IMAGINING AN AUSTRALIAN WORKPLACE CULTURE IN THE AGE OF AI – JUSTIN TAUBER

As discussion mounted around the effects on contemporary and future society of Artificial Intelligence (AI), The Sydney Institute invited Justin Tauber, General Manager, Agentic Technology, Trust & Adoption, Salesforce ANZ and Roger Montgomery – founder and Chairman of Montgomery Investment Management Pty Limited to give their different views of the influence and change to come with AI. Offering both positive and negative impacts, Justin Tauber and Roger Montgomery addressed The Sydney Institute on Wednesday 29 April 2026.

IMAGINING AN AUSTRALIAN WORKPLACE CULTURE IN THE AGE OF AI

JUSTIN TAUBER 

The Sydney Institute has loomed large in my conception of what it means to be a public intellectual in Australia, and I want to thank Gerard, Anne, Jacqueline and all of you for the opportunity to speak today. It’s an honour and a privilege to be here.

I also want to reinforce how glad I am to be doing this talk in Australia. This is an extraordinary place to live and work, a fact that events globally only reinforce every day, and I want to use my time to talk about how to preserve what is extraordinary about it, while navigating the extraordinary transformation being wrought economically and socially by Artificial Intelligence.

This is an incredibly broad topic, however, so in the 15 minutes I have, I’m going to restrict myself to talking exclusively about AI at work. I won’t try to address the impact of consumer AI tools on education, relationships, mental health, or democracy. I also won’t indulge in speculation about Artificial General Intelligence. I’m not interested in whether large language models are conscious.

I’m interested in technology diffusion, because even if model development ceased today, the repercussions of the diffusion of existing AI capabilities through our work environments would be felt for years to come.

The thesis I want to put forward for your consideration is that there is room for different countries to develop different Workplace Cultures in how they respond to AI, and that we can and should take advantage of this opportunity to shape a uniquely Australian response, one that reflects our values and aspirations for our economy and society.

“Well, of course,” you might say. “Who wouldn’t want AI to be aligned to their values and aspirations.” However, there is a big difference between wishing for an Australian AI Workplace Culture, and believing that we have some degree of control over what that looks like.

We are awash with AI utopias and AI dystopias. Will it usher in a new era of abundance, or send us all to hell? Of course, these are attractive stories because they are so simple.

What I find pernicious about them, though, is that they are disempowering. Whatever will be, will be – our only role is to argue about who we think will turn out to be right.

To be sure, we are not all powerful, but we shouldn’t underestimate the level of agency we have with regard to the AI transition. Leadership today requires that we lean into uncertainty – that we imagine a diverse set of plausible futures, and publicly rehearse how we might survive and thrive in all of them. That is how we cultivate resilience, responsiveness and just maybe, hope.

So, I am going to talk about a few critical uncertainties that persist about what the future looks like for AI in the context of work. These fall broadly under the headings of Delegation, Augmentation and Orchestration.

They’ll leave us with a few questions about an AI-powered Workplace Culture

• How will we strike the balance between delegating to an agent and augmenting a human?
• Does our mental model of a knowledge worker need to be updated?
• Where and when will we reconfigure work, rather than just accelerate it?

Delegation

Let’s begin by clarifying the qualitative changes that have taken place in our relationship with these machines. We tend to talk about AI tools in terms of how smart they are, but I don’t think that’s what’s interesting about them. What’s exciting and scary about them is that they’re getting smart enough to delegate to. This is what makes AI agents different from traditional automation. Automation executes predefined processes. Agents can take discretionary action – ask clarifying questions, retrieve missing information, and make decisions without every pathway being specified in advance.

Second, AI systems seem to have no natural limit to the cognitive benchmarks they can meet, rising to PhD level capability in many domains, nor the timeframes they can operate across, rapidly extending task duration from minutes to days.

Third, we are seeing a shift from single systems to multi-agent architectures, where multiple agents play different roles, including evaluation and quality control. In many cases, what we are building are not just systems that act, but systems that review and refine their own outputs.

Collectively, these changes challenge the idea that judgement is an exclusively human domain. And this leads us to take seriously statements like “no job is safe from AI”.

Let me complicate this a little.

We shouldn’t assume that because a capability exists, it will be applied everywhere. Delegation is powered by capability, but constrained by reliability. So, if you want to understand how fast AI will diffuse through the economy, look at the pace of reliability, the speed of trust.

In 2022, Waymo launched its first robotaxi in downtown San Francisco. Four years later, these robotaxis still require human babysitters to take control in complex or unexpected circumstances. Many domains remain difficult to translate into usable context for an AI agent – because of informal processes, contested interpretations, missing or inherently noisy information.

Reliability seems to be increasing linearly, even while capability advance seems exponential. In other words, while capability is a lab property, reliability is a real world property – and recognizing this shifts the conversation from whether a system can do something, to where it can do something.

We also tend to assume that task delegation implies job destruction, but the economics is by no means so simple. The classic example here comes from 2016, when Geoffrey Hinton advised medical schools to stop training radiologists, as the task of reading a scan would soon be fully automated. Thankfully they didn’t, as the role of radiologist has not disappeared, even though AI today can read a scan more accurately than a human.

As researchers at the London School of Economics and Hong Kong University have pointed out – the labour market prices jobs, not tasks. And jobs bundle together tasks that are often difficult to separate – a radiologist doesn’t just evaluate scans, they also interpret clinical context, communicate and translate findings for other disciplines and critically assess processes and tools.

The impact of AI will depend not just on what can be delegated, but on how work is organised. How we resolve that uncertainty may have more to do with how we bundle responsibilities than how we execute tasks.

