How to develop junior professionals when they’re using AI

AI is already changing the shape of entry-level professional work. Many of the tasks traditionally given to recent graduates and junior employees – including research, routine analysis, due diligence and drafting basic client communications – can now be completed faster (and often better) with generative AI.

But while AI has the potential to increase productivity and efficiency while reducing costs, it may also have removed a lot of tasks essential to building the knowledge and judgement needed to take on more complex work.

That leaves employers with a dilemma: if AI does the work traditionally assigned to junior professionals, how do they acquire the experience required to become senior employees?

If that’s where your organisation is at right now, read on.

AI isn’t making expertise less valuable

It is tempting to assume that increasingly capable AI will reduce the importance of human expertise. But emerging evidence suggests the opposite may be true.

A recent Harvard Business Review article reported on research involving 30 organisations across banking and finance, management consulting, technology, media and e-commerce. 

The researchers found that AI wasn’t simply changing how employees performed their work, but changing what employers expected from new hires. And on this front, it seems they now expect much more.

As AI takes away the routine tasks, employers increasingly want junior staff to redesign their workflows around the technology – drawing together knowledge from different sources and disciplines and critically assessing AI-generated material.

In other words, for junior staff, producing a competent first draft or basic analysis is becoming less valuable on its own. Employers want people who can determine what problem should be solved, decide whether the resulting answer makes sense and understand how it connects to the organisation’s broader objectives.

The difficulty is that these aren’t entry-level capabilities, but the domain of mid- and late-career professionals, not least because they depend heavily on context, pattern recognition and experience.

This means many employers are creating a contradiction: raising the bar for junior employees while automating many of the activities that traditionally helped them reach it.

Junior work was never just about producing the work

While the traditional model of professional development may not have been described as an apprenticeship, in practice, that’s often how it worked. 

Junior professionals began with relatively contained tasks: researching an issue, preparing a first draft, analysing information or assembling a presentation. A more experienced colleague reviewed the work, identified what they’d missed and explained how to make it better. As employees encountered more problems and received more feedback, they gradually learned to recognise patterns, anticipate risks and exercise better judgment themselves.

This model gave the organisation the work it needed – even if it could take some time – while also developing the employee.

After all, a junior lawyer researching a legal question wasn’t just doing entry-level grunt work, they were learning how legal principles operated outside a textbook (something they usually hadn’t done at university). 

Similarly, a graduate consultant building a presentation was learning to construct a client-focused argument. Meanwhile, a marketing assistant drafting campaign material was learning how audience, message and commercial objectives fitted together.

Mistakes were a painful but necessary part of that development. Struggling with an unfamiliar problem or choosing the wrong approach and having your work corrected helped employees understand not only what the right answer was, but why it was right too.

Generative AI can now perform much of this work faster and, in many cases, to a higher standard. But when it produces the research, structure and first draft, the employee usually doesn’t engage with the information in the same way.

When working faster means learning less

If someone isn’t engaging, they’re also generally not learning. Just because they can complete a polished report, presentation or analysis or with AI, that doesn’t mean they’ve mastered the concepts, as a recent Anthropic (i.e. Claude) study of 52 mostly junior software engineers sound.

THhe engineers were asked to complete a task using a Python library they hadn’t previously encountered. Half were given access to an AI assistant; the others completed the task without it.

The participants using AI finished slightly faster, (although the difference wasn’t statistically significant). But the more revealing result came afterwards. When their knowledge was tested on concepts they’d used only minutes earlier, the AI-assisted group scored an average of 50%, compared with 67% among those who’d completed the work manually.

The largest difference in the two groups involved debugging, or the ability to identify what had gone wrong in their code and explain why. And that’s precisely the capability developers need if they’re expected to supervise and correct AI-generated code.

But before we condemn AI and say there’s no point using it at all, it didn’t produce the same result for everyone. 

Because, while those who delegated their task to the model tended to learn the least, those who asked conceptual questions, requested explanations and tested their own understanding tended to perform much better. 

These employees retained most of what they’d learned. 

Using AI the right way in your workplace

The Anthropic study suggests the issue isn’t simply whether junior employees use AI, but how they use it.

Employees learn less when they treat AI as a contractor, handing over the task and accepting the finished product. They learn more when they treat it as a knowledgeable but fallible colleague: asking questions, challenging its reasoning and using it to deepen their own understanding.

That means approaching AI with curiosity and a degree of scepticism. Instead of asking it to “write the report”, an employee might first form their own view, then ask the model to:

• identify weaknesses in their reasoning
• explain unfamiliar concepts and further reading
• suggest alternative approaches
• challenge their assumptions
• identify what may be missing, or
• explain why one answer is stronger than another.

They should also be encouraged to test important claims against reliable sources and to be able to explain the final answer in their own words.

This takes more effort than simply generating an output and may also sacrifice some of AI’s immediate efficiency. But it makes employees use their minds and that mental effort is where most learning occurs.

The objective should be to have your employees using AI as a thinking partner that helps them understand more deeply.

Build the right kind of AI use into your policy

One way to embed this in your organisation is to develop your own policy and training for proper AI use. 

This means moving beyond what most organisations do (i.e. approve certain AI assistants and otherwise treat AI as a confidentiality issue), to explain how you expect staff to use AI without bypassing the thinking that develops their professional capability.

For junior employees, that could mean making sure everyone has to form an initial view before consulting AI. It could also mean asking junior staff to explain concepts in their own language when you’re reviewing their work, rather than judging only the polished output.

You should also identify which junior tasks should be automated and which still play an important developmental role so that the learning it provides doesn’t disappear with it.

When used this way, AI can become part of the apprenticeship rather than its replacement, and it can help junior professionals become more capable as well as more productive.

Want more?

Want help implementing effective AI usage in your workplace? Get in touch.