The Right Graduates for the AI Era: A New KPMG Study on Human AI Collaboration Has the Answer

Today’s graduates arrive with top degrees, impressive internships, and high scores in AI literacy. But six months into your graduate programme, half of them are significantly outperforming the rest. The difference isn’t who codes faster or who knows more prompt engineering tricks.

A new study of 523 early-career professionals at KPMG shows that performance in the modern workplace depends on a specific style of human AI collaboration. The joint research with the McCombs School of Business at the University of Texas at Austin reveals that even with identical knowledge and tools, graduates produce dramatically different results based on how they supervise the technology.

If you’re running a graduate programme, this data changes your selection and development strategy. You aren’t just looking for users. You’re looking for managers of intelligence.

The Three Profiles of the AI Era

The KPMG/McCombs study established an AI-only baseline by asking agents to complete tasks without human help. Researchers then measured how professionals performed when working with those same agents. The analysis surfaced three distinct profiles that define the current graduate workforce.

AI Amplifiers (50.1%) outperformed the AI baseline. These individuals didn’t just use the tool. They orchestrated the workflow and refined results across multiple rounds. They treated the technology as a reasoning partner that required direction, oversight, and judgment.

AI Delegators (25.8%) produced results that merely matched the AI baseline. They accepted outputs with little scrutiny and added almost no personal value. While they weren’t the weakest performers, they failed to create any competitive advantage for the organisation.

AI Apprentices (24.1%) performed below the AI baseline. This is the most surprising cohort. These graduates scored higher than Delegators on foundational skills and domain knowledge. They critiqued the AI frequently. However, their critiques often steered the AI the wrong way or focused on irrelevant details. Their traditional skills didn’t translate into better human AI collaboration because they lacked the specific supervisory behaviours needed to improve the machine’s output.

Why Grades and Literacy Scores Are Not Enough

Traditional metrics of capability no longer predict success in a graduate role. The study found that Apprentices matched Amplifiers on critical thinking and domain knowledge. They were just as capable on paper. The difference lies in how they apply those skills to direct a machine.

Senior HR leaders often focus on upskilling graduates in technical AI literacy. While knowing how the tools work is necessary, it is not sufficient. The KPMG data shows that high-performers excel because they treat AI as a collaborator that needs a manager, not just a search engine. They move beyond routine prompting into sophisticated human AI collaboration by focusing on three specific behaviours.

Three Behaviours to Develop in Graduates

If you want to turn Apprentices and Delegators into Amplifiers, your development programme must move beyond tool training. Focus on these three core behaviours.

1. Framing Problems with Extreme Clarity

Amplifiers spend more time at the start of a task. They don’t just ask the AI to write a report. They frame the problem by defining the perspective the AI should take, providing concrete direction, and setting structural boundaries. This high-impact human AI collaboration starts with the human deciding exactly what ‘good’ looks like before the first prompt.

2. Interrogating the Output

Apprentices often fail because they chase the wrong errors. Amplifiers, however, use their domain knowledge to challenge the logic of the AI. They don’t just check for typos. They interrogate the reasoning behind a recommendation. This prevents the ‘accidental manager’ trap, where junior staff pass on work they don’t fully understand. For more on managing this risk, read about how to avoid creating accidental managers in your leadership pipeline.

3. Purposeful Iteration

Delegators accept the first answer. Amplifiers use feedback loops. They refine the work through multiple exchanges, treating the first draft as a starting point rather than a final product. This persistence is a primary signal of sophisticated human AI collaboration. It requires a mindset of continuous improvement rather than a ‘task-complete’ checkbox approach.

How to Spot Amplifiers in the Hiring Process

Identifying potential Amplifiers during recruitment requires a shift in how you interview. Stop asking if they have used AI. Start asking how they managed it.

Look for candidates who can explain how they would critique and improve an automated task. Ask them to describe a time they disagreed with an AI output and how they redirected the tool to get a better result.

Do they demonstrate iterative thinking? A candidate who shows their ‘working out’ and explains the reasoning behind their decisions is more valuable than one who only delivers a perfect outcome. You are looking for the ability to articulate why a certain path was taken. This helps prevent the rise of a toxic boss culture where results are prioritised over the integrity of the process.

The 5-Step Framework for Redesigning Your Graduate Programme

Redesigning your programme for the era of human AI collaboration requires a shift from content-based learning to behaviour-based simulations.

1. Conduct a Skills Check: Baseline your current graduates to see who falls into the Apprentice, Delegator, or Amplifier categories.
2. Implement Scenario Simulations: Use simulations that mirror real client work. Don’t just teach the tool. Ask graduates to improve a mediocre AI output under pressure.
3. Reward Iteration, Not Speed: Shift your performance metrics. If a graduate completes a task in five minutes using one prompt, they are a Delegator. Reward the graduate who takes 20 minutes to iterate the output into something superior.
4. Build a Champion Network: Use your existing Amplifiers as coaches. The study suggests scaling performance by turning these high-impact users into mentors for the rest of the cohort.
5. Redesign Role Responsibilities: Move graduates away from purely administrative tasks that AI can handle alone. Push them into supervisory roles earlier. This supports the ‘great unbossing’ trend, where removing layers of middle management requires junior staff to possess higher levels of judgment and autonomy.

Closing the Gap

The KPMG/McCombs research offers a clear roadmap. The difference between those who struggle and those who thrive isn’t about innate brilliance. It is about a coachable set of behaviours.

By focusing on how graduates direct, evaluate, and improve AI, you ensure your graduate programme delivers real value. Success in 2026 isn’t defined by having the best AI. It is defined by having the best human AI collaboration in your teams.

Identify the Amplifiers. Train the Apprentices. Challenge the Delegators. The future of your workforce depends on the graduates who know how to manage the machines.

Follow the data. Redesign your programme. Start today.