Augmentation

Let’s turn to augmentation. From the 20th century, we’ve inherited a mental model of workers as bundles of cognitive skills – valued for their ability to produce and manipulate information – in the form of reports, designs, plans, policies. This replaced an earlier model of workers as manual labour, brute horsepower – this change was critical because it allowed us to imagine the jobs that today we call knowledge work.

So, we’re used to thinking of augmentation in terms of skills and tools, even if those tools are digital. And using that mental model, AI augmentation looks like deskilling, because we now depend on AI agents to produce the artefacts we used to produce. Worse, augmentation can look like unfinished delegation, wherever recalcitrant humans have resisted automation.

We need a more sophisticated mental model of AI augmentation, because AI augmentation can be quite different from using a tool. It can be a lot more intimate.

What’s becoming clear is that people embrace different levels of integration into their professional practice, even their professional identity. Some people use AI, some people think with it, some think through it.

The Artificiality Institute has done some fascinating research, distinguishing three discrete dimensions of worker-AI integration. Let me summarise their findings for you.

One dimension of integration they call Cognitive Permeability – allowing AI to shape one’s thinking, engaging in a “genuine iterative dialogue — back-and-forth where AI’s responses redirect your reasoning in ways you didn’t plan”. A second dimension measures how much you allow AI to shape goals, frame problems and standards of success. Teachers and scientists embraced this kind of integration, while keeping their reasoning separate.

The final dimension is what they call Identity Coupling. When this is high (I quote): “You can’t describe your professional self without reference to AI. Your workflow, your sense of competence, your career trajectory have all restructured around the relationship.”

You might think that the goal is to maximise all three dimensions. We talk about 10X engineers becoming 100X engineers – as though it’s the engineer that is changing. In a real sense, it may be. But I don’t think that’s true for all. To me, what’s more interesting is the diversity of real integration styles that emerge from choices along these dimensions.

This complicates notions like AI fluency. A common mistake is to think that AI fluency is about understanding how AI works – as though fluency required becoming a data scientist or AI engineer. It might be more accurate to say that AI fluency is about learning how to get AI to understand how you work, and being able to assert appropriate boundaries and set the appropriate style of integration for your role, responsibility and professional values.

An AI workplace culture would take AI augmentation seriously, and guide us on when and how to accommodate diverse augmentation styles.

Orchestration

Finally, orchestration. The diffusion of electricity took over 40 years to impact productivity, and one factor that delayed that impact was the layout of factories. So long as they were constructed around a single central drive shaft – a holdover from the steam era – every workstation had to operate at the same pace, and every power issue affected every workstation. Only once the factory was redesigned to take advantage of decentralised power, and individual electric motors, could the promise of electricity be realised.

This story – where the technology had to wait for us to figure out how to organise ourselves – may turn out to be backwards when it comes to AI.

In his influential book, Reshuffle, Sangeet Choudary argues that economic systems (firms, industries and economies) are organised around managing three types of constraints: scarcity, coordination, and risk. By dismantling these constraints, AI restructures systems.

AI reduces scarcity of expertise – by providing a good enough alternative, or simply decreasing the cost of making that expertise accessible and applicable.

AI removes coordination constraints, by allowing “coordination without consensus” – a combination of instant translation and constant monitoring. Personal agents threaten to disrupt market aggregators because they simply don’t require centralised, standardised information from different providers to orchestrate a holiday.

AI dismantles some risk constraints too – by analysing and assessing events in real time, AI makes compliance activity no longer episodic, sample-based or reactive, but something that can potentially be managed continuously and proactively everywhere.

When these constraints are removed the system reorganises, and value accrues where new constraints emerge.

Here’s an analogy I find helpful. With the arrival of the internet, magicians suddenly found that they could not keep secret their knowledge of how to perform a trick. As a result, today, the magician’s value shifted from knowing tricks, to their skill at integrating them into a performance. In the same way, the value of expertise shifts from what we know, to what we’re prepared to be accountable for, and how we contribute to collective sense-making. The other parts of the radiologists’ bundle.

So, what might be the organising principle of an AI-powered workplace, when cognitive power (rather than electric power) gets redistributed? When coordination is cheap, what new assemblages of capabilities, powers or whole ecosystems might we imagine? What new, more complex products and services might we create, when more exacting standards of quality and safety become affordable?

There’s a line from product management blogger and systems thinker, John Cutler, that I particularly like: “People are overthinking how to use AI,” he writes. “I say, very simply, that they should use AI to do the things they know they should do and aren’t doing.”

At one level this is obvious. AI promises to be a circuit breaker for problems we have – costly coordination, poor customer service, poor employee experience. But there’s a deeper insight in this.

AI expands what is economically feasible, but it doesn’t resolve the tension between competing on cost and competing on value. A focus on cost assumes the value is fixed and pushes us to optimise. A focus on value opens up the possibility of doing new things well – and requires us to reconfigure how work is organised to deliver them. But it’s not just value in a commercial sense.

Companies use agents to:

• Accommodate out of hours, volatile or seasonal demand
• Allow customer-facing staff to focus less on finding and synthesising information and more on building human relationships
• Make compliance less reactive, through pre-emptive nudges

They are seeking an economic benefit, but each also represents an attempt to better embody a value – dependability and inclusion, intimacy and attention, discipline and accountability.

We like to say that values drive value – and I think this is not only true, but helpful. Because different choices about what is worth doing well lead to different ways of organising work.

So, the question is not whether AI will change our workplace culture. It will. But such a culture is not something that emerges automatically from the technology. It is something we shape – through the choices we make about how work is done, what is valued, and who is involved in deciding.