A Deferred Hire Is Not a Lost Job
On 10 November 2026 Gartner set 1.3 million deferred positions beside 2.2 million jobs lost and 1.6 million gained. Counts are not interchangeable.
Christian Kosiek•Published •15 min read

An empty seat does not appear in a layoff count. No one receives a dismissal letter, and no name is added to a redundancy list. On 10 November 2026, in Barcelona, Gartner said that one of the most significant impacts on the job market in 2026 was business executives' decisions to defer the creation and hiring of 1.3 million positions, alongside 2.2 million jobs lost and 1.6 million jobs gained. John Lovelock, Distinguished VP Analyst and Chief of Research at Gartner, draws the line the release needs: a deferred job is not a lost job, because the position is not created. That is his point in the announcement. The sentence is paraphrased here, not quoted, because the release uses a contraction. These figures are not a census of vacancies employers withdrew, and they are not a list of people who were dismissed. The public announcement does not reproduce the detailed method. Gartner says clients can read more in a separate report. That report was not opened for this article.1
A deferred position is a third column
A layoff removes someone from a job that exists. A hiring freeze leaves a posted vacancy unfilled. A deferred hire is a position an employer might otherwise have created, and then does not. None of the three 2026 counts can stand in for the others. A researcher also cannot see every job that was never posted. The counterfactual, whether the position would have existed without AI, is hard to answer, and the announcement does not answer it workplace by workplace.
| Gartner's 2026 line | What the announcement supports | What it does not support |
|---|---|---|
| 1.3 million positions whose creation and hiring were deferred | An estimate of positions not created | A count of dismissal letters |
| 2.2 million jobs lost | A separate estimate, set beside the deferred count | The same 1.3 million, counted again as layoffs |
| 1.6 million jobs gained | A separate estimate of jobs gained in 2026 | A finding that the people in the lost jobs moved into the gained jobs |
By 2030, Gartner says AI will generate 10.6 million new jobs and remove 8.9 million jobs worldwide, a net increase of 1.7 million. The difference between 10.6 and 8.9 is 1.7, which is the announcement's arithmetic. A net increase does not mean every worker benefits, or that the new positions appear in the same region, industry or pay band as the ones that disappear. Helen Poitevin, Distinguished VP Analyst at Gartner, said: "The worst mistake that a business and IT executive can make is to apply lessons from one industry or role to another which can lead to poor errors in judgment and missed targets." The wording is hers, including "poor errors." She also said that organizations most advanced in AI execution view AI not as a replacement for labor, but as a catalyst to reinvent operations, and that they ensure AI's economic value goes beyond replacing labor. Those remarks are not a method for reading one employer's headcount.1
By 2030, Gartner projects that between 0.4% and 1.8% of workers will experience job loss directly attributable to AI. Through 2030, just over 20% of workers (435.5 million employees worldwide) will have to adapt to the new shape of work in their jobs, changing what they do, how they do it, and where their responsibilities stop and end. "Just over 20%" is the release's wording. It is not printed as "more than 20%." The 435.5 million is the figure in parentheses, not an approximation added here. The 0.4% to 1.8% range is a share of workers who experience job loss. It is not the 1.3 million, the 2.2 million or the 10.6 million.1
Three industries, and percentages that do not travel
Gartner outlines three industry types. The announcement does not present those lines as slices that add up to the global 10.6 million created and 8.9 million removed. Adding the published millions does not produce those totals: 6.4 + 1.9 + 0.4 is 8.7, and 6.0 + 0.9 + 0.4 is 7.3. This article does not force a sum, and it does not invent a missing industry to close the gap.
