AI in the Workplace: Eight Workforce Insights from Australian Leaders

A lot is being written about AI and the future of work. When you're responsible for deciding where to invest, what needs to change and how to bring people with you, keeping up with the latest can feel like a job in itself. 

To break through the noise and get to the crux of what organisations are experiencing in practice, Launch brought together executives across technology, business and government from two of Australia’s major employment markets, Sydney and Melbourne. Through a series of leadership roundtables, participants shared their direct experience of implementing AI, redesigning work and leading people through change. 

Their collective insights revealed how closely connected these workforce decisions are. Removing routine tasks raises questions about how junior employees will learn. Increasing output creates more work for the experts reviewing it. Giving teams new tools asks more of managers who are already stretched. 

Across the discussions, eight findings emerged about how work is changing, how organisations are responding and what this means for their people. 

How Work is Changing

1. Roles are changing faster than organisational structures 

In several represented organisations, the work people were doing had already outgrown the way their roles were defined. As AI took on parts of execution, employees spent more time defining problems, directing agents and reviewing results. Those responsibilities called for a broader understanding of the business and the consequences of the work. 

The implications for people can be easy to underestimate. An engineer coordinating several agents is taking on a form of supervision, possibly with little preparation. A familiar job title can conceal a considerable change in expectations. 

Some organisations had appointed dedicated leadership for AI-enabled work and workforce redesign. Others were still exploring what their future structures should look like. “How do agents and people coexist in a workforce?” one participant asked, describing work underway to test different operating models. 

The World Economic Forum’s Future of Jobs Report 2025 estimates that 39% of workers’ core skills will change by 2030. Of every 100 workers, 59 would need training, with 11 unlikely to receive it. These global projections reflect several economic and technological forces, including AI. 

Practical steps to redesign roles as AI changes work 

• Map your team’s recurring tasks, including where people direct or review AI output. Record the skills each task requires; a framework such as SFIA can help. 
• Compare that map with current role descriptions. Update responsibilities, development plans and support for people taking on supervision of agents. 
• Use the task map when reviewing vacancies, and check applicable consultation requirements before introducing significant changes. 

2. Domain expertise and critical thinking are essential 

“Domain expertise has never been more important.” 

Experienced people understand what a good result should look like. They know which questions matter, recognise unsafe assumptions and can tell when a convincing answer needs investigation. That judgement also helps them decide which problems are worth pursuing in the first place. 

Several leaders described the cost of getting that direction wrong. One explained how poorly framed instructions could send multiple agents through weeks of unproductive work, consuming “so much time and so many tokens” while moving in the wrong direction. Another was rebuilding a system after accumulated poor-quality AI work had damaged it. The apparent pace of production had made it harder to see the problems developing underneath. Participants called this an “illusion of speed”. 

Review workload ultimately has to grow as output accelerates. Yet, Australian findings from the University of Melbourne and KPMG’s 2025 study found that 57% of employees who use AI reported relying on output without evaluating its accuracy, and 59% reported AI-related mistakes in their work. 

Expert oversight also needs to continue after deploying an agent. One participant described evaluating deployed agents every week because “with every model release, agents need to be tweaked.” The organisation still needs someone with enough expertise to recognise when performance has drifted and enough authority to intervene. 

As reliance on AI grows, retaining those people becomes a question of operational continuity. 

Practical steps to strengthen expert oversight of AI 

• Match the level of review to the impact of a potential error.  Assign an expert to approve client-facing work and other high-stakes outputs before they are used. 
• Give reviewers written quality standards and examples of acceptable work. Track errors and time spent on rework alongside any time saved. 
• Assign responsibility for testing deployed agents regularly and after model changes, with authority to intervene when results drift. 

3. New entrants need opportunities to understand the work 

“Who’s going to train the young engineers?” 

Routine tasks have traditionally helped people learn a discipline: how a system behaves, why a process exists, what happens when something goes wrong and how to work through the mistakes. Now AI can complete parts of that work before a junior employee has developed the knowledge to assess it. The challenge for businesses now is how to train young people and keep a pipeline to develop future experts. 

