AI skills gap: why the real shortage is inside everyday work

UK organisations are not short of AI headlines. They are short of the time, judgement and workflow design needed to turn new tools into reliable work. That is the tension inside the 2026 data. ONS reported on 2 July 2026 that 29% of UK businesses with 10 or more employees were using at least one AI technology, up 8 percentage points on a year earlier. But later ONS analysis found adoption has been relatively shallow, and only 10% of AI-using businesses reported using AI extensively. ([ons.gov.uk](https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/bulletins/businessinsightsandimpactontheukeconomy/2july2026/pdf))

This matters because the labour response is already visible. ONS says businesses most commonly integrate AI skills by training or retraining existing staff, not by replacing people or hiring large numbers of new specialists. Around 10% report automating or replacing roles with AI technologies, while recruitment of new staff with AI-related skills ranges from around 2% among the smallest firms to 10% among businesses with 250 or more employees. Even so, only 11% of businesses reported that more than half of their workforce had received AI-related training. ([ons.gov.uk](https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/articles/artificialintelligenceinukbusinesses/2023to2026))

What the AI skills gap actually means

The phrase AI skills gap often sounds like a shortage of elite engineers. In practice, the evidence points to a broader problem. Skills England's framework divides AI capability into three domains, technical, non-technical, and responsible and ethical skills, and maps them across entry, mid and managerial levels. It also says non-technical AI skills are the most urgently needed. OECD reaches a similar conclusion from a wider labour-market view, reporting that fewer than 1% of workers need advanced AI skills, while most need digital skills, the ability to use, analyse and interpret data, and a mix of managerial and human skills such as problem-solving, creativity and innovation. ([gov.uk](https://www.gov.uk/government/publications/ai-skills-for-the-uk-workforce/ai-skills-tools-package))

• Technical skills are practical skills such as writing structured prompts, using embedded AI features and monitoring AI-supported tools in context. ([gov.uk](https://www.gov.uk/government/publications/ai-skills-for-the-uk-workforce/ai-skills-tools-package))
• Non-technical skills include evaluating outputs for accuracy and relevance, applying professional judgement, and explaining results to colleagues. Skills England says these are the most urgently needed. ([gov.uk](https://www.gov.uk/government/publications/ai-skills-for-the-uk-workforce/ai-skills-tools-package))
• Responsible and ethical skills include identifying bias, applying data protection and privacy practices, and making accountability visible when AI is used in work. ([gov.uk](https://www.gov.uk/government/publications/ai-skills-for-the-uk-workforce/ai-skills-tools-package))

This is also why the gap shows up differently by role. Skills England's sector work points to finance teams needing stronger governance and interpretation skills, health and care teams needing ethics and output interpretation, and creative teams needing prompt writing, copyright awareness and originality checks. That report is based on workshops and desk research rather than a representative survey, so it is best read as directional evidence, but the pattern is useful: the missing capability is often role-specific judgement around AI, not simply access to a tool. ([gov.uk](https://www.gov.uk/government/publications/ai-skills-for-the-uk-workforce/executive-summary-with-introduction-and-next-steps))

Specialist shortages are still real, and it is worth separating them from the wider workforce issue. A DSIT-commissioned AI Labour Market Survey, published on 28 January 2026 and based on surveys and interviews, found that 97% of respondents identified at least one AI labour-market skills gap, 35% reported difficulty filling AI roles and 28% said technical skills shortages had affected their ability to meet business goals. A separate UK summary report on AI skills for life and work estimated around 158,000 people were in expert, specialist and implementer AI roles in 2024, about 0.5% of the labour force, while AI-related postings made up about 1.7% of all UK job postings between 2021 and 2023. ([gov.uk](https://www.gov.uk/government/publications/ai-labour-market-survey-2025-report/ai-labour-market-survey-2025-report-executive-summary))

Those specialist roles also sit behind a tall entry gate. The same AI skills for life and work report says 99% of AI expert vacancies required at least a bachelor's degree, with 37% asking for a PhD and 29% a master's degree. So the shortage at the top end is not imaginary. But it also shows why a simple plan to hire a few experts will not solve the problem for most employers. ([gov.uk](https://www.gov.uk/government/publications/ai-skills-for-life-and-work-summary-report/ai-skills-for-life-and-work-summary-report--2))

Why the gap persists

The hard part is not simply the absence of courses. Skills England's 2025 workforce study points to a more basic set of obstacles: inconsistent use of the term AI skills, weak foundational digital literacy, fragmented local provision, slow curriculum updates, fragile funding and limited employer understanding of what they actually need. ONS reaches a similar conclusion from another angle, finding that among firms planning to adopt AI, the main barriers include the level of AI expertise and skills, cost, and difficulty identifying business use cases. ([gov.uk](https://www.gov.uk/government/publications/ai-skills-for-the-uk-workforce/barriers-to-ai-skills-development))

