Imagine your team has been tasked with delivering a high-stakes policy paper under intense time pressure. Everyone turns to generative artificial intelligence (AI), and within minutes, it delivers a full draft complete with structured arguments and authoritative-looking citations.
The team then transfers the AI-generated text into the corporate template, polishes the narrative flow and structure, and gives it one final check before sending it up the chain. Leadership takes a glimpse of the polished-looking draft, and – assuming the underlying research has already been verified – signs off and publishes.
This scenario is not hypothetical. South Africa’s Draft National AI Policy had to be withdrawn earlier this year after reviewers found that a number of the citations pointed to journal articles and authors that didn’t exist. Ironically, the mistakes were introduced by AI.
Similarly, last year, consultancy firm Deloitte provided a partial refund to the Australian government after admitting that generative AI had been used to help produce a commissioned document which contained fabricated citations and referenced quotes from sources that simply did not exist. The report cost Australian taxpayers A$440,000.
But the cost of letting AI do our work and thinking also comes at a cost to our brain.
Value of thought
As seen in the examples above, work tasks produced by AI often look polished on the surface but carry significant underlying flaws. This is a byproduct of how the modern workplace has conditioned us to prioritise the final deliverable – the report, the presentation, the assignment – over the actual process.
However, this strictly outcome-driven approach rarely rewards the process by which results are obtained. When we skip the process, we lose the ability to fully grasp what the output actually means.
This will inevitably result in us finding it challenging to critically evaluate the outcomes we consume. More importantly, it will also impact our capacity to discern facts from partial truths and lies.
Ultimately, the true value of knowledge work lies not just in the final deliverable, but in the clarity of thought and the messy, iterative process required to achieve it.
Blindly relying on AI
Recent empirical evidence also highlights a severe disconnect between perceived benefits of using AI and its improvement in quality.
A 2025 study by the University of Melbourne and accounting firm KPMG surveyed more than 48,000 respondents across 47 countries. It found two in three people (66%) use AI on a regular basis, while more than half believe their performance benefits from its usage.
The study specified that the primary driver of AI adoption was the fear of missing out (48%), leading users to prioritise the speed of delivery over the quality of the final output.
Yet 61% of respondents said they have had no AI training, while 60% also reported inappropriate, complacent and non-transparent use of AI in their workplace.
A separate study from July 2026 by the Centre of AI Safety, a San Francisco-based nonprofit research organisation, found that even the top-performing AI agents failed to complete roughly 85% of projects to a standard acceptable for commissioned work.
This indicates that even today’s best AI models still fall short of professional quality on most projects. It also shows we are relying on AI to generate final deliverables without understanding how it gets there.
This blind reliance strips away quality control, replacing genuine knowledge creation with an ever-increasing volume of automated “work slop”.
Improving cognitive skills
So how do we reverse this trend?
The answer isn’t simply banning or limiting AI in the workplace. Instead, we need to redefine our working relationship with AI and the outcomes it produces by taking a more human-centric approach.
First, organisations must set realistic time frames for deliverables that genuinely allow for people to think rather than focus on doing things quicker. There is a need to actively disconnect from AI tools during the brainstorming and structuring phases so ideas and thoughts are driven by actual human synthesis, not just algorithmic prediction.
Traditional whiteboard sessions – where teams physically map out their initial concepts through markers and pens – is a way this may be facilitated.
Second, we must cultivate a culture that is constructively critical of all outputs, whether generated by humans or AI.
This means actively questioning logic, structure, references and underlying assumptions. When teams are expected to defend the behavioural reasoning and methodology behind their work, it very quickly reveals who actually did the thinking and who merely copy-pasted from an AI agent.
Perhaps creating a critique group can create a safe, comfortable environment where feedback and comments can flow freely.
Technology can assist with certain processes, but it is dangerous to assume it can replace our ability to think critically. This is because the most valuable asset we have isn’t the polished final page – it’s the ebbs and flows of the cognitive process that is required to honestly write and evaluate one.
- is a Senior fellow, The University of Melbourne; RMIT University; Deakin University
- This article first appeared in The Conversation