Several participants said they'd pay a premium for an experienced hire who could direct AI effectively. As one put it: “I’m happy to pay that domain expert 100k more if they can help me define which use case is the right use case.” Repeated across a team, those decisions can leave fewer people developing into the next generation of specialists. The immediate hiring choice affects succession and the future supply of expertise. 

Research from the US supports that concern. Stanford researchers analysed payroll records covering millions of workers. Between November 2022 and June 2026, employment among 22–25-year-olds fell by around 11% in occupations more exposed to AI, while growing by around 10% in less-exposed occupations. The researchers found that reduced hiring accounted for most of the divergence. These findings do not establish that AI caused the changes, but they give employers reason to examine what is happening to entry-level opportunities. 

While there is undoubtedly pressure to favour existing AI expertise, one participant viewed that the loss of junior pathways was inevitable. “These are active choices that companies and businesses are actually making,” they said. “You do have the ability to maintain and keep those pipelines.” Employers’ decisions about junior roles will help determine whether people have opportunities to develop into the senior specialists and leaders their organisations will need. 

Their organisation was investing in that development through university partnerships. Students worked on meaningful network projects from their first year, with a pathway into internships and graduate roles. With experienced employees approaching retirement, the organisation had recognised the need to build future capability well before those people left. 

Other participants discussed pairing younger employees with experienced specialists, combining opportunities to experiment with guidance on real work. Making that approach useful requires clear learning goals and time for supervision. As the tasks within junior roles change, employers need to plan how people will gain experience and take on greater responsibility. 

Practical steps to develop early-career talent as AI changes work 

• Identify the foundational tasks juniors need to understand and where automation is removing opportunities to learn. Build supervised practice into their roles, including tasks they should first attempt themselves before leveraging AI. 
• Pair early-career employees with experienced colleagues on real projects. Set learning goals and allocate time for supervision, feedback and discussion of their reasoning. 
• Review entry-level hiring alongside succession plans. Consider university or TAFE partnerships, internships and graduate pathways where important skills will take years to develop. 

How Organisations Are Responding 

4. Value and cost need to be measured honestly 

Leaders reporting gains were open about the effort it took to achieve them. One organisation described a major setback in an internal software project: despite experienced engineers, training and guardrails, it had run well over budget and still could not deliver the functionality required. Substantial engineering work remained before deployment, and the organisation was bringing in other products to address the gaps. 

Another engineering leader estimated that output initially fell to around a fifth of its previous level as the team worked out how to operate with AI. 

"At the start, actually, our productivity dipped massively." 

These experiences brought the costs of learning, rework and disruption into the discussion about returns. An organisation needs to know how the work performs today and what improvement would justify the investment. The discussions distinguished three streams of AI activity, with different ways to assess each: 

Individual productivity was especially difficult to connect to organisational performance. Time saved on email or administration may be useful, but the business still needs to understand what changes in their role and output as a result. Participants questioned what token consumption and output volume could tell them about business value, particularly when the activity was spread across many employees and tasks. 

The costs can become clearer as adoption expands. One participant described a heavy user generating around $3,000 in charges in two days. The organisation was considering different limits by role and demonstrated value: “The majority will have a spending limit, and then the special case will have a higher limit.” A prototype’s economics can change considerably once more employees or customers begin using it. 

Teams were responding by choosing models suited to the task and scheduling appropriate work overnight on slower, cheaper models. They were also assessing whether a use case required AI. These decisions belong in the business case, alongside ongoing review and support costs. 

Practical steps to measure AI value and manage costs 

• Before a pilot, record current performance and agree a business outcome to measure, such as cycle time, quality or cost. Include time for learning and initial disruption. 
• Cost the expected scale of adoption, including consumption, expert review and ongoing support. Assign someone to monitor spend and review exceptions to limits. 
• Set checkpoints to decide whether to continue, change, or stop the work. Use results against the baseline to inform that decision. 

