Everybody Can Be an Engineer and How AI Creates Expertise Flywheel

With AI, everybody can be an engineer. Everybody can be a marketer. Well, sort of. With AI Everybody can, everybody can... Ummm… well.. do better. Much better. 

AI is a lot of things. It is not a game-changer. It is a paradigm shifter. This article is all about how we can leverage subject-matter expertise in AI to improve everyone's ability to do better jobs, with higher quality. Indeed, our quality of life overall can be improved. Stick with me in this article. I'm going to start out a bit abstract and then boil down to a concrete example many of us can relate to. 

Specialization

Typically, job functions in the last several decades have been highly specialized and are heading toward even greater specialization. All the momentum continues along that trajectory. However, AI has started to alter the trajectory.

Specialization is important to have depth in subject areas. Subject matter experts have invaluable skills for producing results and solving complex, deep problems. Think of the problem domains of radiology, nuclear engineering, astrophysics, semiconductor manufacturing, computer engineering, etc. Deep subject expertise takes years, if not decades, of education and experience. 

AI is disrupting long-established trends. With the expressive power of large language models (LLMs) and specialized large reasoning models (LRMs), it is easier than ever to encode subject-matter expertise into these multimodal agentic AI systems. 

Distributed Knowledge, Centrally Codified and Curated 

Codifying subject-matter expertise into systems such as AI models provides many benefits. One of those benefits is that many people can learn from it. From a quality and safety perspective, that subject matter expertise can be reviewed, vetted, and independently verified. This process tends to increase trust and safety in systems. 

Domain Knowledge

Industries involving operational technology (OT), such as manufacturing, energy production and distribution, high-scale distributed systems, logistics management, and related areas, require deep domain expertise. Each field has its own level of uniqueness. However, at a meta level, there's commonality in process design and abstractions that can be applied cross domain. 

Transitivity, Translation, and Tessellation

Distributed knowledge representation enables moving from specific knowledge to one or two levels of abstraction up to meta or meta-meta levels. Here, commonalities can be identified. This is where a great synergistic effect happens. When knowledge is represented in a distributed form and then federated, AI practitioners can make even greater use of the knowledge. And indeed, they can turn this multidomain knowledge into wisdom by associating multiple pieces of knowledge across multiple contexts. 

This transitivity and translation of knowledge lets people learn in one area and apply those benefits to others. With AI and a solid foundation of distributed knowledge codified in these systems, this is not just a promise or a pipe dream. It is a true translation and coverage of knowledge paths through multiple areas of a business. This is the tessellation, the seamless tiling of knowledge across multiple domains. What is learned in one field can be easily translated into another with the aid of AI tools, leveraging codified knowledge. 

But What's the Point? 

This is all abstract. You may ask yourselves, "What's the point?" Simply put, this is about taking expertise from one person or one area and applying that everywhere within the enterprise. It's about raising everyone's expertise in the company. Not just subject matter experts.  

The point is that with the ability to translate knowledge between and across subject domains, practitioners in each field can learn from one another. Indeed, even those outside the specific areas can learn from the process. This is about making the knowledge portable and putting it to work. So great—what does this all mean to advancing technology in a certain area of the business? 

Bringing AI to Bear to Make Life Better 

This is the process:

And viewed as SySML diagram:

The trick and challenge are getting knowledge out of people's heads and into places where others can use it. Knowledge and wisdom have to flow to be useful. In software, engineers and practitioners can use tools like Claude Skills within their environment. These skills and code-specific ways of doing things, like code review, architecture, code maintenance, all of the necessary but sometimes mundane tasks. These skills can apply to other very specialized and deep technical areas. In all cases, the knowledge gets captured for others to use. What are Claude Skils? From Anthropic’s website:

Claude Skills are reusable, specialized instruction packages (folders containing Markdown, code, and resources) that teach the AI to perform specific, recurring, or complex tasks automatically. They enable Claude to adopt specific personas, follow brand guidelines, or use tools—such as creating Excel files, PowerPoint presentations, or Python scripts—without needing repeated prompting.

Vetted skills contribute quality, repeatability, and speed to the process. Once those skills are captured and stored, most of the challenge is solved. The next step is simply applying these skills to existing and new processes. Tool users build momentum and speed using these tools as an integral part of their daily activities. 

It's important to understand which skills AI tools require. Many of us have seen the popular LLMs from companies like Google, Anthropic, Perplexity, and OpenAI. These tools were amazing when they were released, suffered some backlash, and improved in a virtuous cycle. Truly, these are among the most sophisticated tools humans have built. Still, they are merely foundational. Skills are the tools used as powerful levers to create other, still more powerful tools that create still more tools. And this process repeats in a powerful loop. 

