Tech companies turn AI use into a race: programmers are compared by activity

In technology companies, the use of AI for development is ceasing to be an optional habit and is becoming part of internal culture. In some teams, this has gone as far as rankings where employees are compared by their activity in AI tools, and model bills are rising along with it.

How AI entered the workflow

Until recently, generative tools in development were seen as a convenient aid for isolated tasks: finish a function, explain an error, quickly assemble a draft test.

Now the approach is changing. For many teams, AI is becoming a constant layer over everyday work: programmers turn to it when writing code, refactoring, finding bugs, and preparing documentation.

When this practice becomes widespread, managers start looking not only at the final result but also at the very intensity of using these systems.

Against this backdrop, some companies have introduced internal tables showing who works with AI and how actively. In essence, this is no longer about a personal set of habits, but about a new productivity metric.

If teams used to discuss development speed, the number of releases, or closed tasks, now another indicator may appear alongside them — the volume of prompts to models, session frequency, or the share of tasks completed with AI assistance. The very appearance of such rankings points to a shift: AI is beginning to be treated almost like a work resource.

Why this has become a race

When a leaderboard appears inside a company, any tool quickly stops being neutral. It starts to influence team behavior.

Employees see that using AI is not just allowed, but visible and possibly encouraged. In that environment, a natural race emerges: who will master new scenarios faster, who turns to the model more often, and who shows greater engagement.

For some developers, this is a way to move faster, but for others it is additional pressure, because the usefulness of AI starts to be confused with the number of times they use it.

• Write draft code faster
• Reduce time spent finding errors
• Generate documentation and test templates
• Increase personal visibility within the team

The problem is that such rankings measure activity, not quality.

Frequent prompts to a model do not yet mean strong engineering work. A developer may spend less time on routine and benefit, or may simply overload the system with endless clarifications and copy raw output without checking it.

As a result, a metric intended as a way to show progress in adapting to new tools risks turning into competition for the sake of competition.

This is especially noticeable where managers have not yet developed clear rules: what exactly counts as effective use of AI, and what is just empty bustle.

The cost of mass adoption

This mode also has a very down-to-earth side — money.

The more actively employees work with models, the higher the company's spending on compute and subscriptions. If AI becomes an everyday tool for dozens or hundreds of developers, the bill stops being a barely noticeable experimental line item. It turns into a tangible operating burden.

That is why news about rankings is accompanied by mentions of hefty bills: large-scale deployment of AI assistants delivers speed, but almost always increases the cost of every working day, especially when strong models are used.

For business, this creates a new balance.

On the one hand, AI really can remove routine, speed up the search for solutions, and relieve senior engineers of some repetitive questions.

On the other hand, companies will have to calculate the economics more carefully: where automation pays off, and where costs grow faster than the real return.

The logical next step is not just to track employee activity, but to compare it with outcomes: release speed, number of defects, code quality after review, and impact on product metrics.

Otherwise, leaderboards will show a lot of movement, but little meaning.

What this means

AI in development is rapidly moving from a personal tool for enthusiasts to a corporate norm with KPIs, rankings, and budgets.

The next question for the industry is no longer whether to use models, but how to measure their real value — by the number of prompts or by the quality of the result.