Asset-intensive industries, which deal in physical products or deliver services inside capital-intensive physical infrastructures, are described as the largest share of jobs globally. By 2030 they are projected to see 6.4 million jobs created and 6 million lost. AI-driven job losses are projected to affect 0.4% of the global workforce in this industry through 2030.1
The public sector is a second line. The release says AI investments there are unlikely to be sufficient to mitigate future severe skills shortages driven by other factors. By 2030 it projects 0.4 million job gains and 0.9 million job losses, and AI-driven job losses totalling 0.4% of the global public sector workforce. The shortages "driven by other factors" are not the same sentence as the 0.9 million.1
Technology is the third line. Nearly 70% of that workforce is described as having high AI exposure. By 2030 the sector is projected to see 1.9 million jobs created and 0.4 million lost. The AI job-loss impact is projected to represent 1.8% of the global technology workforce through 2030. That 1.8% is the technology workforce's own projected loss share. It sits at the top of the global 0.4% to 1.8% range, and it is not a license to put 1.8% on a hospital, a town hall or a workshop. The asset-intensive 0.4% and the public-sector 0.4% are stated for those workforces. Poitevin's warning is the reason to keep them apart.1
By 2030, a total of 5 million people will have experienced delayed hiring because business and IT executives postponed the creation of new positions. The release says "a total of 5 million," and it names business and IT executives. It does not say "a cumulative five million" as a different statistic, and it does not limit the sentence to business executives alone. The 5 million is the delayed-hiring total through 2030. It is not the 1.3 million for 2026 counted a second time, and it is not the 8.9 million jobs removed.1
The junior rung is a mechanism, not a finding
Consider a hypothetical six-person accounting practice that expected to recruit a junior assistant to prepare reconciliations and draft client correspondence. After AI-supported workflows, the owner postpones the hire for a year. The accountants already on staff review the outputs. No one is fired. The team may get more done. A graduate loses a possible first rung, and the firm loses a possible future manager. The example is hypothetical. It is not a firm in the Gartner announcement, and it is not evidence that this has already happened at a named practice. Software cost, client demand, interest rates, wages and confidence also shape hiring. An unfilled role is not, by itself, an AI casualty.
If the pattern were widespread, the cost would be cumulative. Junior roles often supply the supervised repetitions through which people learn to see a mistake, understand a customer and exercise judgment. If routine tasks move to a tool and nothing replaces that practice, a team can later lack people able to supervise the tool. That paragraph is an editorial argument. Gartner did not publish it as a result. A related question, after someone is in the job, is who teaches judgment after the first draft. Entry-level ads that already ask for senior judgment are a different article, with different sources: the entry-level job that asks for senior judgment.
An IMF working paper cuts against reading the hypothetical as a measured junior-first effect. In the subset of postings that contain a minimum experience requirement, the authors report largely no statistically significant difference in labor demand for junior roles compared with senior roles. That sentence is about postings with an experience field. It is not a finding about every graduate, and it is not the accounting practice.2
Nearly 10 percent is a relative gap in postings
The same paper is not a Gartner forecast, and it does not estimate jobs that were never created. IMF Working Paper 2026/220, by Sakai Ando, Divya Kirti, Willa Lin and Assyifa Szami Ilman, was published on 18 October 2026. The summary says the authors draw on over 700 million online job postings covering 110 economies. The data section describes a Lightcast panel of approximately 700 million postings across those 110 economies, from November 2021 through April 2026. "Over 700 million" and "approximately 700 million" are the paper's two wordings for that panel. This article does not pick a third number. Lightcast compiles vacancies scraped from major recruitment platforms. Occupations are 342 four-digit ISCO-08 codes. Exposure uses the complementarity-adjusted AI occupational exposure measure from Pizzinelli and co-authors (2023).2
The design is difference-in-differences, comparing more exposed occupations with less exposed ones, before and after the launch of ChatGPT in November 2022, inside the same economy, industry and month, with a control for how amenable an occupation is to remote work. Postings declined by nearly 10 percent in more AI-exposed occupations relative to less exposed occupations. That is not a 10 percent fall in total vacancies, and it is not a 10 percent fall in employment. The dynamic estimates say the differential was modest at first and became more negative through 2023 and 2024. By the later part of the sample, the estimated decline is nearly 10 percent. That "nearly 10 percent" is the later point on the event study. It is not a pooled average.2