A Concrete Example

Let's put this in a less abstract form. 

Take the example of product management in software engineering. AI skills in agentic IDEs have afforded incredible capabilities in the last several months. Anyone with terminal access and the ability to encode their knowledge can share it with others and use it to build high-quality software. The tools alone do not bring quality. Like any physical tool, skilled, experienced craftspeople and practitioners must use AI tools. 

It's important to be careful here. I'm not suggesting jobs can be replaced. In fact, I'm not in favor of that. Rather, I'm suggesting that job quality can be improved. People can work on more important tasks and focus their expertise at higher levels, benefiting themselves and those around them. Much like automated manufacturing brought incredible increases in safety, especially in the last hundred years, the automation of tasks, even in software engineering, brings higher levels of productivity and safety both explicitly and implicitly. 

The Diagram Cleaner-Upper Application: A Tool Within a Tool Within a Tool

See the following drawing and chart. This is the input and output of my “Diagram Cleaner-Upper” app. It is intended to quickly turn a rough hand drawing into a clean diagram suitable for documents, presentations, and sharing with colleagues.  AI has cleaned up this rough, hand-drawn diagram on the left in seconds. In less than 2 minutes, I was able to hand-draw this picture. With an AI tool I created in less than 5 minutes, I used Google AI Studio to build an application that cleans up hand-drawn pictures and translates them into clean images using D2 language diagram code that can be further edited and refined. This is a tool I created on the fly in a tool chain. I created this tool today, the same day I'm writing this article. I use this tool in the tool chain, a larger tool to refine my ideas and author this document. When I think about how these tools interact with each other, I see the famous artwork by M.C. Escher, “Drawing Hands.” 

Here is the idea of this whole article in one simple picture:

And that’s the point. It is tools plus expert knowledge that are used to build incredibly powerful tools.  

I used AI to make a tool to take my hand-drawn efforts and rapidly transform it into a tool to talk about the process of using AI to make tools to build more tools. 

What I'm describing here is just a small example of a recursive self-improvement loop. I'm using advanced tools to build AI tools to help me think more clearly about writing articles about using AI tools to create AI tools to help me do my job better. And this article is all about encouraging people to adopt the process of encoding expertise into tools, making them even more powerful to build even more powerful tools. I know it's a bit of a brain twister. So stop. Reread this paragraph. 

Now let's jump back into the main flow. The diagram I've shown is a recursive pattern in which users of AI systems can directly interact with AI tooling encoded with subject-matter expertise. As a result, they can quickly produce vetted, gated, high-quality work. That intermediate work product can then be quickly reviewed by subject matter experts and stewards, both in a straight human-in-the-loop form and by having those stewards and subject matter experts bring the same or other AI tooling into the system for evaluation, vetting, and quality control of the intermediate work product. The following diagram shows the layers:

All this leads to the flywheel effect. It is great momentum to get ideas from people's heads into a form that can be quickly produced. And, this review and vetted product can become a release candidate. Let’s go back to the product manager and the software engineer. The process I'm describing helps alleviate their bottlenecks and reduce their work burden. Rather than reviewing a junior programmer's code and spending more time that diverts from the primary, high-value task, the AI tooling of the flywheel itself helps them springboard ahead faster.

A very interesting and powerful side effect is that even intermediate work built on domain expertise becomes an asset in a vast, growing library of knowledge and skills. The work product doesn't have to be immediately incorporated into a product to be valuable. In one way, it's much like building a vast library of knowledge and wisdom that can be used at any time by anyone to quickly form solutions. 

Under the hood, specialized agents analyze domain- and subdomain-specific factors, including security, scalability, robustness, code quality, architecture, and other facets of the software engineering life cycle. All of these agents are infused with and guided by the honed, careful, and valuable knowledge of the subject matter experts. It's just that it's happening in parallel while these highly skilled, knowledgeable, and valuable experts continue their work, hindered by even higher-value endeavors. 

And, in turn, the whole company ecosystem, both internally and externally, can benefit. Be clear. Pause. For a minute. This is not simple automation of toil. This is not simply automated quality checks and hybrid human-machine improved workflows. These are not incremental improvements. This is an exponential increase in creativity, quality, and productivity. This is rethinking the tradition. This is a radical transformation in how work gets done and in the potential quality and speed of deployment.

Next Steps

You may ask: “This is all great. But how do I get started?”. Like the old adage: How do you eat an elephant? One bite at a time”. 

We are standing at a unique place in history, on the shoulders of giants. We have some of the most well-equipped and capable tools we've ever had. The time is ripe to take advantage of this opportunity to improve our work, our lives, our customers, and our civilization. NOW!