How much does artificial intelligence cost a business when it stops being an experiment and becomes part of daily processes? dev.ua asked the Master of Code Global team about this — the questions were answered by Iryna Nikulina, Delivery Operations Director, Iryna Shevchuk, Information Security Officer, and Bohdan Sergienko, Chief Technology Officer.
How much does artificial intelligence cost a business when it stops being an experiment and becomes part of daily processes? dev.ua asked the Master of Code Global team about this — the questions were answered by Iryna Nikulina, Delivery Operations Director, Iryna Shevchuk, Information Security Officer, and Bohdan Sergienko, Chief Technology Officer.
Yes, we have a separate budget. As of early August, our AI spending has increased significantly compared to last year. We don’t disclose the exact amounts, but subscription investment has increased 3,2x compared to 2025. The main item of our spending is Anthropic — Claude Teams and Claude Enterprise subscriptions.
Iryna Nikulina: AI services costs per employee are part of our operating budget, but we do not disclose specific average monthly figures.
Iryna Nikulina: The undoubted leader is Claude. The service is actively used by both developers for research and development and the business team in their daily work.
Iryna Nikulina: Each role involves tasks that should be performed by a purely human, hybrid (human + AI), or completely autonomous for AI. We delegated to AI those areas where LLMs demonstrate the highest efficiency: analytics, summarization, text content generation, and processing of online call records. Creating emails, developing policies, or preparing commercial proposals no longer take hours of valuable time. The time to complete some tasks has decreased from 1.5 hours to 10 minutes.
For example, previously, after a demo call, the PM was involved in structuring agreements, sending follow-ups, and updating tasks; now AI independently analyzes the recording, prepares a letter, and updates timelines. The manager only validates the final result.
A good example is our business process for preparing a large-scale annual report. Previously, this cycle took a full working week. With the introduction of the first “chats”, the time was reduced to one day, and this year the entire volume of work was completed in less than 2 hours.
Iryna Nikulina: Yes, cost optimization in these areas is taking place. However, these figures remain internal information.
Iryna Nikulina: Integrating AIDD into all our projects and products has significantly accelerated the work of teams and delivery. We evaluate efficiency by a number of technical parameters: the percentage of code generated by AI (from 70% in mature products to 99% in new developments), the level of coverage by automated tests (70–100%), and the degree of coverage of the project context by AI tools (90–99%).
However, the concept of AI-first or AIDD goes far beyond simple code or test generation.
If a developer is wasting resources on endless online meetings and requirements refinements, the gain in code speed becomes meaningless. Therefore, our goal was to fundamentally transform the entire workflow.
The introduction of AI did not result in layoffs due to increased productivity. At the same time, we restructured teams, mostly involving specialists who were unable to adapt to new methodologies and integrate AI into their practice.
Yes, the company has implemented internal rules and policies that regulate the use of artificial intelligence tools. They set standards for working with AI services, defining the categories of information that are allowed to be processed and the requirements for the tools depending on the type of data. In particular, we distinguish between services that are allowed to work with confidential information and those where its processing is not allowed.
The subscription type and model used vary depending on the customer’s requirements. For active spec-driven development, the token costs per developer are 1–3k USD for Anthropic models.
Bohdan Sergienko: It depends not on the level of the specialist, but on the nature of the work: if a senior is working on specifications, then his costs will be higher; if everyone works in the same mode, then the costs will be approximately the same. Token consumption depends on the volume, not on the complexity of the work.
Bohdan Sergienko: A developer who actively generates code spends significantly more resources, since the amount of text processed by the model is usually larger than in queries from other roles.
Bohdan Sergienko: Yes, we do. The size of the limits is confidential information, but it is a means of controlling costs. In situations where a specialist exhausts the quota within the standard workflow, we review and increase it. There were no cases of excessive or abnormal use of the entire set limit within one day.
Bohdan Sergienko: Costs can be considered justified when they ensure the achievement of the planned result. Well-established processes allow the team to work faster, which confirms the feasibility of investments from the point of view of the project’s economy. AI does not work miracles, but in certain types of tasks it helps to work more efficiently and obtain a higher-quality product, which is valuable for the end user and the client.
