Last year, Avital Balwit, Chief of Staff to the CEO at Anthropic - the company behind Claude, ChatGPT’s main competitor - published a thought-provoking article contemplating the rapid rise of AI and its implications. She suggested her career, as she knows it, might have just five years left. While some dismissed this as alarmist, others shared her concerns. For instance, Klarna’s CEO, Sebastian Siemiatkowski, announced a permanent hiring freeze aimed at reducing the workforce through natural attrition, saying: “AI can already do ALL of the jobs that we as humans do. It’s just a question of how we apply it and use it.” Similarly, Salesforce revealed plans to cease hiring software engineers in 2025, citing similar reasons.

While I personally think 5 years is an overly bold prediction, I do agree with the broader perspective that AI will eventually be capable of performing every economically valuable task. Most of us, however, remain unaware - or perhaps in denial - about just how rapidly and profoundly this will disrupt the world of work.

In this article, I explore why we should - at the very least - entertain the possibility that AI could soon replace humans in the majority of economically significant roles, starting with knowledge work. I outline the different levels at which AI will be deployed, examine the roles most resilient to this shift, and highlight the essential skills we must all develop to stay relevant in this rapidly evolving landscape.

Why Superintelligence Will Change the Game

Over the past decade, artificial intelligence (AI) has replicated an astonishing range of human-like capabilities - from creating artwork and translating languages to diagnosing diseases and writing code. With each new generation, AI continues to close the gap on tasks once deemed exclusively human. Many are predicting we could have Artificial General Intelligence (AGI) - an AI matching human intelligence across a vast array of tasks - as early as 2025.

However, AGI is merely a stepping stone. The true paradigm shift comes with Artificial Superintelligence (ASI) - AI that meaningfully surpasses human intellect across virtually all domains. Sam Altman, CEO of OpenAI, predicts that this threshold might only be a few thousand days away. Even if it is not that soon, the “Situational Awareness” essay by Leopold Aschenbrenner, an OpenAI employee, (which I would recommend as essential reading) makes a convincing case that AI is in its infancy and we're on the brink of a significant turning point - where AI can automate AI research - causing its development to speed up dramatically.

Which Types Of Roles Will Be Most Resistant To AI Disruption?

So what happens when companies can deploy thousands of AIs, each with an IQ of 500 (more than 3 times Einstein’s IQ)? If machines start performing cognitive work faster, cheaper, more accurately, and potentially more creatively than their human counterparts, which roles will be left for us?

No role is entirely immune to the transformative effects of ASI, but some will take longer to adapt, and a few will prove highly resistant.

Factors Making Certain Types of Work More Resistant to AI Disruption

1. Distance Between Creators and End-Users

Jobs where workers possess both domain expertise and the ability to develop AI tools - such as software engineering - are more likely to face rapid disruption. Conversely, fields where most subject matter experts lack the skills to create custom AI solutions will likely experience slower automation.

2. Specialised Fields with Few Experts and Limited Data

Highly specialised fields will be harder to automate when there is limited training data, few subject matter experts, and the lack of closely analogous fields for product builders to leverage.

3. Extended Feedback Loops

Tasks requiring extended periods to measure outcomes - such as infrastructure planning - will be slower to automate. Long feedback loops delay AI's ability to refine and improve its performance, hindering the pace at which high accuracy can be achieved.

4. Non-Deterministic Outputs

Recent AI advancements have excelled in tasks where success can be easily measured - such as solving mathematical problems or writing code - because these tasks produce deterministic outputs that are straightforward to evaluate and refine automatically. This clear feedback loop enables rapid improvement.

However, tasks requiring non-deterministic outputs - where outcomes are open to interpretation or shaped by subjective human preferences - present a much greater challenge for AI. Producing high-quality non-deterministic outputs typically involves two distinct steps:

• Divergent Thinking: Generating a broad range of creative and unexpected possibilities.

• Convergent Thinking: Selecting the best or most appropriate option from the generated pool, a step that relies on nuanced human judgment and deep contextual understanding.

