The Question Was Never Whether AI Would Change Work. The Question Is Who Designs the Change.
At the AHRI WA Conference this May, we said some things that needed to be said. Here’s the longer version, with footnotes, findings, and a healthy dose of honesty.
Let’s start with a confession. When we arrived at the AHRI WA Conference, we had a presentation about AI. So did, by our rough count, approximately everyone else. The difference, we’d like to think, is that ours wasn’t about the technology. It was about the work. Specifically, who designs it. And why HR has been sitting in the passenger seat of that conversation for long enough.
The tools are everywhere. The value isn’t.
Here is a fact that should be mildly embarrassing for the entire AI industry. Most organisations already have access to a capable AI assistant at no additional cost, bundled into the Microsoft licence they’ve been paying for. And yet fewer than one in five organisations have broadly deployed AI tools across their workforce. Of those that have, most are struggling to maintain even 25% of licensed users as daily active users.
Which means there are a lot of very capable AI assistants sitting in Microsoft tenants, quietly doing nothing, waiting patiently for someone to ask them to summarise a meeting.
The tools are not the problem. The problem is that most organisations have handed people a set of power tools and forgotten to explain what they’re building. Employees have curiosity. They don’t have clarity. And without clarity about where AI belongs in their actual work, not in theory, not in a use-case list, but in the specific tasks that make up their Tuesday afternoon, AI stays at the surface. A clever shortcut. A slightly faster way to write an email. Not a genuine shift in how work gets done.
- <1 in 5 organisations have broadly deployed AI tools across their workforce.
- 25+ yrs ahead of schedule, AI is developing faster than any expert predicted.
- 91% of high-maturity AI organisations have dedicated AI leadership, vs 37% of low-maturity organisations.
Three blockers. One uncomfortable diagnosis.
In our experience, and backed by a growing body of research, the failure to generate AI value clusters in three places. None of them are the technology’s fault.
1. Employees have curiosity but not clarity
Most people who’ve used AI tools have had a positive experience. They’ve summarised a document, drafted an email, generated some ideas. Lovely. But that’s individual productivity, and it’s the easy part. What employees are missing is a clear answer to the question: where in my actual work does AI belong? Which tasks should I hand off? Which should I keep? Which need redesign? Without that clarity, AI stays at the surface, a clever shortcut rather than a genuine shift in how work gets done. They know how to use the tool. They don’t know what to build with it.
2. Leaders are managing in a world that has already changed
Most leaders today built their careers in a world where managing performance meant managing people; their time, their outputs, and their development. It was a good model. For that world. But it doesn’t map cleanly onto a world where some of the work is now done by AI, where the human contribution is increasingly about judgment and context rather than task completion, and where ‘did they get the work done?’ is no longer a sufficient measure of anything. Leaders who haven’t updated their mental model of what good work looks like are inadvertently blocking value, not through resistance, but through the absence of a new framework. Which, if we’re honest, nobody gave them.
3. HR’s operating model is still running on 2012 assumptions
Job architectures designed around tasks AI can now perform. Performance frameworks measuring outputs without accounting for how they were produced. Learning functions teaching tools rather than building the underlying capabilities people need to work with AI effectively. Governance structures with no view on who owns AI work design decisions. This isn’t a criticism, it’s a diagnosis. The three-legged stool of business partners, centres of excellence, and shared services was genuinely good design for its time. That time was before AI could handle a meaningful portion of what shared services do. We should probably update the stool.
A recent Gartner CHRO leadership briefing found that organisations redesigning work and restructuring their HR operating model are twice as likely to exceed their revenue goals. Which is a polite way of saying that the organisations figuring this out are pulling ahead. Fast.
The capability gap nobody is measuring
When organisations talk about AI capability gaps, the conversation almost always heads straight for technical skills. Prompt engineering. Data literacy. Understanding how models work. These things matter. But they’re not the deepest gap.
The deepest gap is what we call digital literacy in context, it’s the ability to look at your own work, decompose it into its component tasks, and make good decisions about which of those tasks AI should augment, which it should take over, and which need to stay firmly human. That’s not a technical skill. It’s a work design skill. And almost nobody is building it deliberately.
The second gap is AI resilience; the ability to maintain human judgment, skill and well-being as AI takes on more. And this one is more concerning than most organisations realise.
What the Research is Finding on Skill Atrophy
- A 2025 Microsoft and Carnegie Mellon study found that the more people relied on AI tools, the less critical thinking they engaged in, making it harder to summon those skills when needed. Workers with AI assistance also produced a less diverse set of solutions, since AI tends to deliver homogenised answers based on its training data. The researchers described this as a potential ‘deterioration of critical thinking’ itself. Fun.
- Anthropic research published in 2026 found that participants who delegated coding tasks to AI learned the least, while those who stayed cognitively engaged fared better, though still below those who worked without AI at all. In other words, the more you hand off, the less you grow.
