The Rise of AI-Assisted Software Engineering

As they often say in movies, I have a good news and a bad news for you today. 

The good news (although not for everyone): as artificial intelligence develops, the cost of creating software code will fall sharply. Prototypes, integrations, tests, documentation, and many routine functions can be produced faster and with fewer people.

Now for the bad news - again, not for everyone: the cost of correctly deciding what to build, how to embed it in the business, and who will be accountable for the result will rise. This tension is likely to shape the software industry in the years ahead.

I do not intend to make any predictions; in today’s world, that’d be a particularly unrewarding exercise: change is moving so quickly that some conclusions become outdated before they are published. So what follows is not a forecast, but an attempt to consider the trends discussed by technology executives, investors, engineers, and project buyers.

From 2019 to 2026 the market has changed

AI affects software in the first place, but it also affects data centers, chip manufacturing, energy, robotics, logistics, and industrial capital

According to Gartner estimates, global IT spending was about $3.8 trillion in 2019, grew to approximately $4.7 trillion in 2023, and may reach roughly $6.3 trillion in 2026. Investment is growing rapidly not only in software, but also in data centers and AI infrastructure. In 2026, spending on IT infrastructure is estimated at close to $800 billion, while software spending is around $1.5 trillion. Gartner-related 2026 figures

The market has changed not only in size, but also in structure. In 2019, a software project usually began with questions about people, deadlines, budget, and technology stack. In 2026, those questions are joined by questions about computing capacity, data, inference costs, AI security, traceability, and the system’s impact on real business processes.

The good news for customers: code will become cheaper

AI can already accelerate many stages of development:

•           generating routine code;

•           writing unit tests;

•           producing documentation;

•           refactoring;

•           finding errors;

•           building prototypes;

•           migrating standard components;

•           preparing integrations;

•           analyzing logs and technical incidents.

One study of GitHub Copilot estimated time savings of 30–50% on repetitive tasks. Research on GitHub Copilot

This does not mean that an entire software project automatically becomes 30–50% cheaper. Faster coding may simply move the bottleneck to another part of the process.

In a 2026 study, more than 70% of developers reported significant time savings on boilerplate code and documentation. The benefits were considerably smaller in early stages such as requirements analysis and planning. Study of GenAI in software development

The paradox is clear:

The cheaper it becomes to write code, the more expensive mistakes in defining the task may become.

The bad news for customers: good system design will become more valuable

Between a business idea and AI-generated code, organizations will need a more carefully designed layer of work:

•           analyzing the business problem;

•           defining goals and measurable outcomes;

•           describing user and operational scenarios;

•           designing the architecture;

•           preparing data;

•           selecting models;

•           describing integrations;

•           defining security constraints;

•           planning quality control;

•           allocating responsibility between people and AI.

This is why companies that offer assessment, planning, and architecture alongside development may be in a strong position, at least in the near term.

There is an interesting historical analogy. In industrial culture, complex projects first went to a design bureau. Engineers developed drawings, calculated loads, selected materials, and worked out the construction. Only then were the plans passed to builders or production teams.

Perhaps software is moving toward a similar model: first, a design bureau for the digital system; then, AI-assisted production.

1. Buyers will purchase outcomes rather than code

Clients will increasingly care about shorter operating cycles, lower service costs, fewer errors, faster product launches, security, and predictable operating expenses. The market will move from “we need a team of ten developers” toward “we need to create and deploy a capability that will produce a specific business effect within a defined period.”

2. Outsourcing will not disappear but it will change

The greatest pressure will be felt by companies selling inexpensive hours for routine tasks: simple coding, basic integrations, standard testing, and documentation. Complex outsourcing, however, will remain valuable. Demand will continue for AI-native engineering, deep domain expertise, legacy modernization, secure work with corporate data, industrial AI, and accountability for part of the final outcome. The winners will be the companies that can combine people, AI tools, processes, architecture, and responsibility.

