How is AI used in the workplace?
Most workplace AI use is generative text work done by individual employees: drafting emails and documents, summarising meetings and long reports, rewriting for a different audience, and explaining data. In the UK government's AI Adoption Research, 85% of businesses that use AI said they used natural language processing and text generation, and the most common business areas were marketing, administration and IT. Those figures are self-reported by UK businesses in early 2025.
A smaller share of use is AI built into other software, such as suggested replies in a helpdesk, meeting transcription or spreadsheet formula help. A smaller share again is AI that makes or shapes decisions about people, such as screening applications or scheduling shifts. That last category raises the hardest questions about fairness, transparency and employment law, and it deserves far more care than the first.
What are some examples of AI in the workplace?
Everyday examples look modest, which is why they are useful. Each one below pairs a task with the human check that keeps it safe.
An account manager drafts a client update
She supplies her notes and the project plan, then checks every date and deliverable against the plan before sending.
An operations lead summarises a supplier contract
The summary lists the clauses on notice, liability and price changes; he reads each clause in the contract before acting.
A support agent rewrites a reply for tone
The facts come from the knowledge base; the agent checks the policy wording has not changed in the rewrite.
A finance analyst explains a variance
The assistant describes the change in a supplied table; the analyst recalculates the figures and removes any cause the data does not show.
An HR adviser drafts a policy summary
The plain-English version is checked against the full policy, and the full policy remains the document staff rely on.
What are the benefits and downsides of AI in the workplace?
The benefits are faster first drafts, quicker access to the gist of long material and help for people who find writing slow. The downsides are confident errors, sensitive information going into the wrong tool, over-reliance, and worry among staff about what AI means for their jobs. Both lists are real, and a sensible rollout plans for both.
On the downside, NIST's Generative AI Profile names confabulation (confidently stated but false content), data privacy, harmful bias and human-AI configuration, which includes over-reliance and automation bias, among the risks specific to generative AI. On staff sentiment, an Acas survey of 1,023 employees in Great Britain, run by YouGov in March and April 2025, asked about people's biggest concern about AI at work: 26% said job losses, 17% said AI making errors, 15% said that AI is not regulated, and 17% said they had no concerns. The table puts the trade-offs side by side.
| Benefit | Matching downside | What keeps it in balance |
|---|---|---|
| Faster first drafts | Errors that read convincingly | A named reviewer who checks facts against the source |
| Quick summaries of long material | Missing exceptions and qualifications | Read the source for anything you act on |
| Help for slower or less confident writers | Over-reliance and loss of skill | Ask people to explain what they changed and why |
| Easy to start with existing tools | Sensitive data in the wrong account | Approved tools, business plans and clear information rules |
| Staff interest in learning new skills | Worry about job security | Early, honest consultation about what will and will not change |
What should employers do before introducing AI at work?
Employers should set clear rules, consult staff early, train people for the tasks they will actually do and keep a human check on output. Acas advises employers to develop clear policies on AI at work and to consult staff and their representatives, particularly where AI will change how roles operate, noting that changes to responsibilities could mean amending terms and conditions. It also advises that outputs be checked for accuracy, tone and bias, that AI be cited when used, and that existing data privacy policies apply.
If you operate in the EU or serve EU customers, note the AI Act's AI literacy provision. The European Commission's AI literacy Q&A says Article 4 has applied since 2 February 2025, requires providers and deployers of AI systems to take measures to support the AI literacy of their staff, does not require a certificate, and was amended by the Digital Omnibus in mid-July 2026, with AI literacy remaining an obligation but no 'sufficient' level set. It also warns that simply asking staff to read the instructions for use might be ineffective. This page is general information, not legal advice; take advice for your own organisation.
Want to see how this connects to a role?
Explore the planned team trial ↗Can a workplace make you use AI, and can you refuse?
Whether an employer can require you to use an AI tool depends on your contract, the instruction, how it changes your role and the employment law where you work, so there is no single answer. In many cases an employer can reasonably ask staff to use new software as part of the job, much as it might introduce a new CRM. Where AI significantly changes duties, Acas's advice to consult and possibly amend terms and conditions is relevant.
If you have a concern, make it specific. 'I am not comfortable pasting client contracts into a personal ChatGPT account' is a concern your employer can act on; 'I do not like AI' is harder to resolve. Ask which tools are approved, what data may go in, who checks the output and what training is available. If the disagreement becomes serious, Acas in the UK, or an employment adviser or union representative, can help. This is general information, not legal advice.
What does a careful introduction look like?
Quarry Street Studio is a fictional 18-person design agency. Before approving an assistant, the managing director holds a 30-minute team meeting. She explains why (proposal drafting is eating senior designers' evenings), what is approved (one business-plan assistant, two tasks: drafting proposal text from a call brief, and summarising client feedback rounds), what is not (no client files, no work on staff matters, no AI-generated imagery in client deliverables without the client's agreement), and that nobody's role is being cut as part of this change.
A junior designer raises a concern: clients sometimes send work under non-disclosure agreements. The rule becomes 'if a brief is under NDA, do not use the assistant for it'. A month later, the review finds the proposal task working well and the feedback summaries missing nuance, so the second task changes to 'summarise, then the account lead reads the original feedback before replying'. The consultation took one meeting and produced two rules the managing director would not have thought of alone.
When is more AI the wrong answer at work?
When the problem is unclear process, missing expertise or low trust, more AI makes it worse. An assistant cannot fix a team that disagrees about what a good proposal looks like; it will produce drafts nobody agrees on, faster. And if staff are worried about their jobs, more tools without a straight conversation will increase the worry, not reduce it.
Decisions about people, such as hiring, performance and discipline, are where the most care is needed. Do not hand these to a general assistant, and check your legal position before using any specialist tool for them.
How should staff learn to use AI at work?
Staff learn best on their own tasks, with clear rules and feedback on the output, rather than from a general overview alone. The CIPD has highlighted the role of people professionals in creating a safe space to test AI and identifying the points in a task that require human oversight. Our guide to AI literacy covers the broader foundation.
Tosheo.ai is being built to provide role-specific practice for teams of 5 to 200, ending each lesson in a hands-on task with a real AI tool, graded against a visible rubric. Managers are planned to see skill progress and completion, never raw chat logs. It is pre-build; planned prices are on the pricing page.
Sources and scope
DSIT, AI Adoption Research report (fieldwork February to May 2025), published January 2026 ↗
NIST, AI 600-1 Generative AI Profile, July 2024 ↗
Acas, 1 in 4 workers worry that AI will lead to job losses, 28 April 2025 ↗
Acas, One third of employers think AI will increase productivity, 15 May 2025 ↗
European Commission, AI Literacy: Questions and Answers, revised 27 July 2026 ↗
CIPD, Principles and guidance for safe and ethical use of AI in the workplace, 19 February 2024 ↗