The pooled table is a different cut. Across specifications, the impact on highly exposed occupations relative to less exposed ones ranges from a 5 percent to an 8 percent decrease. In the preferred specification, the pooled effect for the combined period after AI is about a 6 percent decrease. Do not replace the nearly 10 percent with 6 percent, or the reverse.2
A third cut uses capability milestones from METR, a nonprofit that benchmarks frontier models, rather than calendar months. The estimated impact becomes more negative as systems move toward the milestone for one-hour tasks, and then largely stabilizes. The paper says a line that pools the period after that one-hour milestone indicates an effect of approximately 8 percent. That 8 percent is not the table's range of 5 to 8 percent, and it is not the event study's nearly 10 percent. The paper's own surprise is that the relative decline did not keep deepening as models became more capable. When the summary says the impact has stabilized, this is the pattern it is summarizing.2
The estimates describe the flow of new postings. The paper says the impact on the stock of employment will accumulate slowly and will depend on separations, turnover and realized hiring. A posting can also be duplicated, left unfilled or written to test the market. That last limit is a general feature of vacancy files. The limit the paper states in its own words is the flow, not a headcount of people already in work.2
Firms with stronger demand for AI skills show faster growth in postings, even in more exposed occupations. The authors build that demand from 51 emerging AI skills: general familiarity, such as ChatGPT or Copilot; expertise, such as prompt engineering or AI agents; and the ability to connect AI with internal data. Demand for those skills has increased rapidly since 2023 and is concentrated in occupations with higher exposure. Faster posting growth at firms that ask for those skills is not the Gartner figure of 1.6 million jobs gained, and it is not proof that every exposed occupation is hiring.2
Two other articles on this site use other evidence and should stay separate. Smaller startup teams are not proof of an AI cut is a LinkedIn comparison of founder cohorts. AI creates jobs, and that is not a transfer for everyone is about new work that does not automatically absorb the people who lose a role. When a job does end, the next search can take longer, which is a different object again: the next search after an AI job loss. None of those pieces is the 1.3 million, and none of them is the relative 10 percent.
What to write down before the vacancy disappears
For a small business or a nonprofit, the useful question is not whether AI will raise employment in the world. It is which work is changing in this team, and what happens to the way people learn it. The list below is an editorial checklist. Gartner and the IMF did not publish it as a method.
- Before postponing a vacancy, write the original case: the tasks, the hours, the service demand and the training purpose of the role.
- Then record whether the tool removes that workload, moves checking onto senior staff, or creates new work such as checking data quality.
- Look at error rates, time spent verifying outputs, customer outcomes, overtime and whether anyone is still learning the work.
- If drafting is faster and correction is slower, the labor saving may not be there. Say so in the note.
- If the junior vacancy goes, name the replacement for that practice: a rotation, a supervised case review, shadowing, or a narrower paid apprenticeship.
- Separate a time-limited trial, with a date and review criteria, from a line that quietly leaves the budget.
A six-month trial that can be reversed is a different decision from deleting the role. A nonprofit can apply the same note to volunteer coordination, grant writing and case administration, and still has to protect confidentiality and name who is accountable. Those scenes are illustrations. They are not in the Gartner release or the IMF paper. The woodworking shop in the image is not the accounting practice.
The net figure cannot name who adjusts
Public debate often sets jobs created against jobs destroyed. Gartner's November announcement is a reason to keep a third column: positions whose creation was postponed. That column does not establish that AI has already removed millions of opportunities a statistician can count. It shows why a net figure of 1.7 million cannot say who bears the adjustment. The IMF result is about relative demand for postings, not about that third column. Using one study to prove the other would repeat the mistake Poitevin names: applying a lesson from one object to another.
For a public reader, youth entry, apprenticeship openings and moves between occupations are worth watching beside layoff counts. This article does not supply those series. For an employer, the practical test is whether the productivity gain is real before a junior role disappears. For a worker, domain knowledge that can be shown, the ability to check an output, and the ability to adapt remain useful. Those last sentences are editorial judgments. They are not Gartner scores.
How hilighter can help
hilighter does not estimate how many jobs were never opened. It does not reproduce a Gartner or IMF figure, and it does not decide whether a vacancy should be postponed.
