The jobs data is out, and the American private industry lost another 32,000 jobs in the month of November. In this context, the topic of AI and its impact on the labor market inevitably comes up.
In the opinion of many, AI is displacing humans in high-paying knowledge intensive occupations – software development, sales engineering, financial analysis, and many others that were considered automation-proof.
First, we lost to computers in chess. Now we are losing in Python, JavaScript, Photoshop, Blender, and Excel.
Yet, there is another group that believes in Jevons paradox. AI needs access to vast amounts of data, energy, compute, and deep learning expertise. As the use of AI increases, it will give rise to new jobs in data management, nuclear and clean energy, supercomputing, and professions that we can’t even imagine, like space-based data center management.
Eventually, the net effect will be positive. AI will destroy jobs in some occupations but will create many more jobs in others.
The keyword here is “eventually.” There are bills to be paid here and now. So what does one do while waiting for the “eventually” to transpire.
The answer to the question, not surprisingly, is “follow the money” or, more precisely, “follow the value”.
Knowledge work is not monolithic. It can be broken down into activities. These activities can usually be abstracted as
- Intake: Understand the requirements
- Research: Perform deep research over any number of data sources
- Synthesis: Synthesize research findings into a work product
- Review: Test the product to ensure it satisfies the requirements
- Ship: Deploy the work product
This flow maps to many different occupations.
Software Development:
- Intake: Interview the client to understand the requirements
- Research: Research existing solutions in closed and open source
- Synthesis: Design and implement a new solution
- Review: Test the implementation: unit testing, integration testing, etc.
- Ship: Deploy the solution
Investment Analysis:
- Intake: Receive a confidential information memorandum
- Research: Perform research and due diligence: commercial, legal, financial, operational, etc.
- Synthesis: Synthesize an investment committee memo
- Review: Deliver the memo in front of the investment committee
- Ship: Complete the financial transaction
You get the idea. Almost every knowledge-intensive workflow can be abstracted as a sequence of Intake > Research > Synthesize > Review > Ship.
It’s plain to see now why vibe coding has become Gen AI’s number one killer app. If a human can clearly express the requirements, AI can easily automate all steps in the developer workflow: understand the requirements, research existing code bases, synthesize new code, and validate the solution. If the solution does not pass validation, AI can repeat this cycle thousands of times until it does.
Most vibe coders now just enter prompts and lean back in their chairs watching AI do its magic. After a few false starts and iterations, they get an app deployed in the app store.
Similarly, there are many solutions now available to automate the work of an investment analyst. Anthropic has been recruiting quant and finance professionals and positioning Claude and Excel integrations toward Wall Street use cases. With some of these solutions, one can enter the name of company and, a few moments later, get a complete draft of an investment committee memo.
In summary, the research and synthesis steps of knowledge work are increasingly commoditized in many occupations. The costs and value of reading and summarizing financial reports, looking up snippets of code, writing new code, and composing investment committee memos are quickly approaching zero.
Consequently, the relative value of other steps is increasing:
- Prompt engineering – the art of communicating requirements to AI models
- Context engineering – making clean and relevant data available to the models
- Validation and testing – ensuring that AI output satisfies the requirements
In essence, the purely human skills of connecting with customers, understanding their needs and desires, and translating them into AI model prompts are growing in importance. Critical thinking, the art of estimating the value and validating correctness of AI-generated artifacts, becomes crucial.
But is this enough?
If prompt engineering, context engineering, and testing were sufficient, then financial markets, for instance, would not be able to function.
Financial markets are zero sum games – for every seller there must be a buyer. They are also highly efficient: the same amount of information is available to all players. In this context, AI models, representing the compressed form of the entire human knowledge, must be generating substantially similar buy/sell recommendations, preventing any kind of transactions.
From experience, we know that this is not the case. Markets remain fully functional despite extensive automation and quant trading.
So, what are we missing? There must something else that is extremely valuable.
The answer to that question is methodology. Unique methodology makes every final work product unique despite automation and commodity components, like raw data and AI models.
In the case of financial management, unique methodology, such as all-weather diversification, can be used to construct unique portfolios and financial products. An AI is merely an execution mechanism turning this methodology into financial products delivered to customers.
Methodology becomes the primary market and skill differentiation mechanism in the age of AI and knowledge work automation. Unique methodology can be applied at every step in the knowledge work process, but it is particularly effective during the research and synthesis phase.
Methodology is best implemented through human-guided task decomposition and orchestration.
There are many ways to execute the research step, for instance. In the classical deep research scenario, an AI agent analyzes the requirements and develops a plan for researching relevant topics using the data sources available. It executes the plan by searching the web, reading documents, and communicating with business systems. Every time we run the research step, the AI agent may generate and execute a different plan.
We can control the research process by breaking it down into subtasks and providing task-specific prompts, data sources, and tools. For instance, one could use the PESTEL method to research a company’s macroeconomic environment and craft a separate prompt for each element of the PESTEL analysis. In this case, PESTEL is the methodology that makes the research process unique and distinct.
PESTEL is a well-established methodology, and instead of defining it for the AI model, we could just instruct the model to use this methodology for research. To do this, however, the human driving the AI model must (a) know of the PESTEL methodology (b) know how it compares to other methodologies.
Established frameworks are good, but a proprietary methodology could be a true competitive moat. A company could develop such a framework by codifying the tacit knowledge and experiences accumulated in multiple customer engagements. There are many examples in the management consulting domain: McKinsey 7-S Framework, the BCG Growth-Share matrix, and many others.
Consider the Talent-to-Value framework developed by CEO Works. Rather than starting with organizational hierarchy, it identifies the 5% of roles that deliver 95% of value, then systematically matches talent to these positions. An AI can execute the analysis – gathering performance data, mapping dependencies, running simulations – but the framework itself, refined through thousands of engagements, is what makes the output strategically valuable.
One might argue that established methodologies like PESTEL or Porter's Five Forces will themselves become commoditized as AI models absorb them through training. This is precisely why proprietary methodologies – developed through unique client engagements, refined through continuous feedback loops, and closely guarded – become increasingly valuable. They represent the compressed wisdom of specific organizational environments and lived experiences that AI cannot easily replicate from public training data.
In summary, in the age of AI-driven knowledge work, competitive advantage concentrates in three domains:
- Foundational AI Skills – Prompt engineering, context engineering, and critical validation of AI outputs
- Methodological Expertise – Deep knowledge of established frameworks (PESTEL, Porter's Five Forces, etc.) and when to apply them
- Proprietary IP Development – The ability to codify tacit knowledge into unique methodologies and orchestrate AI-powered workflows that execute them
As another 32,000 jobs disappear from the monthly report, the question isn't whether AI will transform knowledge work. It already has. The question is where you will position yourself in the new value system. Those who develop, refine, and orchestrate proprietary methodologies won't just survive the transition, they will capture disproportionate value in the AI-enabled economy. The tools are commoditizing. The thinking remains invaluable.