3. Smaller teams will create much more

One specialist may manage several AI agents, while a small team produces output that once required dozens of employees. This does not mean that people will automatically disappear: context, conflicting requirements, architecture decisions, security accountability, and organizational change are difficult to delegate completely to AI.

4. Requirements quality will become a strategic asset

AI works well with a well-defined task. If requirements are incomplete or disconnected from how the company actually works, AI will simply produce a useless result faster. The value of business analysis, acceptance criteria, architectural constraints, exception handling, data quality, and measurable success criteria will therefore increase. The METR study found that experienced developers working in familiar complex codebases actually completed tasks more slowly when using AI tools. METR study reported by Reuters

5. Software products will become more polarized

Simple products will be easier to create and copy. Basic chatbots, elementary workflow applications, and services without unique data, integrations, or market trust will come under pressure. Products embedded in critical processes will become more valuable: cybersecurity, industrial platforms, healthcare systems, financial infrastructure, logistics software, developer tools, and solutions for regulated industries. The product’s protection will be more than code; data, integrations, domain knowledge, trust, and customer history will matter just as much.

6. Physical infrastructure will become part of the software economy

AI is no longer only a digital phenomenon. It requires electricity, data centers, chips, cooling, networking equipment, batteries, manufacturing, engineering talent, and capital. Projects involving digital twins, IoT, predictive maintenance, robotics, computer vision, connected products, edge computing, and factory automation will therefore expand.

7. Energy will become a new cost factor

According to the IEA, global data-center electricity consumption may more than double by 2030. The impact will be even stronger in regions where large AI data centers are concentrated. IEA energy and AI reporting Buyers will calculate both, development cost and the ongoing cost of operating a system: the price of one request, GPU capacity, inference location, and whether a smaller model can be used. The new engineering question will be: How many computing resources must be spent to produce one unit of useful business value?

8. Infrastructure industries will generate the strongest demand

The strongest need for software modernization is likely to appear in energy and utilities, manufacturing, logistics, semiconductors, data centers, financial services, insurance, healthcare, telecommunications, and companies producing connected products. Manufacturers of physical goods will compete not only for the best product quality, but also for digital services, operational data, remote monitoring, personalization, and predictive maintenance.

9. Not every industry will be able to afford AI modernization

If major AI companies actively compete for electricity, specialists, construction capacity, and investment capital, lower-margin companies may face higher costs, limited access to power, and more expensive financing. The statement “AI grows while everything else declines” is not true, and the future will show this: AI may help other sectors reduce costs through better transport, energy-grid, manufacturing, and maintenance optimization. Today, the key question is: who will receive the productivity gains, infrastructure owners or the wider group of companies using AI? No doubt, we’ll soon have the answer to this question, as well.

10. Control will become more important than speed

Speed without control can become a new form of risk. A GitLab study found that developers are writing code faster, while code review and validation are becoming new bottlenecks. Companies also struggle to determine which code was created by people, which by AI, and who should be accountable when something goes wrong. GitLab findings Spending will therefore increase on code review, testing, security, observability, traceability, AI governance, and model audits.

So, what will happen to the software industry?

I would not say that AI is going to destroy the software development market. It will more likely divide it into two levels. At the lower level, code will become abundant, accessible, and relatively inexpensive. Competition will intensify, prices will fall, and some routine roles will disappear.

At the upper level, the value of those who can define the right problem, design the system, prepare the data, embed technology in the organization, ensure security, and connect the project with measurable business value will increase. That is why my good-and-bad-news summary is: AI will sharply reduce the cost of creating software code. At the same time, we’ll witness growing cost of making the right decision about which code should be created in the first place.

The future of software may belong to those who design the systems that people, AI, or a combination of both can build successfully. The key competitive advantage will be the ability to answer three simple questions:

1.         What real problem are we solving?

2.         How should the system work inside the business?

3.         Who is accountable for its outcome?

AI cannot yet answer these questions reliably on its own. Human expertise in software engineering will therefore not disappear. But it will become narrower, more valuable, and far more accountable.