What a team can do is write down the work before the seat disappears. Map skills by role, including the checks a person still has to make. Compare people with those requirements in skills and talent insights. Record a correction or a finished exercise in skills evidence and recognition. Set personalized growth paths so learning is not only the vacancy that was not opened. A manager's review of that evidence sits in manager assessments. The AI job impact view is indicative. It is not a Gartner forecast, and it is not the IMF posting gap. The Free plan covers up to three job roles. The number of employees is not capped.
Sources
- Gartner, Gartner Forecasts AI Will Drive a Net Increase of 1.7 Million Jobs by 2030, Barcelona, 10 November 2026. Figures used from the announcement, not from a reading of the chart image: in 2026, deferral of the creation and hiring of 1.3 million positions, alongside 2.2 million jobs lost and 1.6 million jobs gained. By 2030, 10.6 million new jobs and 8.9 million removed, a net increase of 1.7 million. A total of 5 million people will have experienced delayed hiring because business and IT executives postponed new positions. Between 0.4% and 1.8% of workers experience job loss directly attributable to AI by 2030. Through 2030, just over 20% of workers (435.5 million employees) adapt to the new shape of work. Asset-intensive: 6.4 million created and 6 million lost, with AI-driven losses affecting 0.4% of the global workforce in this industry. Public sector: 0.4 million gains and 0.9 million losses, with AI-driven losses at 0.4% of the global public sector workforce; AI investment is described as unlikely to offset severe skills shortages driven by other factors. Technology: nearly 70% of the workforce with high AI exposure; 1.9 million created and 0.4 million lost; AI job-loss impact 1.8% of the global technology workforce. Lovelock's distinction between deferred and lost jobs is paraphrased because the release uses a contraction. Poitevin is quoted on applying lessons from one industry or role to another. The client report named in the release was not opened. The three industry million-counts are not added up to the global totals. The chart titled "AI Cumulative Job Impact Forecast, Worldwide, 2025-2030" is mentioned on the page; values were not read off the image.
- Sakai Ando, Divya Kirti, Willa Lin and Assyifa Szami Ilman, How is Labor Demand Changing as AI Improves? New Global Evidence, IMF Working Paper 2026/220, published 18 October 2026, DOI 10.5089/9798229061551.001. The full text on the IMF eLibrary was opened. Summary: over 700 million online job postings, 110 economies; demand declined by nearly 10 percent in more AI-exposed occupations, and the impact has stabilized as AI became more capable; firms with stronger demand for AI skills show faster growth in postings, even in more exposed occupations. Data section: Lightcast, approximately 700 million postings, 110 economies, November 2021 through April 2026. Introduction: difference-in-differences, nearly 10 percent relative to less exposed occupations after ChatGPT in November 2022. Event study: by the later part of the sample, nearly 10 percent. Table 1: 5 to 8 percent across specifications; preferred specification about 6 percent in the pooled post-AI period. Figure 1(b), METR capability milestones: approximately 8 percent after the one-hour milestone, with the relative decline then largely stable. Those three percentages are not interchangeable. Appendix on seniority: largely no statistically significant difference between junior and senior labor demand in postings that state a minimum experience requirement. Fifty-one emerging AI skills. The estimates are a flow of postings, not the stock of employment. This paper does not estimate Gartner's deferred positions.
Sources last reviewed: 19 November 2026. The 1.3 million deferred positions, the 2.2 million jobs lost and the 1.6 million jobs gained are three 2026 lines, not one. The 2030 net of 1.7 million is 10.6 minus 8.9. The total of 5 million delayed hires through 2030 is not the 1.3 million. "Just over 20%" and 435.5 million employees are the release's pair for task change, not for job loss. The 0.4% to 1.8% range is a share of workers, and the industry 0.4% and 1.8% lines stay on the workforces the release names. The IMF nearly 10 percent is relative to less exposed occupations and is not the preferred specification's about 6 percent, nor the approximately 8 percent after the one-hour capability milestone. The accounting practice, the checklist and the woodworking image are illustrations. The Gartner client report was not opened. Neither an AI-detector score nor the absence of Google penalties can be guaranteed.