Signal or hallucination?

But is this time different? Recent headlines have linked AI to significant job cuts across technology firms such as Wisetech, Block, and Atlassian. However, these corporate announcements require context.

While workforce reductions appear significant, they follow a period of rapid hiring during the pandemic and post-pandemic recovery. In these high-profile cases, employment levels remain materially above pre-2020 baselines even after announced cuts, suggesting the contractions may reflect a ‘normalisation’ of headcount rather than AI impacts.

Broader evidence of an AI-led employment contraction is limited. At the aggregate level, labour markets remain resilient with little indication of structural unemployment¹. In the US, AI does appear to be having a negative impact on the employment outcomes of young people, particularly those entering the workforce², in occupations most exposed to AI automation. However, occupations that are augmented by AI continue to record employment growth across demographics³. Overall, AI is currently more visible in narrative than in measurable labour market outcomes.

Base case: augmentation over automation

In our view, the most plausible base case is that AI functions predominantly as an augmentative technology rather than a wholesale substitute for human labour. Central to this conclusion are the incentives and constraints that shape real world economic behaviour.

Firms optimise for growth, not just cost

Businesses optimise not only for cost minimisation but for output quality, growth, resilience, and competitive position. AI can reduce the cost of specific tasks, but its economic value is often highest when combined with human judgement, oversight, and contextual understanding. In practice, this favours a complementary relationship where workers become more productive, decisions improve in quality, and the scope of activity expands.

This will not be a universal experience. Some businesses, particularly those under margin pressure, may use AI primarily to reduce headcount and extract cost rather than to expand. But that response often says as much about the condition of those businesses as it does about the technology itself. Businesses don’t shrink to greatness, and contraction by weaker operators may ultimately create room for stronger firms to invest, grow, and take share.

Automation thresholds will increase

Recent academic research highlights the importance of task quality in shaping how firms deploy automation⁴. AI is likely to be adopted first in tasks where it can meet existing output quality at low risk. As these lower-value or more routine tasks are automated, human effort is freed up and reallocated toward activities that are more judgement-intensive, relationship-driven or context-specific.

This reallocation improves the quality of the remaining human-led tasks: more time and focus = better output. The quality threshold AI must then reach is increased, making it harder for further automation to occur⁵. This self-reinforcing augmentation loop sees people focusing more and more on higher value activities, in turn increasing the value of their labour.

Full delegation introduces risks

While AI can enhance productivity across a range of tasks, the case for fully delegating long, complex, or high-stakes work remains less compelling. Microsoft research finds that as task length and complexity increase, AI output quality deteriorates and document contents can become corrupted⁶. Errors compound across steps, objectives can drift over time, and inconsistencies can be difficult to detect.

Similar issues are evident in software development and cybersecurity. AI-assisted coding tools can accelerate development, but fully delegated code generation has been shown to contain more security vulnerabilities than human-written code⁷.

These dynamics reinforce the ongoing importance of human oversight and limit the extent to which full automation can be deployed. While the technology will likely improve, it will take time and resources. Productivity gains are therefore likely to be uneven and may be slower to materialise than implied by some of the more optimistic claims.

High unemployment will not be allowed to persist

At a macro level, there are social, economic, and political limits to how long high unemployment can persist without response. South Korea (late 90s) and Spain (early 2010s) are useful modern examples: in both cases, severe labour market deterioration was accompanied by social strain, political pressure and, ultimately, policy adjustment.

Elevated unemployment and the dislocation that accompanies it tend to provoke institutional adaptation aimed at restoring labour market stability. For that reason, the more plausible outcome is slower, more adaptive adjustment rather than a sustained period of mass unemployment.

Downside case: weaker demand but no sectoral collapse

A more pessimistic scenario could arise if the pace of AI-driven disruption materially outstrips the economy’s capacity to adapt. Even in that case, however, the implications for office demand are unlikely to be linear.

A reduction in labour input doesn’t necessarily require a one-for-one fall in employment. Given society’s limited tolerance for sustained unemployment, weaker labour demand would more likely be absorbed through changes in working patterns than through mass job losses. One potential mechanism is a reduction in average working hours, such as the adoption of a four-day work week. This would adjust overall labour supply while avoiding a commensurate decline in employment.

For office markets, fewer hours worked wouldn’t necessarily translate into a proportional reduction in space demand. Offices, like stadiums, are typically configured to accommodate peak occupancy (e.g. anchor days) rather than average utilisation. And in a future where the human-centric elements of work such as collaboration, innovation, and culture-building become more important than the routine, having sufficient fit-for-purpose space will be far more important than maximising workplace density.

Supply dynamics would also act as a moderating force. Elevated construction costs, tighter financing conditions and feasibility constraints are already limiting new development, reducing the risk that weaker demand outcomes translate into structural oversupply. At the same time, ongoing population growth should continue to support aggregate economic activity and space needs over time, even if AI reduces labour intensity in some functions. Over time, the withdrawal or conversion of obsolete stock would provide further stabilisation.

