Graduates entering the workforce in 2026 will have learned in a fundamentally different way. They haven’t just studied with artificial intelligence – they’ve studied through it.
New research from Anthropic, based on nearly 600,000 anonymised university student interactions with the Claude AI model, reveals the scale and complexity of this shift. These students haven’t merely used AI for answers – they’ve co-created with it, shaped ideas in dialogue with it, and integrated it into the way they think and learn.
For employers designing the next wave of graduate programmes, this has profound implications. Traditional assumptions about skill development, assessment, and work readiness are being disrupted. To stay relevant and competitive, early careers recruiters must understand how these students are evolving – and how hiring and development practices need to evolve in response.
This article sets out what’s changing, why it matters, and how organisations can begin to adapt.
1. What the Data Tells Us
Anthropic’s Education Report reveals not just that students are using AI, but how deeply it has been embedded into the university experience. These are not one-off interactions; they represent new modes of cognition and content creation that are becoming second nature. Understanding this landscape is essential for predicting the behaviour and expectations of the 2026 graduate cohort.
1.1 High engagement in technical disciplines
The research shows that Computer Science students alone, though making up just 5.4% of US graduates, account for nearly 39% of AI-driven conversations. This is no coincidence. Technical disciplines are not just early adopters – they are AI incubators. Students in engineering and natural sciences are engaging in real-time coding, modelling and simulation activities with AI as a co-pilot.
This is partly a reflection of tool fit. Claude, like other large language models, is particularly strong in helping users work with structured data, troubleshoot code or automate tasks. But it also speaks to a mindset shift: these students are not waiting for institutional approval to use AI – they are already integrating it into their learning routines.
Implication for recruiters: When hiring from technical disciplines, be aware that AI fluency may be assumed, not taught. These candidates may expect to work with AI tools from day one – and may be more likely to ask how your teams approach AI during interviews. For non-technical functions, recruiters may need to probe more deeply into how candidates have been exposed to or experimented with these tools.
1.2 Delegation of higher-order thinking
Perhaps the most telling insight from the report is how AI is being used to support cognitively complex tasks – not just to recall information, but to construct, analyse, and evaluate.
- Creating: Students use Claude to co-write essays, generate practice questions, summarise readings and refine presentation scripts.
- Analysing: AI is used to interpret legal cases, explore ethical dilemmas, troubleshoot code and simulate scientific experiments.
- Applying and understanding: Less common, but still present, especially where AI helps bridge gaps in subject knowledge or reinforce foundational skills.
This is not rote learning. Students are engaging AI in activities that once defined independent thought. It’s not simply that they can complete the assignment – they’re using AI to help structure how they think about the assignment.
Implication for recruiters: Graduates may arrive with more advanced capabilities in synthesis, structuring and output generation than previous cohorts. But their development may be uneven. Where AI has been used uncritically – or as a replacement for learning – core skills may be weaker than they appear. Traditional interview methods may miss this entirely.
1.3 Four new modes of interaction
Anthropic’s research categorises AI use into four modes along two axes: intent (solve vs. create) and style (direct vs. collaborative).
- Direct Problem Solving
→ “Explain this theory,” “Give me the formula,” “What is the answer to...” - Direct Output Creation
→ “Write me a summary,” “Generate a cover letter,” “Create a lesson plan” - Collaborative Problem Solving
→ “Can you help me work through this step-by-step?”
→ “What if I approach it this way?” - Collaborative Output Creation
→ “Let’s refine this draft,” “How can I improve this paragraph?”
→ “Does this flow logically?”
These modes were evenly distributed across all conversations, with around a quarter of interactions falling into each. That balance suggests students are as likely to co-develop and refine with AI as they are to use it for quick answers.
Implication for recruiters: A graduate who has worked collaboratively with AI may have a better grasp of iterative problem solving and feedback loops – especially valuable in design, product, writing and customer experience roles. Your assessment processes should surface not just what candidates know, but how they explore and develop ideas.
1.4 Disciplinary divides in AI fluency
Not all students are arriving with equal experience. The report shows stark contrasts:
- STEM disciplines leaned toward collaborative and exploratory use.
- Business and Humanities showed more transactional engagement, often using AI for answers or basic content generation.
- Education (interestingly) skewed towards high levels of output creation – possibly reflecting usage by teachers and student-teachers alike.
These differences matter. A Computer Science student may have spent hundreds of hours debugging code with AI. A History graduate may have used AI to help summarise articles. A Business student may have used it to polish presentations – or to remove plagiarism risk from assignments.
Implication for recruiters: The question is no longer whether graduates have used AI – it’s how they’ve used it, and how that shapes their readiness for your specific roles. Recruiters must move beyond surface-level questions like “Have you used AI?” and dig into when, why and how it was used.