5. Governance needs clear ownership inside everyday work 

Employees are being encouraged to experiment with AI while organisations are still establishing approved tools, policies and risk settings. Leaders described the difficulty of giving people room to explore while maintaining oversight of what they were building, the information they were using and the decisions they were making. 

Some organisations were already bringing tools back under central oversight after an initial period of broad access. Employees had developed similar capabilities independently, creating duplication and questions about maintenance and responsible use. Elsewhere, leaders were considering how their controls would need to develop as their organisations grew. 

The gap between adoption and oversight creates conditions for shadow AI: using AI outside an organisation’s knowledge, approval or controls. In the University of Melbourne and KPMG’s 2025 Australian findings, 60% of employees who use AI hid their use at work, and 48% used it in ways that contravened company policies. Only 30% of employees said their organisation had a policy guiding generative AI use.   

Ownership becomes particularly important when an experiment turns into something the business relies on. Participants recognised this from earlier experiences of staff-built systems that became critical processes. An employee-built agent can create similar dependencies, leaving questions about who checks its performance, approves changes and maintains it when its creator moves on. 

Some organisations were assigning agents named owners and defined responsibilities, much as they would manage a member of a team. Others were incorporating AI into existing risk checks. Participants also argued for building governance into systems and workflows from the outset, so employees had a clear, practical way to use AI within agreed boundaries. 

The Australian Government’s Guidance for AI Adoption provides resources including policy and AI-register templates. Its six practices cover accountability, impacts, risks, information, testing and monitoring, and human control. OAIC guidance also addresses the privacy implications of selecting and using AI products, including due diligence and oversight. These resources can help organisations translate their expectations into processes employees can follow. 

Practical steps to establish AI governance and accountability 

• Create an AI register that records each tool or agent, its purpose, the data it uses and a named person accountable for it. Include staff-built tools the business already relies on. 
• Clearly set out which tools are approved and how data may be used, in language every employee can understand. Give them a clear contact for questions and proposed experiments. 
• Agree how each system will be tested, monitored and escalated when something goes wrong. Define who can pause its use and how affected work will continue. 

What It Means for Talent and People 

6. Curiosity and problem solving are becoming stronger hiring signals 

As AI becomes part of more roles, participants were looking for evidence that candidates would be willing and able to learn how to use it. Personal interest and experimentation helped those in the hiring process to assess how someone might respond to unfamiliar tools, changing expectations and the inevitable mistakes involved in learning. 

One participant asks all candidates, “What have you built yourself?” They were interested in the attempt, including when the result was unfinished or unsuccessful, and whether it revealed someone who was teachable. They described looking for people open to “that uncomfortable journey of learning it, making those mistakes”. 

Launch consultants have seen this in their own work, with clients asking to see what candidates were building and experimenting with outside work, recognising that their current roles may offer limited opportunities to demonstrate those capabilities. 

Participants were also seeing employees develop capability through self-directed learning. One leader described people progressing from little familiarity with the tools to deploying a useful application within two weeks and presenting its value to the board within six weeks. The organisation had provided access to tools and opportunities to experiment, allowing employees to demonstrate capabilities beyond their existing responsibilities. 

Professional experience remained important in these discussions. Participants valued people who could combine field knowledge with a willingness to try unfamiliar approaches. The hiring signal was evidence that someone could apply what they knew while continuing to learn. 

Practical steps to assess and develop AI learning capability 

• Include curiosity and learning ability in assessment criteria where roles are changing. Agree what evidence would demonstrate those qualities alongside the professional capabilities required. 
• Ask candidates about something they have tried, the problems they encountered and what they learned. Accept examples from work, study or personal experimentation. 
• Give employees opportunities to develop their AI capability through relevant work, with access to approved tools, guidance and feedback. Look for existing employees whose capabilities may extend beyond their current responsibilities. 