While AI often excels at divergent thinking, it struggles significantly with convergent thinking. Consequently, it tends to perform poorly in tasks requiring a decisive, one-shot non-deterministic solution - such as negotiations, therapy, or aesthetic decision-making. In such cases, AI frequently produces generic or broadly acceptable outputs rather than truly optimal ones, highlighting its limitations in handling complex, subjective scenarios.

5. Complex Attribution of Success or Failure

In high-level tasks, such as strategic decision-making, outcomes are often intangible, long-term, and influenced by multiple factors, including luck and execution quality. The potential impact of poor decisions is theoretically immense, leading organisations to invest time and resources into ineffective initiatives repeatedly. While it is well-established that businesses frequently make similar critical errors, the return on investment of AI-based decision support systems remains unclear and almost impossible to prove. This may deter budget holders from investing, resulting in slower adoption.

6. High Stakes and Stringent Regulations

Professions operating in high-stakes and regulated environments - such as healthcare, autonomous vehicles, and criminal justice - are subject to intense scrutiny and legal constraints. Even when AI can demonstrate superior safety or accuracy compared to humans, public mistrust and regulatory barriers will likely require continued human oversight.

7. Manual Labour (Especially Intricate Tasks)

The disruption of manual labour depends on advancements in robotics, which face unique challenges compared to knowledge-based AI. High prototyping costs, limited automation in research, and lengthy feedback loops slow progress significantly. Tasks requiring precise, complex movements - such as plumbing, electrical work, and hair styling - will be especially challenging to automate and, as a result, will take much longer to face disruption.

8. Human Touch Valued

Roles that depend on genuine human connection - such as social workers, therapists, and frontline hospitality staff - are going to be particularly resistant to AI replacement.

It seems like this factor is going to be the most resilient of all. Even as AI advances in mimicking human nuances, many consumers will likely prefer authentic human engagement, making this a sustainable and meaningful differentiator.

Levels of Workplace AI Adoption

As AI grows increasingly sophisticated, organisations will delegate a broader range of tasks to these systems. As they do this, they will be able to do more work, better and with fewer employees.

To better understand this trend, I’ve drafted a seven-tier framework that outlines how AI adoption is likely to evolve in the workplace. Currently, most companies operate at the foundational Levels 0 and 1. Looking ahead, Levels 4 and 5 reflect the much-predicted emergence of Unicorns (companies valued at over $1 billion) with 10 or fewer employees .

The Seven Levels of Autonomous Organisations

(Naming inspired by the Six Levels of Autonomous Driving)

Level 0: No AI

No AI tools used. Humans perform all tasks unassisted.

Level 1: AI Assists Humans with their Tasks

AI supports humans to do their current tasks more efficiently and effectively. Humans review and approve every AI suggestion, maintaining full accountability. e.g. coding copilots.

Level 2: AI Executes Tasks with Human Oversight

AI autonomously carries out tasks but requires human approval before finalising outputs or performing consequential actions, ensuring human accountability. e.g. generating automated reports or medical diagnoses.

Level 3: AI Autonomously Executes Tasks

AI executes tasks without needing human approval for consequential actions, so long as it operates within predefined parameters. e.g. customer service chatbots. While AI may occasionally make mistakes that humans typically would not, these errors are deemed acceptable due to the AI's overall performance and cost savings.

Level 4: AI Autonomously Executes Projects or Entire Roles

AI autonomously manages entire projects or fulfills specific roles (e.g. self-driving taxis), handling execution independently within predefined parameters. At this level, AI begins taking on higher-level tasks such as proposing goals and execution plans. It may also delegate specific tasks to humans where it deems human input more effective or appropriate. Humans are required to sign off on key plans, monitor performance, and step in when significant issues arise.

Level 5: AI Autonomously Manages Business Functions

AI oversees entire business functions (e.g. Digital Marketing). At this level, AI formulates strategies and proposes objectives while delegating tasks to humans when necessary. Human sign-off remains essential for these plans and any major decisions that arise, ensuring strategic alignment.

