AI Strategy· 8 min read

AI Evolution Will Make Professionals More Wanted

AI is making it easier to build software. I believe its evolution will make genuine technical expertise more sought after and more expensive — and turn craftsmanship into a premium.

Oleksii Novodvorskyi

Managing Director · Nashville, TN

AI is making it easier to build software. I believe its evolution will make genuine technical expertise more sought after and more expensive.

When almost anyone can generate an application, a working application becomes less impressive on its own. The important questions move deeper: Does it solve the right problem? Can it evolve without falling apart? Does someone understand its behavior well enough to take responsibility for it?

Producing code is becoming more accessible. Developing that understanding still takes time.

The gap between creating and understanding

AI gives beginners an extraordinary opportunity to explore software development. An idea that once required months of learning can become a working prototype in a day or two. Experienced engineers can also use it to explore alternatives and accelerate implementation.

But the speed of creation can exceed the speed of understanding.

A developer can produce a system without knowing why its components were chosen, how they interact, or which assumptions hold it together. When something breaks, the response can become another prompt, followed by another patch. Eventually, nobody has a coherent explanation of the whole.

Addy Osmani explored this problem in his December 2024 essay, “The 70% problem: Hard truths about AI-assisted coding.” He described how rapid initial progress can give way to difficulty completing a dependable product. His percentages illustrate an observed pattern rather than a universal measurement.

That distinction matters commercially. A prototype demonstrates possibility. A product must survive reality.

Human authorship carries something valuable

Engineers who have spent years writing software develop more than familiarity with syntax. They build mental models of how systems behave. They learn how a convenient shortcut becomes a constraint, how a small change travels across components, and how an elegant solution can fail under an unexpected condition.

Writing code can be part of the thinking itself. An engineer discovers a flawed assumption while implementing an interface, recognizes unnecessary complexity while restructuring a module, or changes the design after seeing how awkwardly two concepts fit together.

That process creates an intimate understanding of the product.

It also leaves room for human judgment and taste: deciding what deserves attention, where simplicity matters, which inconvenience users will feel, and which technical compromise the business can afford.

A developer with that understanding can reshape a system deliberately. Someone who has accepted an implementation they cannot explain has less freedom, even if the first result arrived faster.

Simon Willison captured a related point in his October 2025 essay “Vibe engineering”: “AI tools amplify existing expertise.” He connected effective use of AI with established practices such as planning, testing, documentation, and review.

My argument goes a step further: experience gained through authoring software develops the judgment needed to decide when generated code deserves to become part of a product.

A prototype can finish its job without becoming the product

Consider a team that spends one or two days building an AI-assisted prototype. It lets stakeholders explore the workflow, challenge the idea, and discover what the future product should do.

That is valuable progress. The prototype has reduced uncertainty.

Once the concept is accepted, the team faces a separate decision: what implementation should support the actual product?

For a prototype built primarily for speed and discovery, the strongest answer may be to start again. The team carries forward what it learned about the product while designing the architecture, data model, security boundaries, and operational behavior deliberately.

Continuously patching the prototype can preserve assumptions that were useful during exploration but inappropriate for long-term use.

Some prototypes are designed to evolve, and rebuilding is not automatically the right choice. Professional judgment lies in recognizing whether the existing foundation deserves further investment.

Accepting the concept should never substitute for evaluating that foundation.

Better AI will increase the ambition

This argument cannot depend entirely on today’s AI limitations. Models will improve. They will handle more complexity and make fewer of the mistakes we currently spend time correcting.

Martin Fowler offered a useful perspective in his June 2025 article “LLMs bring new nature of abstraction.” He compared their impact with the transition from assembly to high-level programming languages, while highlighting the additional challenge of working with non-deterministic tools.

My expectation is that stronger tools will also expand what businesses attempt.

Ideas previously rejected as too expensive will become feasible. Companies will pursue more integrations, more customization, and more ambitious products. As the cost of implementation falls, expectations will rise.

That creates more opportunities for people who understand systems deeply enough to make difficult decisions about them.

The valuable professional will keep learning as the tools change. Their advantage will come from their ability to understand the problem, form a sound design, and recognize whether the implementation fulfills its purpose.

Scarce expertise will command a premium

Here is the economic consequence I expect: software production will expand faster than the supply of people capable of taking responsibility for complex software.

Code can be generated almost instantly. Technical judgment develops through experience, feedback, and confronting consequences. Companies cannot manufacture that experience by purchasing more AI subscriptions.

If demand for that expertise grows faster than its supply, competition for it will increase. So will its price.

Genuine technical seniority will become a scarce and expensive capability. Businesses will compete for professionals who can turn abundant implementation capacity into dependable products.

This also makes developing the next generation of engineers more important. Juniors need opportunities to reason, investigate, write, debug, and understand. Those experiences build the expertise the market will later need.

An industry that automates entry-level work without preserving a path to mastery risks making its future shortage worse.

The Rolls-Royce future of software

I expect this evolution to create a premium market for deliberately authored software.

The Rolls-Royce analogy fits: customers paying for exceptional craftsmanship, individual attention, and a product shaped around their needs.

“Handmade software” could become a meaningful expression of that value. Its appeal would come from knowing that skilled people thought through the details, understood the interactions, and shaped the product with intention.

AI may participate in that process. The premium would rest on the depth of human authorship: important decisions emerge from understanding, and the people responsible can explain and defend them.

Manually written code can still be poor. Craftsmanship must be visible in the quality of the result: coherent behavior, thoughtful design, maintainability, and care for the person using it.

As functional software becomes more abundant, those qualities will become stronger reasons to choose one product over another.

AI will continue to lower the cost of producing code. I believe it will also increase the value of professionals capable of creating something exceptional with it, around it, or by their own hand.

The easier software becomes to generate, the more valuable it becomes to know what is worth building and how to build it well.

References

Enjoyed this essay?

I read every reply and the best pieces start as reader questions.

Write to me