- Gartner now predicts that 50% of global organisations will mandate ‘AI-free’ skills assessments for employees by 2026 — an acknowledgement that critical-thinking atrophy from GenAI use is becoming a structural risk, not a theoretical one.
- The World Economic Forum estimates that 39% of existing skill sets will be transformed or become outdated between 2025 and 2030. Which means that the skills conversation HR is having today is already about yesterday’s workforce.
We are, in short, at risk of building workplaces that become more efficient in the short term while quietly hollowing out the human capabilities that make them valuable in the long term. HR is the function best placed to catch this before it becomes a crisis. This requires HR to be in a rather different conversation than the one it’s currently in.
“The organisations that struggle with AI over the next decade are unlikely to do so because of technology limitations. They will struggle because they fail to redesign work.”
Zest’s three-step playbook
At AHRI, we walked through the methodology we use with clients when helping organisations design AI-ready work. We’ll be honest: it’s not revolutionary. But it is deliberate, sequenced, and grounded in what actually works. Which, in our experience, is more valuable than revolutionary.
Step 1: Map the work
Before any technology decision, understand where the organisation actually sits on its digital journey, not where the strategy document says it should be. We use an HR Digital and Tech maturity assessment for exactly this reason. Then decompose roles and workflows into component tasks and apply a risk-and-value lens to each. High value, low risk: early candidates for AI delegation. High value, high risk: human in the loop, minimum. Low value, low risk: automation opportunity. Low value, high risk: worth questioning whether the task should exist at all. This last category tends to generate some interesting conversations.
Step 2: Decide the human-AI mix
For each task, make an explicit decision: Keep (human-led; judgment, relationships, ethics, accountability), Co-lab (human + AI, where AI augments and the human decides), or Delegate (AI-led, human sets parameters and reviews outputs). This must be facilitated with the people closest to the work, not decided in a meeting room by people who haven’t done the work since 2019. It’s deeply personal to each organisation’s culture, its risk tolerance, and its actual ways of working. Done well, it’s one of the most clarifying conversations a team can have. Done badly, it’s a spreadsheet nobody looks at.
Step 3: Protect careers and well-being
This is the step most organisations skip. Partly because it’s harder. Partly because by the time you’ve mapped the work and decided the mix, everyone is a bit tired. But it’s the step that determines whether AI-enabled work is actually good work, or just faster work that quietly exhausts and deskills the people doing it. Design deliberately against skill atrophy. Manage cognitive load: AI often adds review burden, not reducing it. Redefine the EVP for a world where human contribution looks different. And name the psychosocial risks that AI-enabled work creates: role ambiguity, surveillance anxiety, loss of meaningful work, and job security fear. These are compliance issues in Australia’s evolving WHS framework, not soft issues to be addressed at the next team offsite.
The HR operating model that needs to follow
No playbook survives contact with a broken operating model. We say this with affection for everyone who has ever built a beautiful framework and watched it quietly expire because the organisation had no structure to sustain it.
The three-legged stool, business partners, centres of excellence, shared services, was genuinely good design. For a world of relatively stable work where the main job of HR was to support business-as-usual operations. That world is gone, and the model needs to follow.
The AI-ready HR Operating System we presented has five components. At the centre: a Workforce Design Council, starting small and cross-functional, evolving into a standing governance body for AI work design decisions. Around it: HRBPs as Work Designers and Change Coaches (with deep business literacy, not just relationship managers who show up to hard conversations); an AI and Insights Hub as a shared HR/IT centre of excellence; Experience and Product Owners for each major HR domain (talent, performance, learning, employee services, workforce design, organised around outcomes, not processes); and People Operations that is AI-enabled rather than AI-replaced.
The fundamental shift is this: from HR as a function that responds to business needs to HR as a function that shapes how the business works. That is a significant identity shift for many HR teams. It requires new capabilities, new relationships, and in some cases new confidence, confidence that HR has the expertise and the mandate to lead these conversations, not just support them. Spoiler: we do.
Building confidence, not just use
There’s a distinction that gets lost in almost every AI capability conversation, and it’s important. Adoption is about usage. Confidence is about judgment. An employee can use AI every day, drafting, summarising, generating, and still make poor decisions about when to use it, how to verify its outputs, where to push back, and what to keep human. A hammer can be used enthusiastically in the wrong place. AI is similar, but with more consequences.
The AI Confidence Ladder is a tool for assessing where people actually are, designed for both leaders working with their teams and HR assessing organisational readiness. Three rungs.
Aware
Ad hoc, informal AI use. No shared language. Curiosity and anxiety in equal measure. People are experimenting, but they’re doing it quietly and alone, like a teenager trying a new hobby before they’re ready to tell anyone about it. Leaders: set guidelines, create safety to experiment. HR: map current usage and gaps.
