AI Training Reaches Bosses, Not Beginners

A login is not a lesson. In a small organisation the AI policy may be one sentence: everyone should start using the tools. The director attends a strategy session. The newest employee is handed an account and told to experiment. Both can say they have access. Only one has been shown how to judge the result.

That scene is hypothetical. The imbalance has a survey behind it. It is not proof that AI training causes career inequality. It is a warning that an employer can speed up adoption and still leave the people who most need supervised practice to improvise.

The training rates do not form a neat ladder

On 26 October 2026 ZipRecruiter published Better Work, More to Learn: The 2026 AI Worker Report, written by Nicole Bachaud. The page says the fieldwork was a nationally representative online survey of a PureSpectrum panel of more than 1,500 verified US workers, from 8 to 21 September 2026. It does not publish the weighting, the response pattern or a margin of error for each subgroup. The figures below are what people reported. They are not an audit of employer training records.¹

Senior leaders, defined on the page as directors, vice presidents and the C-suite, report any AI training at their company at 83.5%. Individual contributors are at 46.4%, and entry-level workers at 45.3%. The page calls the two lower groups "far less likely to receive AI-specific training". The adjective is not identical: "any AI training" for senior leaders, "AI-specific training" for the others. The report presents them as one comparison. This article does not upgrade that into a finding that the courses were the same length, the same quality, or assessed.¹

Self-teaching is a different sentence, and the clause matters. 37.2% learn entirely on their own because their employer provides no training. The page then says the self-teaching burden falls on entry-level workers (54.7%) and on women (45.4%, against 29.9% of men). The "because the employer provides no training" clause is attached in the text to the 37.2% figure. It is not repeated as a separate definition of the 54.7%. These gaps identify who is practising without support. They do not explain why a particular person was promoted or left behind.¹

Use is not a perfect ladder either. 26.6% of entry-level workers say they do not use AI at all. Among individual contributors the share is higher, 30.5%. Only 5% of senior leaders say they never use AI at work. The people in the middle of the chart are not simply a step between the newest hire and the board.¹

Encouragement is not a training programme

Employers can mistake permission for preparation. In the same worker survey, 62.8% say they receive any form of AI training from their employer, and 39.3% say their employer actively encourages AI use. The key findings say only 28.9% receive enough training. The body of the same page says only 28.9% receive extensive training. This article does not treat "enough" and "extensive" as two statistics. It also does not subtract 28.9 from 62.8, or 39.3 from 62.8, to invent a single deficit.¹

Expectations are rising on both sides of the market, in two surveys that must stay separate. 49.9% of the worker respondents report higher productivity expectations linked to AI. ZipRecruiter's employer report, More Jobs, Higher Bar, is a different sample. The page is dated 29 July 2026. Fieldwork was 11 to 18 June 2026, with more than 1,000 verified US talent-acquisition professionals and hiring managers. 57% of those employers say they have raised baseline productivity expectations because of AI. The two percentages do not measure the same people, and they do not show how much finished work actually increased.¹²

The employer page also describes how thin the formal offer is. 22% provide mandatory AI training for all employees. Another 23% offer it only to specific departments. 34% rely on optional resources, and 17% provide no training. The page also says the remaining 55% are the optional-or-none group. Those published shares do not add to a clean 100 (34 plus 17 is 51). This article quotes them and does not force a total.²

Two further employer findings belong next to the worker gap, still as reported experience. 38% say they have already moved basic data entry and processing off entry-level workers and onto AI. 31% say AI has raised experience requirements for entry-level roles. If the starter tasks move and the training does not, the first job asks for judgment the old first job used to build.²

A manager who asks a team to produce more with AI, without naming the tasks, the quality standard or the support, is making a quiet bargain. Employees supply the learning time. The organisation expects the productivity. That bargain may be tolerable for someone who controls their calendar. It is a weak design for a junior colleague, a shift worker, or anyone who cannot safely paste client data into an unapproved tool.

Senior leaders may feel the benefit first

The same worker survey finds that 76.5% of senior leaders say AI improved their work satisfaction, against 39.8% of entry-level workers. That sits beside the training and usage gaps. It does not demonstrate that the training caused the satisfaction, or that entry-level staff would report the same benefit from an identical course.¹

Job content is a plausible part of the difference, and it is an editorial reading, not a result the survey tested. A senior manager may use AI to condense a briefing whose subject they already understand. A junior employee may be asked for the briefing itself, before they have learned which details matter, which sources hold, or which mistakes are expensive. The tool can make both faster. It does not give both the same capacity to judge the draft.