The chart below puts potential downside scenarios into context and illustrates that desirable stock (i.e. prime) should be relatively resilient even in the event of a severe reduction in demand. Flight-to-quality and the withdrawal of secondary assets from the market could, over time, wholly absorb a contraction in excess of 30%. Under such a scenario, investment outperformance would become increasingly dependent on asset selection and the alignment of building attributes and tenant experience with occupier needs.

Implications for office markets

Employment and demand mix

Under our base case, the office-using workforce continues to grow in aggregate as productivity gains support economic expansion and population growth underpins underlying demand.

AI adoption is likely to be slower in parts of the economy where compliance, accountability and implementation constraints are more significant (e.g. government). The effects may also be slower to emerge in industries such as construction and manufacturing, where office-based roles are more closely tied to physical operations.

By contrast, back-office processing functions that are more standardised and repeatable face greater automation pressure. Even if these roles are materially disrupted, the impact on the broader Australian office market should be limited given the diversity of the domestic demand base. In some offshore markets, where these functions account for a larger share of demand, the effects could be more pronounced. In Metro Manila for example, the Business Process Outsourcing sector accounted for 64% of leasing demand in 2025⁸.

Smaller occupiers, more fragmentation

We believe AI may shift occupier demand toward smaller businesses and drive a more fragmented tenant base. Lower start-up costs and the emergence of new business models may encourage new entrants, while existing small and medium-sized enterprises are, in our opinion, more likely than large corporates to use productivity gains to support growth rather than bank the savings through cost-outs. That tendency partly reflects the more dynamic and multi-functional nature of roles within smaller organisations, which makes them less amenable to automation than the more process-oriented roles in larger firms.

Smaller occupiers are also far less able to hand back space or sublease part of their footprint, reducing the likelihood of short-term speculative contractions. This theme was evident during the pandemic and post-pandemic recovery, when smaller occupiers accounted for a greater share of net office demand and were more likely to expand than contract.

Across office markets, this would support greater demand for smaller, more adaptable floorplates that are better suited to this size of tenant, along with assets that offer attractive shared amenity such as boardroom facilities and collaboration spaces. It may also increase the value of flexible leasing structures and speculatively fitted suites, given smaller occupiers are less likely to have the internal resources to plan and deliver bespoke fitouts.

Space use and workplace design

Densification has been gradually occurring for decades, and we don’t expect AI to materially reverse or accelerate this trend. The more meaningful shift is likely to be in workplace design, as floorplates evolve away from desk-heavy layouts toward more meeting rooms, collaboration spaces, and shared amenity to support in-person interaction for complex and valuable work.

This may increase the importance of technology integration and bandwidth capacity within buildings, as occupiers place greater value on seamless connectivity and the ability to support more data-intensive ways of working. Similarly, greater reliance on AI-enabled workflows is likely to elevate data security and control considerations for some occupiers, shaping preferences toward buildings with secure, resilient technology environments.

Quality and market polarisation

These shifts are likely to make office demand less homogeneous. Firms will adopt AI at different speeds, in different ways, and with different workplace requirements. Buildings that can accommodate varied and evolving occupier needs, both through physical adaptability and management flexibility, should attract a greater premium.

Location and quality preferences are also likely to strengthen rather than weaken. High-skill, knowledge-based work will continue to benefit from agglomeration, reinforcing the role of major CBDs. As the office becomes more focused on enabling high-value work outcomes, fit-for-purpose space and proximity to key stakeholders are likely to become higher priorities relative to rent minimisation.

We expect this to result in a wider divergence between prime and secondary assets, with higher-quality buildings better positioned to capture demand and lower-quality stock facing increasing pressure from functional obsolescence.

1. Evaluating the Impact of AI on the Labor Market: Current State of Affairs. Gimbel, Kinder, Kendall & Lee (Oct-25)
2. Young workers’ employment drops in occupations with high AI exposure. Atkinson & Yamco (Jan-26)
3. Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. Brynjolfsson, Chandar & Chen (Nov-25)
4. O-Ring Automation. Gans & Goldfarb (Jan-26)
5. You will comply with the AI. Ellis (Westpac) (Feb-26)
6. LLMs Corrupt Your Documents When You Delegate. Laban, Schnabel & Neville (Apr-26)
7. Human-Written vs. AI-Generated Code: A Large-Scale Study of Defects, Vulnerabilities, and Complexity. Cotroneo, Improta & Liguori (Aug-25)
8. Metro Manila office market shows strong 2025 performance. JLL (Feb-26)

Disclaimer

This material is prepared for discussion only and should not be relied upon for any other purposes. It has been prepared on a good faith basis but its contents have not been formally verified and no Cromwell entity or person accepts any duty of care to any person in relation to the information it contains. It should not be considered to be investment advice, marketing material or a promotion or offer of any Cromwell fund, product or services. Any person that wishes to invest in any Cromwell fund, product or services should refer to the relevant information or legal documents produced in relation to such opportunity before making any investment or other decisions. This document reflects the views of its author as at June 2026.