Active
Deliberate AI use in specific workflows. A shared keep/co-lab/delegate view is forming. Skill atrophy risk is starting to show up, usually noticed first by the people doing the work, rarely acknowledged by the people managing it. Leaders: facilitate work mapping, introduce AI-free skill checks. HR: identify highest-value redesign opportunities.Adaptive
This is AI embedded in how work is designed, not just how it’s done. The team reviews its human-AI mix as AI capability evolves, which is to say: regularly. Governance feels like good practice, not compliance theatre. Leaders: coach ongoing redesign cycles. HR: measure AI resilience, not just AI use.
Most teams sit between Aware and Active. Most HR functions sit at Aware, aspiring to Active. Very few organisations are genuinely Adaptive, and that is fine. Adaptive requires structural maturity that takes time to build, and anyone claiming to be there already is probably describing a pilot team of twelve enthusiasts. The goal for the next twelve months is Active, consistently, across the enterprise. That’s ambitious enough.
The talent remix: what happens downstream
When you redesign work, you don’t just change tasks. You change skills, roles, careers and pipelines. This is the part of the AI conversation that makes HR genuinely uncomfortable, not because it’s new information, but because it requires HR to have honest conversations about workforce consequences before the workforce is already in consequence.
Harvard research tracking 245 million job postings found that junior role hiring flatlined from Q1 2023, precisely when generative AI emerged at scale, while senior hiring continued to grow. The apprenticeship model, where people built foundational capability through entry-level work, is under real pressure. The leaders who need to develop the next generation of talent learned their craft in a world that is no longer available for bookings.
The Talent Remix Map asks HR to think in three columns: at-risk roles (high volume, low variability, template-driven work that AI can largely absorb), the transition zone (where AI augments but humans remain essential, and where most roles currently sit, usually without realising it), and growth roles (work designers, AI integrators, people-facing roles where trust and empathy are the competitive advantage). The specific roles in each column will vary by organisation and industry. The column structure holds almost everywhere.
HR’s job isn’t to protect jobs. It’s to protect people’s ability to grow into the jobs that matter next. That means building a forward view before roles become redundant, not after, creating real pathways with genuine support rather than reskilling programs that look good in annual reports but don’t connect people to anything real, and redesigning learning and development around capability rather than tools. The half-life of any specific AI tool is short. The half-life of good judgment, relational intelligence and ethical reasoning is considerably longer. Build for the second one.
The 90-day path: where to actually start
Everything above can feel overwhelming if you try to do it all at once. The temptation, particularly after a conference, is to go back to your organisation, write a strategy document, present it to the executive team, wait for approval, form a working group, and discover six months later that nothing has actually changed and everyone has moved on to the next thing. We recommend a different path.
Ninety days. Three phases. One workflow. That’s the entry point. The fastest way to build credibility for the work design agenda is to demonstrate it in something real, contained, visible, and genuinely meaningful to the business. Then use that proof point to expand.
Phase 1: Choose (Days 1–30)
Pick one high-visibility workflow where AI is already touching the work or where pressure to use it is building. Recruitment is a good example. Case management. Performance reviews. Onboarding. Run an HR Digital and Tech maturity assessment to understand where the organisation actually sits. Assemble a small cross-functional team: HR, a line manager, and a couple of people who actually do the work. Not a steering committee. A working team.
Phase 2: Sprint (Days 31–60)
Map the tasks at the component level. Run the keep/co-lab/delegate facilitation with the team. Identify risks of skill atrophy and build AI-free skill checks into the new design, structurally, not as a nice-to-have. Then run the new design with real people doing real work. The first design will not be the final design. That’s not a failure. That’s the point.
Phase 3: Structure (Days 61–90)
Stand up a lightweight Workforce Design Council, three people is enough to start. Assess where the team sits on the AI Confidence Ladder using the sprint as evidence. Define three metrics: AI use, human capability health, and workforce design progress. These three together tell a far more complete story than adoption rates alone, and they’re the metrics that will actually matter in twelve months.
“Structure turns a sprint into a discipline. Without it, the work disappears when the energy does.”
The question HR needs to own
We closed our AHRI presentation with a thought that deserves repeating.
For years, HR has been positioned, and has sometimes positioned itself, as the function that manages the consequences of decisions made elsewhere. The restructure that HR implements. The redundancy process that HR manages. The culture problem that HR is handed and asked to fix, ideally by Friday.
The AI era is a genuine opportunity to change that. Not by claiming territory or winning turf wars with IT, but by bringing something that no other function has: the expertise, the ethics lens, and the human-centred design capability that sits at the intersection of work, people, and performance.
The organisations that will create the most value from AI are not the ones deploying the most technology. They’re the ones designing work most deliberately. And HR, when it steps into that space with confidence, is the function best equipped to lead that design.
The question was never whether AI would change work. It was always who designs the change. We think it should be HR.