Companies have already bought AI tools. Why is there no breakthrough?

““We bought the tools, employees use artificial intelligence (AI) every day, and there is no breakthrough. Maybe we should measure in more detail?” I have heard these sentences in more than one conversation with managers over the past six months.

 

But they were not first uttered in 2026. An almost identical complaint was repeated by factory owners about 130 years ago – having just installed the most revolutionary technology of its time, the electric motor, and seeing that productivity… had hardly budged.

 

Before electricity became established in industry, the heart of a typical factory was a single steam engine: it turned a central shaft installed throughout the building, and a labyrinth of leather belts led from it to each machine tool. The machines were not lined up according to the work flow, but the closer they were to the shaft, the more half of the energy was lost in the friction of the belts, and if one belt broke, the entire factory would come to a standstill.

 

Then came the electric motor—and the owners attached it… to the same central shaft. The so-called group drive: a new technology, assembled according to an old design. Economic historian Warren Devine, who has studied this period in detail, summarizes it mercilessly: there was practically no productivity jump. In 1900, electricity accounted for only about 10% of the power of US factories.

 

It was not a more powerful motor that unlocked value, but a different way of thinking: what if each machine had its own motor? When machines no longer had to line up under a shaft, the factory could be arranged according to the workflow—and thus the conveyor belt was born. By 1930, electricity already accounted for about 80% of the power of factories. Electricity began to create value not when factories changed their energy source, but when they redesigned the factory itself to the new nature of energy. It took forty years.

 

Now let’s look at how most companies are “using AI” today.

 

The employee sits in his usual environment and talks to the AI ​​as if it were a smarter assistant. In software teams, engineers try to squeeze agents into the existing process - hoping that they will work the way people would work. Each step is accelerated by a dozen percent, but the “factory” itself - processes, team composition, work organization - has not changed. It is a group drive, only the 2026 version.

 

And then the most dangerous part happens: the employee, not seeing much difference, returns to old habits, and the manager, not seeing results in the indicators, concludes - “AI is overrated, let's wait.” Both draw a logical conclusion from their experience. And both are wrong, because they tested not a new work design, but a big engine on old shafts.

 

The author of the comment leads a company of 200+ software engineers, developing products for the Scandinavian and Dutch markets, is a board member of two companies and the author of the training for managers “Code is cheap. Leadership isn't.” He is currently conducting a survey of Lithuanian software managers on the topic of agent engineering, and will present the results on November 18 at the “Verslo žinios” conference “Business 2027”.

 

There is also the opposite side of the trap – managers who saw the breakthrough and rushed to it. In May 2026, Meta laid off about 10% of its employees, transferring some of them to AI teams, and the next month, at an internal meeting, Zuckerberg admitted: the use of agents in programming did not yield the expected results. If an organization that plans to invest up to 145 billion USD in AI infrastructure this year miscalculated, then anyone can. Ford turned quality control over to AI systems, laid off experienced engineers—and then hired back 350 veterans. Klarna, which replaced about 700 customer service workers with AI, admitted it went too far and started hiring people again.

 

These cases are no exception: 55% of managers who laid off employees because of AI now admit they were wrong. The numbers explain why. A 2026 study by Faros AI, which analyzed two years of work data from 22,000 engineers, records a real leap in productivity: work completed per engineer increased by 66%. But the same study also shows the cost: errors increased by 54%, and the ratio of incidents to changes more than tripled. Velocity was measured. Maturity was not.

 

This is what connects all costly mistakes: decisions about people were made based on the first indicators of speed in a discipline in which the organizations themselves were still new. This discipline has a name – agentic engineering. Its essence: a person no longer writes code and performs tasks himself, but orchestrates AI agents – defines the task, designs quality checks and is responsible for the final result.

 

This is not “more AI inside the same process” – this is a different process, redesigned around agents in the same way that an individual drive redesigned a single-engine factory. I observe this change closely, leading a company of 200+ engineers: within a few months, software development was rewritten, and tasks that we entrusted to a person just last year can now be performed by agents from start to finish.

What should an IT company or department manager do? Three practices.

 

First – constantly ask the question and think: if this product or process we were to create today from scratch using agents, would our current problem even exist? It frees us from the boundaries of old processes and changes the thinking model.

 

Second, at least one real experiment: not “10% of time for everyone”, but a real work process redesigned from start to finish using agents. There is a lot to learn from this.

 

Third, measure the team’s maturity in agent engineering, not activity: not how many tools have been purchased and how much they are used, but how much responsibility can be entrusted to agents while maintaining high quality.

 

This period is special for Lithuania. Working in this area, I see how far behind we are from what is happening in Silicon Valley. Factory owners had four decades, the cloud computing revolution lasted about 25 years, and the agent era is moving even faster, and the gap between companies is now accumulating not in years, but in months. The electric revolution did not win those who bought the engine first. The winners were those who redesigned the factory first.

 

Comment author – Agnius Paradnikas, CEO of UAB “Visma Tech.”