The most common reported uses are exactly those starter tasks: research and information gathering (52.1%), summarising documents or transcripts (41.6%), and drafting or editing text (40.9%). They show up in sales, administration, fundraising, customer support and project coordination. Each one still needs a judgment about relevance, accuracy, confidentiality and tone.¹

For how tasks move before job titles do, see AI changes tasks inside jobs before it changes the titles. For the gap between stretching a task and having mastered the work behind it, see why going beyond an apprentice's training is not the same as mastery. When the new constraint is checking what a system produced, see why human review becomes the bottleneck once agents are fast.

The World Economic Forum's publication page of 22 June 2026, for a white paper written with PwC, makes a related argument without testing a training method. It says that globally, more than one in three young workers are employed in occupations with medium to high exposure to AI-driven task change. The framework it introduces has four dimensions: job access, job design, talent pipelines and education system alignment. The useful shift is from "Can AI do this starter task?" to "What capability did the starter task used to build?" The full paper was not re-read page by page for further statistics.³

The skill is wider than a prompt

On 5 June 2026 the OECD published the policy brief AI and skills: What we know so far. Less than 1% of workers need advanced AI skills such as programming or model development, a figure the brief takes from Green and Lamby (2023). Most people need digital skills and the ability to use, analyse and interpret data, plus managerial and human skills such as problem solving, creativity and innovation.⁴

The same brief says skills shortages hold adoption back, especially in smaller firms. Around 40% of employers in manufacturing and in finance and insurance who have not yet adopted AI say skills are the main reason, with sample sizes of 1,400 and 653, from Lane, Williams and Broecke (2023). More than half of SMEs that are not yet using generative AI say the same, citing an OECD survey published in 2025. The brief also says that more than half of workers using AI report receiving employer-funded training, again from the 2023 employer and worker surveys, and that workers who received training were more likely to report better performance and working conditions. That is not a seniority split, and it is not a 2026 experiment. The brief says its most recent data date from late 2024. An older finding that many AI users got some employer-funded training does not answer whether a new hire in 2026 got the same support as a director.⁴

Geography is a separate caution. The European Commission's Joint Research Centre lists AI and Skills (JRC148616) as published on 15 October 2026, by Alice Bertoletti, Alessia De Santo, Yannis Galanakis and Wouter Vergote. The publications-record abstract says that in 2024-2025, 30% of EU workers reported using AI at work, and that by 2025 20% of enterprises had adopted AI technologies. Worker use ranges from close to 40% among white-collar workers to 6% in elementary occupations. The official PDF did not open on 1 November 2026: the repository rejected the request. The abstract was read on the publications record and the DOI page.⁵

A Directorate-General for Employment news page of 23 October 2026 summarises the same brief as around 30% of workers and repeats the occupation contrast. It also says about 30% of EU enterprises with ten or more employees have adopted AI. That enterprise figure does not match the abstract's 20%. This article uses the abstract's 20% and does not merge the two enterprise numbers into one claim.⁶

ZipRecruiter's US worker survey says 82.5% report some AI use at work. Neither 82.5% nor 30% is a universal adoption rate. The surveys cover different places, periods, questionnaires and samples. Their difference cannot, by itself, prove that adoption jumped by a particular amount.¹⁵

The gender split in self-teaching is also easy to over-read. Women in the ZipRecruiter worker survey are more likely than men to say they teach themselves (45.4% against 29.9%). That is not the same finding as the employment gap in AI-exposed US occupations. For that separate analysis, see why the AI jobs gender gap is widest without a degree.

A small team can test one workflow

Consider a hypothetical nonprofit with twelve people. The grants coordinator wants help summarising eligibility rules. The service team handles sensitive personal information. A one-hour tour of product features is easy to schedule. It does not answer which documents may be uploaded, how a claim should be checked, when the tool must not be used, or who signs the final decision.

A better start is one low-risk workflow. The grants coordinator could extract requirements from a public funding call, compare the output with the original, and note the kinds of errors. A supervisor reviews two examples and says what a good check looks like. The service team practises on invented case notes, with an explicit ban on putting real client information into an unapproved system.

The point is not the number of prompts. It is visible evidence of skill: can the person spot a misleading summary, explain why an answer is unreliable, and recover when the tool fails? A short supervised exercise can teach more than a library of videos nobody has time to apply. This is a proposed way of working. The ZipRecruiter survey did not test it.

Three design choices follow from the evidence above. They are operating suggestions, not findings.