7. Organisations are rethinking how they secure and retain capability 

Leaders described reviewing vacancies against how the work had changed before deciding whether to recruit a replacement. Some requirements could be met through automation, internal development or external support. Others had disappeared as delivery changed. One leader reported margins on certain services rising from 30% to 50% over one to two years and was firm about the roles affected: “Those jobs are never coming back.” 

Uncertainty about future structures was also influencing how organisations engaged people. One participant expected to assemble internal teams around projects and supplement them with more contractors and freelancers. Others observed greater reluctance to commit to permanent headcount while organisations tested their requirements. This discussion raised a concern about who would invest in developing people when organisations accessed more capability through contingent workers. 

Where organisations were recruiting, participants described difficulty defining the skills required. Titles such as “AI engineer” and “forward deployed engineer” could conceal different combinations of technical, functional and design capability. 

Identifying that capability in applications brought further challenges. A participant described recruiters returning to manual screening after AI-written CVs caused applicants to rank similarly highly in automated assessments. Greenhouse’s November 2025 US findings reflect the pressure on hiring processes: 65% of hiring managers had caught applicants using AI deceptively, and 34% of recruiters spent up to half their week filtering spam and junk applications. Participants emphasised the continuing importance of human assessment and relationships in understanding who could perform the work. 

Retaining those people was another concern. Participants anticipated more movement as employees who had stayed put through economic uncertainty became willing to change jobs. They expected employers’ promises about development to face greater scrutiny, with staff questioning: “What are you actually doing to support our growth?” Participants expected opportunities to learn and develop to become more important in employees’ decisions about whether to stay. 

Practical steps to recruit and retain skills for AI-enabled work 

• Review vacancies against the work and skills required. Agree whether to develop capability internally, recruit permanently or engage contingent support, and define the outcomes before selecting a job title. 
• Check that screening and assessment methods can distinguish relevant capability in AI-assisted applications. Use direct discussion and an appropriate work exercise to explore candidates’ experience and reasoning. 
• Review development and retention plans for people with critical skills. Build knowledge transfer into contingent assignments, with named employees responsible for retaining and applying that knowledge. 

8. The capacity to absorb change is being tested 

Managers are carrying responsibility for AI adoption alongside their existing workload, often while developing their own understanding. They are also supporting employees who feel anxious about the future or are struggling to change familiar ways of working. The pressure to increase output is landing on teams already stretched. Microsoft’s April 2025 Work Trend Index found that 53% of leaders believed productivity needed to increase, while 80% of the global workforce, including leaders, reported lacking sufficient time or energy to do their work. 

An executive discussing large operational teams described what this meant for managers: “The minute they open their doors, their capacity is full, and then we ask them to do more.” Their organisation used capacity planning, wellbeing support and psychosocial risk assessments to understand and respond to those pressures. 

They described change management as a larger challenge than adopting the tools. Weekly team meetings gave employees opportunities to discuss what was working, what they disliked and where additional guardrails were needed. The team continued to adjust how it worked, even after substantial productivity gains. 

Participants also raised gaps in board and executive AI literacy. One described the risk of senior decision-makers relying heavily on managers to explain developments while those managers were already responsible for implementation. Building understanding at the top of the organisation helped ensure that expectations, resources and the pace of change reflected what teams were experiencing. 

Practical steps to manage workload and support AI adoption 

• Review managers’ existing workloads before assigning AI initiatives. Decide what can be paused, reduced or reassigned, and where dedicated change support is needed. 
• Hold regular team check-ins about workload, fatigue and what is working. Use that feedback in psychosocial risk assessments and follow through with changes to the work. 
• Build board and executive AI literacy around the systems being introduced. Give decision-makers visibility of the demands on teams so they can make informed decisions about resources, pace and priorities. 

Explore what these findings mean for your team 

Launch offers a free presentation and discussion of the roundtable findings for leadership, HR and technology teams. We share how organisations are approaching AI adoption, the challenges they are encountering and the practices emerging from their experience.