply a smaller workforce.

Keep humans in the loop without making them the bottleneck

AI can perform tasks that do not rely exclusively on fixed rules, such as interpreting an inquiry, classifying a document or developing a recommendation. Because its outputs are not deterministic, oversight is necessary. But requiring manual approval for every action can erase much of the benefit.

The goal is to move from handling every case manually to overseeing a set of processes. That requires explicit decisions about which actions can run autonomously, which require approval and which will be checked through sampling. Those decisions should reflect risk, reversibility, data quality and measured performance.

Accountability is not optional: Fewer manual approvals do not mean less control. Autonomy requires limited permissions, records of actions, quality checks and a clear escalation path to a human.

For example, an agent can be designed to gather documents and draft a response within an existing service system, while financial compensation or contractual commitments remain subject to approval. Employees can then spend their time on consequential decisions rather than copying information between systems.

Two tracks need to advance together

Employee AI literacy and enterprise agent development are not alternatives. They are complementary capabilities with different adoption challenges.

On the literacy track, employees learn to provide context to a model, specify requirements, evaluate answers and protect information. The challenge is not simply familiarity with a tool. It is changing working habits and applying professional judgment to output that looks convincing.

On the agent track, the enterprise embeds capabilities into existing processes. An agent operating behind a service system may require less behavioral change from employees than a separate chat tool, even if it is technically more complex to build. It still requires training: employees need to know when to intervene and how to report a failure.

Scaling adoption requires shared infrastructure for permissions, data connections, testing, cost monitoring and version management. IT will increasingly need to manage agent lifecycles as well: assigning roles, defining authority, evaluating performance and retiring agents when necessary.

Turn the forecast into an operating plan

There is no need to wait for certainty about the global labor market. Start with a clearly defined business process, bringing together the functional leader, operations, HR and IT.

  1. Map the work: Identify recurring tasks, decision points, risks and knowledge currently held by employees.
  2. Establish a baseline: Record handling time, quality, cost and workload before choosing a solution.
  3. Define roles and controls: Specify what the agent will do, what stays with employees and when intervention is required.
  4. Train and test in practice: Use real scenarios to evaluate performance before expanding authority.
  5. Let outcomes determine expansion: Scale only after measuring a business improvement without an unacceptable loss of quality.

This process should include an upfront decision about how to use the time released: meeting unmet demand, improving service, deepening customer relationships or adjusting staffing levels. Without that decision, hours saved on paper can disappear into other work without a clear economic effect.

Choosing the right people to lead the change matters too. Familiarity with a popular tool is no substitute for relevant education, an understanding of model limitations and experience with business processes. Value comes from combining technical knowledge, industry expertise and management capability, not from an impressive demo alone.

Gartner’s forecast is neither a reason for complacency nor a justification for premature layoffs. It is a reminder that changes in the composition of work are likely to be broader than changes in job counts. Leaders need to answer not only how many employees they will need, but what work those employees will do, what they will be accountable for and how they can create more value with AI.