Australian employers are no longer asking whether Artificial Intelligence belongs in their workforce. The question is whether they can identify people who will use it well. A polished CV or title now reveals less about practical capability.
The demand is real. PwC Australia reports that AI-skill job advertisements doubled between 2024 and 2025. As AI tools become part of everyday work, a polished application alone is less useful evidence of practical capability. [1]
Do not ban AI or add a generic “AI experience essential” requirement. Design a fair, role-specific process that separates applied capability from superficial familiarity.
Start with the AI capability the role actually needs
“AI skills” is not a single skill. A product manager may need to evaluate AI-assisted research and trade-offs. A software engineer may need to integrate an AI service, validate generated code and work safely with data. The right standard must follow the work.
Identify the two or three workflows in which the hire will use AI, then specify the non-negotiable human responsibilities: validation, decision-making, privacy, security and accountability. This turns an ambiguous requirement into observable capability.
PwC recommends role-specific AI fluency: use cases, review points, risks and success measures. Apply the same principle to selection. [1]
| Role type | Evidence of applied AI capability | Human capability to test |
|---|---|---|
| Software engineer | Accelerates development, reviews outputs and integrates tools safely | Architecture, secure engineering judgement, debugging and code review |
| Data or analytics professional | Explores data and accelerates analysis while checking reliability | Data literacy, methodology, interpretation and communication |
| Product manager | Applies AI to discovery, prioritisation and workflow design | Customer insight, commercial judgement and stakeholder alignment |
| Cyber security specialist | Understands AI-enabled threats and applies automation responsibly | Risk assessment, incident judgement and governance |
Screen for skills, not buzzwords or job titles
The strongest AI-ready candidate may not have “AI” in their current title. They could be an analyst who improved reporting through automation or a developer who established code-review guardrails. Conversely, a CV listing every major model and platform does not demonstrate sound judgement.
A skills-based approach is a stronger starting point. Jobs and Skills Australia notes that titles, qualifications and keyword matching can obscure valuable transferable, cognitive and technical capabilities. [2]
At shortlisting stage, ask for concise evidence. Invite candidates to describe one AI-assisted workflow: the objective, their approach, what they verified, the result and what they would improve. This distinguishes experimentation from accountable practice without a lengthy unpaid assignment.
Make the rules for AI use clear before every assessment
Candidates should not have to guess whether AI is allowed in a take-home exercise, coding challenge or presentation. Unclear rules undermine fair comparison.
State the permitted level of AI use in writing. For a take-home task, allow realistic tool use but require disclosure of the workflow and review method. For a live problem-solving exercise, assess unaided thinking if that is the capability required. Apply the rule consistently.
The Department of Employment and Workplace Relations applies a similar principle: AI may support preparation, but candidates must demonstrate their own capability in interviews and skills-based assessments. [3]
A practical standard: assess the work as it will be done. If the role involves AI-assisted work, assess the candidate’s ability to use, interrogate and own that output. If the role requires unaided real-time judgement, test that directly.
Use a realistic work sample, not a trivia test
AI tools change quickly; memorising model names or prompt patterns is a poor proxy for professional capability. Use a job-relevant work sample instead.
For a developer, provide an imperfect AI-generated code snippet and ask the candidate to identify risks, improve it and explain their review process. For a data role, offer a sanitised data set and a draft AI insight, then ask what they would validate. For product, project or marketing roles, ask for a responsible AI-enabled workflow with success measures and human escalation points.
Keep the exercise time-boxed and proportionate. A 30- to 45-minute simulation followed by discussion often reveals more than an open-ended weekend task. Judge reasoning, relevance, limitations and decision communication—not a polished first output.
PwC finds that AI-exposed roles increasingly value human-intensive abilities such as judgement, empathy and creativity. Test the professional behind the tool, not merely their capacity to produce content quickly. [1]
Probe for judgement, governance and learning agility
Use the same core questions and pre-agreed scoring for every shortlisted candidate, so confidence or an AI-polished application does not dominate.
Ask candidates when they would not use AI, how they would validate an output, what data they would not enter into a public model, and how they would handle a persuasive but incorrect recommendation. Follow up with a real example.
Look for practical fluency, independent thinking, responsible practice and communication. Candidates need not know every tool in your stack; they should show curiosity, learning agility and the confidence to challenge an output when facts, context or ethics demand it.
| Scorecard dimension | Strong evidence | Warning sign |
|---|---|---|
| Practical task fit | Connects a tool or workflow to a business outcome | Talks only about tools, not the work or result |
| Human judgement | Validates outputs, explains limits and makes trade-offs | Treats AI output as inherently correct |
| Responsible practice | Identifies privacy, security, copyright and bias issues | Cannot explain data boundaries or accountability |
| Communication and learning | Explains reasoning clearly and adapts when challenged | Relies on jargon or cannot discuss their contribution |
Design for capability, not a perfect answer
A good AI assessment shows how a candidate approaches uncertainty. The best hires can frame a problem, choose tools deliberately, verify outputs and own the recommendation.
This is particularly important for early-career hiring. PwC reports that the most AI-exposed junior roles increasingly require judgement, leadership and stakeholder management. Employers who only test prior tool experience may miss high-potential candidates. [1]
For Australian organisations building AI-ready teams, a transparent, skills-first process identifies credible capability, protects fairness and communicates the standards your organisation expects.
Build Your AI-Ready Team with Confidence
Redwolf + Rosch can help you define critical capability, access specialist IT and digital talent, and strengthen your hiring process for a fast-moving market.