Level 6: AI Autonomously Manages Organisations

AI runs the organisation by making high-level strategic decisions and managing all operational aspects under human board-level governance, ensuring alignment with the organisation's mission.

Who Will Excel in the Age of SuperIntelligent AI?

As the ROI becomes hard to ignore, an increasing number of companies will employ higher levels of AI delegation for more of their roles / functions. As they do this, humans employees will find they spend more of their time doing the following three things:

1. Managing AIs (to ensure they are aligned with and effectively executing their goals);

2. Collaborating with other humans;

3. Providing a human-touch to end-customers.

A generic piece of advice people are throwing about at the moment is the idea that - in order to thrive in the workplace of the future - we all need to learn how to use AI effectively. But what does this actually mean?

I’d suggest there are five things we will need to be good at.

1. Knowing When to Use AI

Having an intuition to know when to use AI and when to do a task alone. And also knowing which level of delegation / collaboration would be most appropriate.

2. Choosing the Right AI

Deciding on the most suitable AI tool - or combination of tools - requires careful consideration. This includes evaluating whether to create a custom solution tailored to specific needs or to adopt an off-the-shelf product.

3. Understanding How to Get the Best Outcomes from AI

Getting the best results from AI involves clearly defining desired outcomes - whether by setting unambiguous goals, providing examples, using fine-tuning, developing structured workflows, crafting detailed prompts or something else. It also requires supplying the right resources, such as accurate data and relevant context.

4. Interpreting the Outputs of AI

Applying critical thinking and decision-making skills to assess & action AI outputs. This skill is particularly important when interpreting the output of AI owned or managed by someone else.

5. Adaptability

Remaining flexible is essential in an era where it’s difficult to predict how AI tools and capabilities will evolve. Success will depend not only on staying open to change but also on the ability to identify and prioritise the right skills and knowledge to learn as new advancements emerge.

Parting thoughts

For hundreds of thousands of years, humans have been the most intelligent species on Earth. The rise of superintelligent AI will mark a paradigm shift unlike anything we've encountered before, and its true impact on the workforce is easy to underestimate.

If AI surpasses humans in speed, cost, and accuracy, will it lead to widespread unemployment, or will it fuel the rise of countless smaller, highly efficient companies? The answer is likely a combination of both. Lower barriers to entry will foster the growth of micro-enterprises, but many workers may struggle to adapt, resulting in higher unemployment rates.

To succeed in this new era, focus on becoming the kind of person who can thrive in a micro-enterprise powered by AI. Master the ability to select the right AI tools, align them with your goals, and set them up for success. And develop the skills to critically evaluate AI outputs, collaborate with others, and provide a human touch that AI cannot authentically deliver.

For centuries, economists assumed humans were rational actors – making logical choices to maximise benefits while minimising costs. However, in the latter half of the 20th century, as real-world behaviour was scrutinised more closely, it became clear that this rational model didn't capture the complexities of human nature. Behavioural economics emerged to challenge these assumptions, revealing the psychological biases and flawed thinking that often shape our actions.

While many businesses acknowledge that their customers don’t always act rationally, they often fail to apply the same critical thinking to their own decision making.

Why We're All Flawed

The reality is that we, as humans, are not as adept at making decisions as we like to believe. And this is especially true for those of us designing and developing new tech products. We operate in environments filled with high uncertainty – something we are biologically ill-equipped to handle. This is because our brains are wired to favour quick decisions over deliberate, well-considered ones.

Instead of thoughtful analysis, we often rely on intuition and mental shortcuts – heuristics – which can lead to flawed conclusions. We tend to be overconfident, overly influenced by authority figures, and focus on the information that is readily available rather than seeking out contradictory evidence. We also prioritise short-term gains over long-term benefits and frequently misidentify the true problem that needs solving.

How People Try to Solve This

These human flaws are well-known and universal. Yet, in my 15 years in product management, I've seen few companies take more than superficial steps to address them.