Protect a little paid practice time. Do not assume people will learn at night. Pair a junior colleague with someone who knows the job, not only someone who knows the newest interface. Keep a sample of the person's own reasoning next to the assisted result, so speed can be seen without losing the question of whether they understood the work.

Ask managers to show the standard. Each manager provides one AI-assisted task that worked and one error they caught, including the source, the decision rule and when to stop. A manager who cannot explain the check is not ready to treat frequent use as competence.

Count access, not only attendance. Record which roles received practice, coaching and feedback, and which were left to self-teach. Look at shifts, language, disability access and temporary contracts. The International Labour Organization's publication page for Lifelong learning and skills for the future, dated 5 May 2026, says quality training that leads to recognised skills is especially out of reach for low-qualified and informal workers and for people in smaller enterprises. The full report was not re-read page by page. The page is enough to treat access as a design question, not as a test of personal motivation.⁷

What this survey cannot tell a leader

The worker findings are self-reported and cross-sectional. There was no random assignment of training. Seniority may also stand in for age, industry, occupation and freedom to choose tasks. The publisher calls the panel nationally representative. The public article does not give the detail needed to judge the weights.

25.1% say they plan to leave their job because of how their employer views or uses AI. A further 30.9% report elevated concern. The page says that concern can be too much use, too little use, missing guardrails, or ethics and privacy. Intentions are not observed resignations. It would be misleading to forecast that a quarter of the workforce will quit because of AI.¹

44.6% say AI is making their technical skills outdated faster. Among people who use AI daily, that share is 67.3%. That is a perception of skill turnover, not a measurement that those skills lost value. It is a reason to ask better questions, not a reason to declare that a profession has an expiry date.¹

One further satisfaction figure is easy to drop, and it should not be. 52.6% say AI made work more enjoyable, against 5.5% who say it made work less enjoyable. The positive report is real. A tool becomes a career advantage only when people have the knowledge and the authority to use it with care.¹

Only 34.3% believe they can master AI skills without help. 40.2% say they could if given proper training. The page presents that as room for employers to formalise learning. It is still a belief about future mastery, not a completion rate.¹

AI literacy is a duty to support learning, not a guaranteed score

For organisations in the European Union, the legal line is more specific than a general warning about "AI at work". This article is not legal advice.

Article 4 of the EU AI Act, in the text the Commission service desk shows as amended, says providers and deployers shall take measures to support the AI literacy of their staff and other people who operate AI systems on their behalf. They must take account of technical knowledge, experience, education and training, the context of use, and the people or groups the systems are used on. The same text says this obligation does not require anyone to guarantee a specific level of AI literacy for any individual. The Commission and the Member States shall support those efforts, in particular for SMEs. The service desk says the wording reflects the EUR-Lex consolidated text as at 27 July 2026, including Digital Omnibus amendments. It is an official explanation, not a substitute for the regulation.⁸

A workplace exercise in which someone checks a summary against a public document is not, by that description alone, a high-risk system. Annex III lists, among others, AI systems intended for recruitment or selection, and systems intended to decide terms of a work relationship, promotion or termination, to allocate tasks based on individual behaviour or personal traits, or to monitor and evaluate performance. It also lists certain uses in educational institutions, including systems that determine access, evaluate learning outcomes or monitor students during tests. Classification depends on what the technology actually does. The service desk says Annex III rules apply from 2 December 2027. Prohibited practices, some transparency duties and the rules for general-purpose AI models are enforceable from 2 August 2026.⁹¹⁰

Questions leaders actually ask

Does 83.5% mean senior leaders were certified?

No. The report says 83.5% of directors, vice presidents and C-suite respondents had any AI training at their company. It does not say the training was extensive, assessed, or comparable to what entry-level staff received. The same page uses 28.9% once as "enough" training and once as "extensive" training.¹

Should a small employer raise the productivity quota because 57% of employers did?

No. The 57% comes from hiring managers and talent-acquisition professionals in a June 2026 employer survey, not from a measure of output. The worker survey's 49.9% is a different sample reporting higher expectations. Neither figure says which tasks got faster, or whether the check still happens.¹²

Will a quarter of the workforce leave because of AI?

No. 25.1% said they plan to leave because of how their employer views or uses AI. That includes concern about too much use, too little use and weak safeguards. It is a stated intention in one US panel, not a count of resignations.¹

How hilighter can help

hilighter does not run AI courses, certify literacy, score who should be trained, or predict who will resign. The practical question this survey leaves open is narrower: which roles received supervised practice, which skills a generated draft does not prove, and who is accountable for the judgment that remains.

hilighter helps teams map skills by role, compare people with those requirements in Skills & Talent Insights, record proof of capability in Skills Evidence & Recognition and set personalized growth paths. Its AI Job Impact view is indicative. It is not a prediction that a role will disappear. A free plan covers up to three job roles, with unlimited employees.