• Employers often do not know whether they need baseline AI literacy, role-specific tool use, or governance skills, so they buy vague training or postpone it. ([gov.uk](https://www.gov.uk/government/publications/ai-skills-for-the-uk-workforce/barriers-to-ai-skills-development))
• Many learners still struggle with basic digital tasks, which means some AI courses start one level too high. ([gov.uk](https://www.gov.uk/government/publications/ai-skills-for-the-uk-workforce/barriers-to-ai-skills-development))
• Provision is fragmented, which makes it hard to compare courses or build progression routes. ([gov.uk](https://www.gov.uk/government/publications/ai-skills-for-the-uk-workforce/barriers-to-ai-skills-development))
• Cost and time matter, especially for SMEs, and underinvestment is common when leaders have no clear AI plan for their business. ([gov.uk](https://www.gov.uk/government/publications/ai-skills-for-the-uk-workforce/barriers-to-ai-skills-development))

There is a second reason the problem feels muddled. Employers are adopting AI for ordinary operational reasons, such as reducing administrative burden, improving communication and making better use of data, but many still treat training as an optional add-on rather than part of process redesign. That mismatch helps explain why adoption can rise while confidence, depth of use and training coverage lag behind. The final sentence is an inference, but it is grounded in the pattern across the UK evidence. ([gov.uk](https://www.gov.uk/government/publications/ai-skills-for-the-uk-workforce/ai-skills-tools-package))

What employers should do next

The practical response follows from the evidence above. If most organisations are adapting through internal training, while deep use remains limited, the next move is to treat AI capability as a workflow design issue, not a side project for a few enthusiasts. ([ons.gov.uk](https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/articles/artificialintelligenceinukbusinesses/2023to2026))

1. Start with tasks, not job titles. ONS says improving business operations is the most reported use of AI. Skills England's adoption pathway begins with identifying practical use cases such as reducing administrative burden, enhancing data use and improving communication. A useful audit asks where work already involves drafting, searching, summarising, routing or checking, and where human judgement is still essential. ([ons.gov.uk](https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/articles/artificialintelligenceinukbusinesses/2023to2026))
2. Split training into three layers. Most organisations need a baseline layer for everyone, a role layer for specific teams, and a governance layer for managers and decision-makers. That structure follows directly from Skills England's framework of technical, non-technical, and responsible and ethical skills across different job levels. ([gov.uk](https://www.gov.uk/government/publications/ai-skills-for-the-uk-workforce/ai-skills-tools-package))
3. Put managers in the first cohort, not the last. Entry-level users need practical confidence, but mid-level and managerial staff are the ones who guide others, redesign workflows, monitor use and handle accountability. If managers cannot judge where AI is useful or risky, adoption stays patchy and staff training becomes generic. ([gov.uk](https://www.gov.uk/government/publications/ai-skills-for-the-uk-workforce/ai-skills-tools-package))
4. Use pilots with an assessment gate. Skills England's pathway explicitly separates awareness, exploration, assessment and experimentation. That is a sensible sequence. Before rolling out licences widely, test one workflow, document the failure points, and decide what staff need to know about quality, privacy and escalation. ([gov.uk](https://www.gov.uk/government/publications/ai-skills-for-the-uk-workforce/ai-skills-tools-package))
5. Measure whether work changed, not whether people attended training. This is an editorial inference rather than a direct survey finding, but it follows from the fact that adoption depth remains modest and training coverage remains low. For most teams, the useful measures will be output quality, turnaround time, error rates, rework, escalations and compliance issues, not course completions alone. ([ons.gov.uk](https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/articles/artificialintelligenceinukbusinesses/2023to2026))

What this means for workers

For individuals, the message is less dramatic than many job-market summaries suggest. You do not need to become an AI specialist to stay valuable. The more durable mix is domain knowledge plus AI literacy: knowing how to prompt, check, question, document and escalate. OECD defines AI literacy broadly as understanding, using and monitoring AI applications with critical reflection, without needing to build models yourself. ([gov.uk](https://www.gov.uk/government/publications/ai-skills-for-the-uk-workforce/ai-skills-tools-package))

The wider pressure is not going away. The World Economic Forum says 59 out of every 100 workers globally are expected to need training by 2030, and 85% of surveyed employers expect upskilling to be a core workforce strategy. OECD also reports that more than half of workers using AI receive employer-funded training. Put together, that suggests the organisations most likely to cope well will be the ones that make training routine, role-specific and accountable. That final sentence is an inference, but it is a cautious one. ([weforum.org](https://www.weforum.org/publications/the-future-of-jobs-report-2025/digest/))

If you lead a team, the clearest next move is small and concrete. Pick one high-volume task, decide what good judgement looks like, then train people to use AI within that boundary. The AI skills gap becomes much more manageable when it stops sounding like a talent myth and starts looking like ordinary work design. ([gov.uk](https://www.gov.uk/government/publications/ai-skills-for-the-uk-workforce/ai-skills-tools-package))