The augmented decade: how artificial intelligence will reshape the way we work

Published 16 September 20266 min read

Written by

Fahad IshtiaqFounder, Vela

Which jobs will artificial intelligence replace? The concern deserves attention. For a business deciding where to invest, a more useful starting point is what its people could accomplish with better access to knowledge and assistance.

That question predates the current AI boom. In 1960, psychologist J. C. R. Licklider described people and computers contributing different strengths to thinking and decision-making (Licklider, 1960). Douglas Engelbart developed a related argument in 1962: computing could expand people’s ability to understand and solve complex problems (Engelbart, 1962).

Their idea, augmentation, now has practical support in workplace research. AI has helped people resolve customer problems, complete writing assignments and perform consulting tasks more effectively. The strongest business case starts with that additional capability: what could your people do with expertise available at the moment they need it?

Businesses can put that capability to work by redesigning tasks and developing the judgement to use it well.

Where assistance improves performance

A field study following the phased introduction of an AI assistant to 5,172 customer-support agents found that issues resolved per hour increased by about 15% overall. Less-skilled and less-experienced agents improved by about 30%. With assistance, agents two months into the role performed comparably to colleagues with more than six months of experience who worked without it (Brynjolfsson et al., 2025).

Expertise became available during the customer conversation. Newer staff could apply effective practices while they were still learning the job. The most experienced agents benefited much less. Where would access to that kind of help make the greatest difference in your team?

In a preregistered experiment with 453 professionals, access to ChatGPT reduced the time spent on assigned writing tasks by 40%, while independently assessed quality rose by 18%. Lower-performing participants benefited more, narrowing the performance gap (Noy & Zhang, 2023).

Consulting research shows why the task matters. Among 758 Boston Consulting Group consultants, GPT-4 users completed 12.2% more tasks within the model’s capabilities and finished them 25.1% more quickly, with higher-quality results. On a task outside those capabilities, AI users were 19 percentage points less likely to find the correct answer (Dell’Acqua et al., 2026).

The researchers call this a jagged technological frontier. Success on one task offers limited assurance about another, even when they appear similar. A business needs to test the work it intends to delegate and determine how errors will be detected. Presentation quality alone is an unreliable guide to correctness.

Start with the task

A job contains several activities, each with different requirements. Some can be automated; others benefit from assistance; some still demand a person’s decision. Automating part of a role can therefore augment the person performing it.

IMF researchers estimate that almost 40% of global employment is exposed to AI, rising to about 60% in advanced economies. Roughly half the exposed jobs in advanced economies could benefit from AI complementarity (Cazzaniga et al., 2024). Exposure describes the relationship between AI capabilities and job tasks. It is not a forecast of job losses.

Statistics Canada’s analysis, published in 2024 using May 2021 employment data, classified 29% of workers as highly exposed with high potential complementarity, 31% as highly exposed with low complementarity, and 40% as less exposed. University-educated workers were particularly exposed (Statistics Canada, 2024).

The scale of exposure makes AI a business-wide question. The useful starting point is still a particular workflow.

Consider a hypothetical support team. AI might retrieve an approved policy and draft an answer, while an agent decides whether the policy fits the customer’s circumstances. If the system produces fluent responses using outdated information, faster drafting may increase rework. If it retrieves reliable information and flags exceptions, the agent may resolve cases more effectively. The design of the process determines what the additional capability accomplishes.

What happens to the time saved?

Task improvements do not automatically produce proportional gains across a business or an economy. Their value depends partly on what happens next.

Research linking Danish adoption surveys to administrative records found no detectable average effect on earnings or recorded hours two years after ChatGPT’s launch, despite task changes and reported productivity benefits (Humlum & Vestergaard, 2026). In US survey data from November 2024, respondents reported savings equivalent to about 1.4% of total work hours (Bick et al., 2025).

Does faster completion release capacity for valuable work, improve service, or move a document more quickly into the same approval queue?

