For the past two years, the public conversation around AI has been dominated by one question: will it replace jobs?
It's an understandable worry for many. Tools like ChatGPT and Claude can write, research, summarise, code, and synthesise information at a speed that would have seemed impossible only a few years ago. When people see those capabilities for the first time, the conclusion often feels obvious.
And for the last few years it hasn’t helped that both Sam Altman (OpenAI) and Dario Amodei (Anthropic) fueled intense anxiety about an "AI job apocalypse." Amodei publicly warned that AI could wipe out half of all entry-level white-collar jobs, while Altman stated that entire entry-level categories were at severe risk.
If AI can do human work, surely it will replace people in the workplace? What we are seeing in our work within public and private sector entities large and small is that the reality inside many organisations is very different. A realisation that over the last few months has changed Altman and Amodei’s narrative as well where Altman recently admitted he was "pretty wrong" about AI eliminating entry-level roles, while Amodei shifted his stance to suggest automation will expand human workloads rather than destroy them.
In instances where AI training programs have been held for large corporate teams, a surprising contradiction often comes up. Many employees who have completed foundational AI training still do not have access to the basic tools needed to apply what they have learned.
In some cases, participants will arrive at advanced AI courses having done little to apply AI to solve their operational and business problems since their previous training. Not because they didn't want to, but because they couldn’t as they simply didn't have access to the tools or been provided time to innovate, experiment and learn.
That experience shines a light on a challenge that receives far less attention than automation or job displacement, which is that most businesses are still struggling with AI enablement leading to a massive AI capability overhang (the significant gap between the advanced theoretical capabilities of modern AI systems and how they are actually deployed and utilised in the real world).
We talk about AI as if every business is operating at the frontier. The reality is that there is a vast chasm between those who are truly AI-native and those who are not.
The size of that gap is far larger than many businesses and IT leaders realise.
The assistant versus replacement debate
One reason job displacement is such a dominant topic is that modern AI models are undeniably impressive. As a perspective, the model you are currently using will be the worst that it will be.
Viewed in isolation, its capabilities can feel threatening, but capability alone does not equate to replacement.
An AI model can automate repetitive manual tasks, and can assist with writing, coding, analysis, or research. What it cannot do is provide organisational context, align stakeholders, exercise judgment, understand politics, manage relationships, or make decisions on behalf of a business.
Without people directing and prompting it, AI is just a tool.
Yes, AI can write code better than almost anyone on the planet, but without a person behind the machine prompting, directing, evaluating, and validating the output, it cannot accomplish anything meaningful on its own.
The missing ingredient is still human judgment and the context we provide.
Why productivity gains don't necessarily eliminate jobs
A more useful way to think about AI is not as a replacement for people but as an accelerator. When one part of a process becomes dramatically faster, bottlenecks will inevitably crop up elsewhere.
Research is a good example. Finding and synthesising information that once took days can now take minutes. However, that information still needs to be evaluated, interpreted, socialised, implemented, and acted upon.
The bottleneck simply moves to another place. This has been seen repeatedly in software development. AI can help teams build initial versions of products far faster than before. What follows, however, is a flood of feedback, change requests, testing requirements, stakeholder input, and new ideas.
The faster teams move, the more opportunities they create. As a result, rather than reducing work, AI often sees it grow.
Many organisations are discovering that increased productivity leads to expanded expectations. If a team can deliver two projects instead of one, the response is rarely to work half as much. Usually, the response is to take on more work, so the principle is not doing more with less (i.e. fewer people) but doing more with the same.
That changes jobs, but it doesn't necessarily remove them.
The real risk may be at entry level
However, it’s wrong to think that there are no workforce implications at all, and the area that deserves the most attention may be entry-level roles.
Many professions rely on apprenticeship models where junior employees develop their skills and competence through repetition and experience. If AI significantly reduces the volume of routine work or traditionally junior tasks such as initial research or team administration, fewer entry-level positions may be required.
Have a conversation with an intern, and you will likely find they understand the idea of AI-assisted work, but will ask a simple question: "Without having vast amounts of practical experience, how do I know if the answer is right?” And they’d be spot-on with this reflection.
Without years of experience, validating AI-generated outputs becomes difficult.
This is another emerging challenge. AI may democratise access to expertise, but experience still matters. In many cases, it matters even more because someone must determine whether the machine's answer is correct.
The shadow AI problem
Organisations attempting to control AI adoption through policy alone are setting themselves up for disappointment. Whether they like it or not, shadow AI already exists.
Employees are using AI tools whether companies formally approve them or not - many just choose not to disclose the fact.
Research has identified what some call an "AI disclosure penalty." This is when employees who admit to using AI may be perceived as less capable, less hardworking, or less deserving of recognition. That creates a dangerous dynamic: People continue using the tools but become less transparent about them, and they look for workarounds.
Here, the answer is not stricter policies, but creating environments where people can experiment safely, learn openly, and discuss AI usage without fear or stigma.
What AI skills mean
Perhaps the most interesting question firms face today is one that HR teams increasingly ask: What exactly are AI skills?
The answer is probably not prompt engineering. In fact, the most valuable AI capabilities are surprisingly human.
The ability to define the right problem.
The ability to provide context.
The ability to evaluate outputs critically.
The ability to use data responsibly.
The ability to apply judgment.
And perhaps most importantly, the ability to work transparently.
These are the skills that determine whether AI creates value or simply generates more content. The businesses that will benefit most from AI over the next decade will not always be those with the most advanced models. They will be the ones who give employees access to the tools, time to experiment, guardrails to use them safely, and a culture that encourages transparency.
Before worrying about whether AI will replace workers, many companies still have a more immediate challenge to solve, which is the need to enable them.
Both leading experts in their fields, De Gregorio is a partner leading AI Advisory and Enablement and Goddard is a partner leading Software Development at technology and management consultancy, iqbusiness