It is true that structured decision-making frameworks can help guide people away from common pitfalls, but behaviour change is hard – so when they are adopted, they are often used inconsistently and quickly forgotten.

And, while many teams understand that things like customer feedback, data analysis, and experimentation can reduce subjectivity, they often lack the confidence or skills to plan and implement these strategies effectively.

How AI Will Help

From the extensive R&D I’m doing in my current project, I have a high degree of confidence that artificial intelligence will play a key role in improving decision-making in the future. Here’s a summary of my mental model, specifically tailored to tech product decisions:

1. DELEGATE TASKS TO AI

AI will take on many time-consuming or specialised tasks:

• Autonomous Research: AI will independently plan and execute research tasks, sourcing data, analysing it, and delivering actionable insights in response to user questions.

• Pattern Discovery: It will mine data to uncover hidden trends, correlations, and outliers that might otherwise go unnoticed, helping teams identify opportunities and risks.

• Data Summarisation: AI will condense large volumes of structured and unstructured data – such as metrics and customer feedback – into clear, actionable reports.

• Metric Monitoring: It will track key performance indicators in real time, alerting stakeholders when anomalies are detected or when unexpected opportunities arise.

• Personalised Dashboards: AI will generate dashboards and visualisations tailored to individual users, highlighting the most relevant metrics and insights.

• Automated Documentation: It will document decisions automatically, capturing key details such as rationale, trade-offs, and outcomes in a clear and accessible format.

• Stakeholder Support: AI will handle routine stakeholder questions and challenge superficial objections.

• Outcome Tracking: It will monitor the results of decisions over time, linking outcomes back to the original rationale and providing feedback loops for continuous improvement.

2. AI AS A THOUGHT PARTNER

It will work collaboratively to enhance decision quality in the moment:

• Problem Reframing: AI will challenge team members whether they asking the right questions, proposing alternative ways to frame problems and uncover opportunities.

• Context Clarification: It will answer complex or unclear questions, providing deeper understanding and context to help teams move forward with confidence.

• Ideation Support: AI will facilitate divergent thinking by suggesting alternative solutions, generating left-field ideas and asking thought-provoking questions.

• Subjectivity Flagging: It will highlight when decisions rely too heavily on subjective judgement and suggest ways to incorporate more data-driven approaches.

• Analysis Paralysis Detection: AI will identify when teams are overanalysing, highlighting where the cost of further analysis outweighs the benefits.

• Bias Awareness: It will identify cognitive biases – such as confirmation bias or availability bias – that could distort decision-making and highlight their impact.

• Scenario Exploration: AI will model different scenarios, showing how decisions might play out under varying assumptions, helping teams anticipate potential outcomes.

• Trade-Off Evaluation: It will help teams weigh trade-offs systematically, considering both short- and long-term impacts.

3. AI AS A MENTOR

And it will provide high-level direction, opinionated feedback, and support ongoing growth:

• Decision Frameworks: AI will recommend and walk users through decision-making frameworks tailored to the context.

• Uncertainty Navigation: It will suggest strategies for approaching ambiguous situations, helping teams build confidence when dealing with high uncertainty.

• Strategic Alignment: AI will challenge decisions and actions that are misaligned with the overall strategy, ensuring any deviation is well-justified and supports organisational goals.

• Stakeholder Management Coaching: It will coach team members on managing challenging stakeholder interactions, including addressing objections and aligning differing priorities.

• Personal Development Mentoring: AI will support learning from past decisions by encouraging reflection on the decision process, execution and outcomes.

Conclusion

Human brains are inherently ill-equipped for making decisions in environments filled with uncertainty. Leveraging AI to bridge this gap will provide a competitive advantage for those who adopt it effectively.

🤝 If you’re a Product leader interested in using AI to empower your teams to make better decisions, let’s connect.

Recommended Reading

• The Four Villains of Decision Making by Shane Parrish

• Leaders as Decision Architects by John Beshears and Francesca Gino

• Thought Partner by Gregory Shove