Forecasts offer possibilities rather than a settled answer. Acemoglu estimates an initial upper bound of 0.66% for the cumulative increase in total factor productivity over ten years, reduced further when harder-to-learn tasks are considered (Acemoglu, 2025). OECD economists estimate an annual contribution of 0.25 to 0.6 percentage points to total factor productivity growth over a ten-year horizon (Filippucci et al., 2024).

McKinsey estimates a contribution of 0.1 to 0.6 percentage points to annual labour productivity growth through 2040, conditional on adoption and productive use of workers’ released time (Chui et al., 2023). Goldman Sachs projects a potential increase of about 7% in global GDP over a ten-year period (Goldman Sachs, 2023). These measures and assumptions differ, so their headline figures cannot form a simple league table.

Businesses can act without resolving that forecasting debate. They can measure whether a particular implementation improves the outcomes that matter to them.

Adoption takes work

The 2026 AI Index from Stanford’s Institute for Human-Centered Artificial Intelligence (Stanford HAI) reports AI use at 88% of surveyed organisations in 2025, alongside $285.9 billion in US private AI investment. Agent deployment nevertheless remained in the single digits across nearly all business functions (Stanford HAI, 2026). Trying AI and integrating it into operations are separate achievements.

The wider transition also requires infrastructure and skills. The International Energy Agency projects data-centre electricity consumption rising from 485 terawatt-hours in 2025 to approximately 950 by 2030, with supply constraints affecting expansion (International Energy Agency, 2026). Employers surveyed by the World Economic Forum expect 39% of existing worker skills to change or become outdated by 2030. That expectation reflects several forces, including AI (World Economic Forum, 2025).

These constraints help explain why change will proceed unevenly. IMF researchers also warn that differences in access and readiness could widen inequality within and between countries (Cazzaniga et al., 2024).

Expertise still needs somewhere to grow

The support-agent findings suggest AI can help people learn. Another question follows: will businesses continue to provide the work through which experience develops?

Stanford researchers examining US payroll data through June 2026 found no evidence of widespread economy-wide displacement. However, employment among workers aged 22 to 25 in AI-exposed occupations stood 19% below the level implied by keeping pace with less-exposed peers. Reduced hiring accounted for much of the gap, which was concentrated in occupations where AI usage substituted for human tasks (Brynjolfsson et al., 2026).

The authors describe these findings as early, descriptive indicators. They do not establish that AI caused the gap. The Danish study also found no detectable effect of employer chatbot adoption on early-career employment in its sample (Humlum & Vestergaard, 2026).

The organisational concern remains worth examining: if junior learning opportunities disappear while demand for experienced judgement rises, how will future experts develop?

History shows how the creation of new work can reshape the outcome. Autor and colleagues find that most US employment in their study is in job specialties introduced since 1940 (Autor et al., 2024). Acemoglu and Restrepo describe how new tasks can counterbalance the displacement of existing work (Acemoglu & Restrepo, 2019).

Anthropic’s usage data show both approaches already in use: collaborative patterns account for just over half of sampled Claude.ai conversations, while automation dominates its first-party API traffic (Anthropic, 2026). The choice of how to use additional capability is already being made, task by task.

Making additional capability useful

Four priorities put augmentation into practice.

Test a defined task. Compare performance with and without AI using representative cases. Include exceptions and establish who can recognise a wrong answer.

Plan how the benefit will be used. Decide whether saved time should improve responsiveness, expand capacity or allow more attention to difficult cases. Measure that outcome alongside speed.

Develop judgement deliberately. Give staff opportunities to practise, question outputs and learn from corrections. Assign responsibility for checking consequential decisions.

Evaluate the complete workflow. Track quality, rework, resolution and customer outcomes. A quicker first step is useful when the whole process improves.

AI can extend what people accomplish. Give a team better access to expertise, design work around that capability and help people develop the judgement to use it. Then ask the question that matters: what can they now do well that they could not do before?

Sources

Every figure in this article was checked against these sources.

All posts