Sources

1. ZipRecruiter Economic Research, Nicole Bachaud, Better Work, More to Learn: The 2026 AI Worker Report, 26 October 2026. Fieldwork 8-21 September 2026, PureSpectrum panel of more than 1,500 verified US workers, described by the publisher as nationally representative. Figures used here include training by seniority (83.5%, 46.4%, 45.3%), self-teaching (37.2%, 54.7%, 45.4%, 29.9%), non-use (26.6%, 30.5%, 5%), any employer training (62.8%), the 28.9% "enough" / "extensive" wording, encouragement (39.3%), productivity expectations (49.9%), satisfaction (76.5%, 39.8%), task uses (52.1%, 41.6%, 40.9%), workplace use (82.5%), enjoyment (52.6%, 5.5%), leaving intentions (25.1%, 30.9%), perceived skill turnover (44.6%, 67.3% of daily users), and beliefs about mastery (34.3%, 40.2%).
2. ZipRecruiter Economic Research, Nicole Bachaud, More Jobs, Higher Bar: The 2026 AI Employer Report, page dated 29 July 2026. Fieldwork 11-18 June 2026, more than 1,000 verified US talent-acquisition professionals and hiring managers. Figures used here: 57% raised baseline productivity expectations; training offer of 22% mandatory for all, 23% specific departments, 34% optional, 17% none (the page also says the remaining 55% are optional or none; the shares are quoted, not forced to 100); 38% moved basic data processing off entry-level roles; 31% raised entry-level experience requirements.
3. World Economic Forum, with PwC, Artificial Intelligence and the Future of Entry-Level Work, publication page dated 22 June 2026. This article uses the page summary: more than one in three young workers in occupations with medium to high exposure to AI-driven task change, and the four-part framework (job access, job design, talent pipelines, education system alignment). The full paper was not re-read page by page.
4. OECD, AI and skills: What we know so far, 5 June 2026. The PDF was the text checked. Underlying studies are cited in the brief as Green and Lamby (2023), Lane, Williams and Broecke (2023), and OECD (2025) on SMEs. The brief says its most recent data date from late 2024.
5. European Commission, Joint Research Centre, Alice Bertoletti, Alessia De Santo, Yannis Galanakis and Wouter Vergote, AI and Skills, JRC148616, 15 October 2026, DOI 10.2760/7588639. The publications repository returned an access rejection on 1 November 2026, so the full PDF was not opened. The abstract used here (30% of EU workers in 2024-2025, 20% of enterprises by 2025, close to 40% of white-collar workers, 6% in elementary occupations) is the abstract on the publications record and the DOI page.
6. European Commission, Directorate-General for Employment, Social Affairs and Inclusion, AI and skills in Europe - Closing the adoption gap, 23 October 2026. Agrees on around 30% of workers and the occupation contrast. States about 30% of EU enterprises with ten or more employees, which does not match the JRC abstract's 20% of enterprises. This article does not merge those enterprise figures.
7. International Labour Organization, Lifelong learning and skills for the future, publication date 5 May 2026. This article uses the publication page, not page-level statistics from the full PDF. The page says quality training is especially out of reach for low-qualified and informal workers and for those in smaller enterprises.
8. European Commission, AI Act Service Desk, Article 4: AI literacy. The page says the text reflects the EUR-Lex consolidated version as at 27 July 2026, with Digital Omnibus amendments marked, and that the summary is not legally binding.
9. European Commission, AI Act Service Desk, Annex III, points 3 (education and vocational training) and 4 (employment, workers' management and access to self-employment).
10. European Commission, AI Act Service Desk, When does enforcement start? (Annex III rules from 2 December 2027; prohibited practices, certain transparency duties and general-purpose AI model rules enforceable from 2 August 2026).

Sources last reviewed: 1 November 2026. ZipRecruiter figures are the publisher's panel surveys, not audited training records. The 28.9% line is worded "enough" in the key findings and "extensive" in the body. The employer training shares are quoted as published and do not sum to 100. The JRC PDF could not be opened; the enterprise figure in the 23 October news page (about 30% of enterprises with ten or more employees) conflicts with the abstract (20% of enterprises). The WEF and ILO full documents were not re-read page by page. EU dates follow the Commission service desk and are not legal advice.