Are We Outsourcing Thinking?

Photo by Roger Ce

One of the most important questions around AI is no longer whether it can make individual workers more productive. A well-used AI tool can help a good developer move faster, a good analyst test more ideas, and a good marketer sharpen messaging.

The more important question now is what happens when AI stops being used to support thinking, and starts being used as a substitute for it? The distinction is subtle, but is utterly central to the future of ‘work’.

At one level businesses are understandably moving quickly. The productivity opportunity, and the cost pressure, are real. The competitive pressure is real. No management team wants to look back in three years and realise they moved too slowly while competitors rebuilt their operating models around AI.

Yet in the rush to adapt, there is a risk that given the rapidly changing landscape companies could make rash decisions that will have lasting negative effects, or require unwinding.

AI is forcing every business to ask a difficult question: what work actually needs a human, and why? ‘Work’ is context, judgement, memory (personal and corporate), relationships, accountability, taste, intuition and the ability to discern why something matters in the first place.

If those things are stripped out too quickly, the short-term efficiency gain can become a long-term organisational cost.

The New Efficiency Cycle

We are already seeing the first phase play out.

Meta, as an example, is undergoing another significant internal AI-driven reorganisation, moving thousands of employees into new AI-focused roles while also preparing further layoffs, with reports suggesting the company is trying to flatten management structures and redirect more resources towards AI investment.

As we explored previously, Spotify is also doubling down, describing AI as a major driver of productivity gains with the company aiming to keep headcount broadly flat while doing more with the same base.

The tension between short term decisions and long term impact is important. The market rewards efficiency and investors like discipline- but a business can reduce headcount faster than it can rebuild knowledge, and adopt tools faster than it can redesign processes.

The promise of AI is leverage. The danger is believing leverage removes the need for depth.

The End Cost Paradox

The landscape is changing so quickly we are still only just starting to see the fruits of decisions made in the last 12 months, if at all.

One example is a fast emerging financial paradox. For the past two years, the AI narrative has often been framed as cost-saving. Fewer people, faster work, lower marginal cost, more output. That logic is powerful, and in some cases it will be true, but it is not universally true.

AI is not ‘free’ labour, and comes with very real token costs. Axios recently reported that some companies are seeing AI-related spend exceed employee salary costs in certain areas, with token usage and compute costs becoming a material budget issue.

That does not undermine the AI thesis, but it does add nuance as implementations mature. The question becomes less “can AI replace this work?” and more “where does AI create a genuinely superior economic and operational outcome?”.

A company that cuts people first and works out the AI operating model later may find itself paying twice: once in lost organisational knowledge, and again in ever-more-expensive AI infrastructure that does not yet have the right data, workflows or human supervision around it.

The irony is that companies may remove the very people who understand the processes deeply enough to automate them well.

AI as an Amplifier

This is where the question of “outsourced thinking” becomes so important.

We’ve discussed before how AI is a great amplifier- and in particular that it not only amplifies intelligence- but the lack thereof. It amplifies good judgement, but it also amplifies poor judgement. Experience, but also inexperience. It amplifies wisdom but it also amplifies confusion, and a lack of specific knowledge.

AI makes it easier to produce- arguably more than at any time in history. But this does not automatically make it easier to know what is worth producing, how is best to produce, or even to know whether what is produced is any good!

This matters enormously for the future of work, because much of professional development has historically happened through the friction of doing. Junior employees learned by drafting the first version, sitting in meetings, getting feedback, making mistakes, seeing how decisions were made, and slowly developing pattern recognition.

If AI removes too much of that friction too early, the risk is not simply fewer entry-level jobs, but a thinner pipeline of judgement. Less wisdom. Less specific knowledge.

The Future of the Labour Market

Yuval Noah Harari writes about the possibility of an economically “useless class”, birthed by the separation of intelligence from consciousness that ever more sophisticated algorithms are heralding. His argument is deliberately provocative, but the underlying question is still practical: what happens when non-conscious systems can perform more and more tasks that previously gave people economic value?

The fact that new jobs may appear does not mean they will appear quickly enough, in the right places, or for the same people. Nor does it mean the first generation of AI-native companies will naturally invest in developing the people whose previous work has been automated away.

Some of that capability will remain human, and some will sit in AI systems. Some will sit in data infrastructure, workflow design, operational process and institutional knowledge. The challenge for leadership is to understand how those resources interact, rather than treating AI as a simple substitute for labour.

The distinction matters because not all work is the same. Some tasks can be automated with relatively little loss. Others carry judgement, context, accountability, relationship management, creativity or organisational memory. A company that fails to distinguish between those categories may become more efficient on paper while weakening the knowledge base that made it effective in the first place.

The risk is particularly acute at the junior end of the workforce. If early-career employees are no longer asked to do the first draft, the first analysis, the first reconciliation or the first piece of research, then companies need to think carefully about where learning happens instead. 

The companies that navigate this well will not be the ones that blindly lean most heavily on AI, or those that resist it out of caution. They will be the ones that understand what kind of resource each part of the business actually needs, and build a model where human capability and machine capability compound rather than quietly erode one another.

The Practical Path Forward

The companies that do this well will be the ones that connect AI adoption to the real structure of the business, starting with data and operations.

Before a company can successfully put Claude, ChatGPT, Gemini or any other model on top of its business, it needs to understand what the model is being placed on top of. Where is the data? Who owns it? Is it clean? Is it accessible? Are definitions consistent? Are workflows documented? Are permissions clear? Is there a reliable source of truth?

Without that foundation, AI becomes another layer of abstraction over an already messy system.

The next step is listening to employees.

In many businesses, the AI strategy already exists in isolated fragments. It is sitting in the prompts, hacks and unofficial workflows employees have built for themselves. Leaders should be asking where AI is already saving time, where people are nervous about using it, where duplication exists, and where teams are already solving the same problem in different ways.

That information is valuable- it reveals the real pain points, and it shows where adoption is likely to stick. It also helps avoid the common mistake of imposing a top-down AI strategy that sounds impressive but does not reflect how work actually happens.

Then comes training- not a single webinar, policy document or vague encouragement to “use AI more”. People need to understand what these tools are good at, where they fail, how to challenge outputs, how to protect sensitive data, how to build repeatable workflows, and how to use AI without slowly surrendering their own judgement.

Experimentation should be encouraged, but it should be channelled. The goal is not to stop the flywheel that is already turning, but to get everyone pushing it in the same direction.

The Real Competitive Advantage

AI is already changing the structure of work, and of thinking.

Some roles are disappearing, some are being redesigned. Some companies will move too slowly whilst others will move too quickly and discover that efficiency without knowledge is brittle.

The real competitive advantage will not come from simply using AI. Everyone will use AI. It will come from knowing where to apply it, how to govern it, how to train people around it, how to preserve judgement, and how to build an organisation where human expertise and machine capability reinforce each other.

The danger is not that AI thinks better than us, it is that we stop practising the kind of thinking that made us useful in the first place.