The AI colleague is best understood as a set of capabilities that appear inside normal work systems. It may be a writing assistant in an email platform, a summariser in meeting software, a forecasting model in a dashboard, a recommendation engine in a customer relationship management system, or an AI agent that completes several steps in a workflow. This makes AI adoption less visible than earlier technology changes. A team may not “buy an AI system”; they may find AI already embedded in the software they use for communication, analysis, customer service, recruitment, finance, cyber security or learning.
This explains why prevalence is difficult to measure. Official UK research shows both growth and unevenness. DSIT (2026) reported that 16% of businesses were current AI adopters, but among those adopters, natural language processing and text generation were used by 85%, and 80% used AI at least weekly. ONS (2025) found that business use of AI had risen from 9% in September 2023 to 23% in late September 2025. These figures show that AI is becoming more common, but they also show that many organisations are still at an early stage. McKinsey (2025) similarly found that reported organisational AI use has broadened, while many organisations remain in pilot or experimentation phases.
The work impact is task-based rather than job-based. This reflects the human and machine collaboration perspective described by Daugherty and Wilson (2018), where value comes from redesigning work around complementary strengths. AI can draft, search, summarise, classify, translate, code, identify patterns, generate options and recommend next actions. These capabilities may remove small blocks of repetitive work, allowing people to spend more time on judgement, relationship management, problem solving and quality assurance. CIPD (2026a) argues that organisations need to look beyond standard governance focused only on data security and human accountability, because AI can also change expertise, role design and workforce stability. The “colleague” therefore needs a managed role in the workflow: what it can do, what it cannot do, when it must be challenged, and who owns the outcome.
Productivity gains are possible but not automatic. CMI (2026) reported that 70% of UK managers saw some productivity improvement from AI, but only 5% said the gains were transformational, while many organisations remained in testing or pilot stages. This matters because AI can create an illusion of progress: faster drafts, quicker summaries and more outputs do not automatically improve performance. Practitioners should ask whether AI has improved quality, reduced rework, increased consistency, shortened cycle time, improved customer outcomes or released people for higher-value activity. Without these measures, AI may increase activity while leaving the real work problem untouched.
The strongest future-of-work opportunity is human-AI collaboration. Microsoft (2025) describes emerging “frontier firms” built around organisation-wide AI deployment, human-agent teams and new forms of digital labour. However, the practical lesson is not that every organisation should automate as quickly as possible. It is that roles will need to change. People may become prompt designers, output reviewers, exception handlers, data stewards, workflow owners, AI trainers, ethics reviewers or “agent bosses” who supervise digital tools. The World Economic Forum (2025) predicts significant skill change by 2030, with AI and big data among the fastest-growing skills and employers expecting many workers to need upskilling.
Risk increases when AI is treated as a colleague without controls. AI can produce plausible but inaccurate information, reflect biased data, expose confidential information, recommend unfair action or weaken professional judgement through over-reliance. ICO (2023) makes clear that AI processing involving personal data must still meet data protection requirements, including fairness, transparency and accountability. NIST (2023) frames AI risk management around govern, map, measure and manage, while ISO/IEC 42001 supports an organisational management system for responsible AI use (ISO, 2026). These approaches point to the same conclusion: AI needs ownership, evidence, monitoring and review.
The future of work is therefore not a simple choice between humans or machines. It is a design challenge. Practitioners should identify where AI supports work, where it changes decision rights, where it creates dependency, and where human capability must be protected. They should make approved tools visible, train people to use them well, set rules for data and confidentiality, check outputs before decisions are made, and measure whether the technology is improving outcomes. Used responsibly, the AI colleague can reduce friction and extend capability. Used carelessly, it can speed up poor decisions and make accountability harder to trace.
Action Point
Choose one role or workflow where AI is already used or likely to appear soon. Break the work into tasks, then mark which tasks could be drafted, analysed, recommended or automated by AI. For each task, identify the human judgement needed, the data risk, the quality check and the evidence that would show whether AI improves the work.