2. Why This Matters for Employers
With these shifts in mind, it’s clear that traditional early careers recruitment models are no longer sufficient. The emergence of AI as a core learning companion demands a corresponding update in how we define potential, evaluate readiness and build skills post-hire.
2.1 Process matters more than product
Graduates may arrive with brilliant portfolios – of work they cannot fully explain. Or with underwhelming CVs – masking deep fluency in problem-solving with AI.
AI blurs the lines of authorship. Was the code written independently or collaboratively with Claude? Did the essay come from deep understanding, or from reframing AI suggestions? In this context, output alone is no longer an accurate proxy for ability.
Recruiters must now find ways to evaluate the thinking behind the work.
2.2 The skills illusion
AI can help students leap ahead – but it can also disguise gaps. A graduate may seem proficient in analysis but lack the rigour to replicate that thinking unaided. Others may have depended heavily on AI throughout their studies and not developed a deep sense of personal capability.
The risk is overestimating performance in structured assessments or written tasks that align closely with AI’s strengths. Meanwhile, those who have used AI responsibly – to learn, refine and question – may not stand out without a way to demonstrate their critical engagement.
2.3 New expectations from candidates
Graduates who’ve learned alongside AI won’t expect to leave it behind in the workplace. They will anticipate environments where AI is part of the workflow – and will judge employers partly on whether they’re keeping up.
They may also challenge assumptions during recruitment:
- “Why can’t I use AI in this task?”
- “How do your teams use AI in practice?”
- “What support do you provide to build AI fluency on the job?”
Being caught off-guard by these questions is not just a risk to brand reputation – it’s a signal of disconnect from the talent market you’re trying to reach.
3. Rethinking Readiness: Strategies for 2026
Preparing for this shift does not require ripping up your recruitment process. But it does require targeted adjustments – especially around how you assess capability, build confidence in hiring managers, and support graduates post-offer.
3.1 Attraction: Update your message
Show that your organisation understands how the world of learning – and working – is changing. That doesn’t mean leading with AI in every post, but it does mean reflecting an environment where curiosity, experimentation and tool fluency are valued.
3.2 Assessment: Look for interaction, not isolation
Candidates will not be working without AI, so why test them in a vacuum?
Start by introducing assessment tasks that reflect AI-rich scenarios – not to trap candidates, but to observe their judgement. Did they ask the right questions? Did they verify AI suggestions? Did they challenge flawed outputs?
Several early careers teams are now trialling these “AI-in-the-loop” exercises as part of more holistic evaluations. These don’t just test knowledge – they surface how a candidate approaches uncertainty, feedback and intelligent support.
3.3 Manager enablement: Equip interviewers to ask better questions
Managers are already seeing AI-shaped outputs in applications – from personal statements to work samples. They need guidance on how to probe beneath the surface.
Good questions might include:
- “How did you use any tools to support this work?”
- “If the AI had given you an alternative suggestion, how would you have evaluated it?”
- “Can you walk me through your thinking from prompt to final output?”
3.4 Onboarding: Support mixed AI fluency
Not all graduates will be AI-proficient. Some may come from disciplines or institutions with stricter usage policies or fewer opportunities to engage. Build a simple diagnostic into onboarding to understand where each hire sits on the AI fluency curve, then support accordingly.
This might include:
- Introductory training on prompt engineering or responsible use
- Pairing more confident users with less experienced peers
- Embedding AI use cases into early project work
4. Questions to Ask Now
As you review your early careers strategy, these are the questions that matter most:
- How are AI tools being used in the learning environments of our target candidates?
- Are we still assessing capabilities that AI can now replicate?
- Have we updated our hiring criteria to reflect how modern learning shapes real-world readiness?
- Do our assessments surface process and judgment – or just outputs?
- Are our hiring managers equipped to interpret AI-shaped work?
- What would a graduate expect us to know – and offer – when it comes to AI?
5. Final Thoughts: From Preparedness to Advantage
The students graduating in 2026 will not only bring new skills – they will bring new norms. They’ll expect to work with intelligent tools, to collaborate across boundaries, and to be judged not just on what they know, but how they think.
For organisations willing to adjust, this is an opportunity. By updating assessment approaches, investing in early capability building and aligning with how students actually learn, employers can gain a clearer, fairer view of emerging talent.
Several organisations are already experimenting with assessments that incorporate AI into the process – evaluating interaction, judgement and applied thinking in real time. These innovations offer a glimpse of what hiring could look like in a world where AI is not a variable to control, but an environment to navigate.
The early careers landscape is changing. The question is no longer whether your next hires have used AI. It’s whether you’re ready to meet them where they are – and select for what truly sets them apart.
For employers exploring how to evolve graduate hiring in an AI-shaped world, tools such as genAssess offer a way to assess thinking, not just output – and to meet candidates where they are, not where we assume they’ve been.