For founders and technology executives, the pressure to invest in AI is constant, yet the returns often depend on decisions that have little to do with the latest model. Alexey Tulia, Executive Leader at Coinspaid Dev, argues that the companies best prepared for AI are the ones that can change course quickly and know exactly who is accountable when machines act.
According to DEV Community, Tulia made his case during the AI Impact in Engineering panel at Tech Race Summit 2026 in Warsaw, a discussion about how AI is redefining the responsibilities of engineers and CTOs. His perspective comes from leading engineering at Coinspaid Dev, an independently owned and operated software company specializing in blockchain infrastructure development. The firm has more than 120 engineers, over 11 years of industry experience and dedicated software engineering, infrastructure, security and R&D teams whose systems operate across more than 20 blockchain networks.
Tulia’s first message to technology leaders concerns money. AI budgets, in his view, should be attached to a specific problem the organization needs to solve, with a clear link between the spending and the business result. What makes those investments pay off is often the less visible groundwork that allows new technology to be introduced without disrupting operations. He named several building blocks that give a company that room:
- strong APIs that let new tools connect to existing systems;
- reliable data that both AI models and people can trust;
- automated testing that catches problems before customers do;
- observability, so teams can see how systems behave in real time;
- security practices that keep expanded access under control;
- flexible architecture that can be reshaped as requirements evolve.
Two further points complete his investment logic. The first is capacity. A roadmap that commits every engineer to planned features leaves nothing for experimentation, so a team cannot test a promising tool when it appears or respond when priorities move. Entrepreneurs who have watched a competitor adopt a new technology faster will recognize the cost of that rigidity. The second is independence from vendors. Work on architecture and on reducing lock-in rarely produces revenue in the quarter it is done, and it can be hard to defend in a budget meeting. Its payoff arrives when a provider stops fitting the company’s needs or an earlier assumption turns out to be wrong, and the business can switch or rebuild without starting from scratch. Tulia captured the philosophy in one remark: “I don’t need to predict the future perfectly. I need to make being wrong cheap.” For a leader placing bets in a fast-moving market, that means designing systems in which a wrong call can be corrected at a manageable price.
The same thinking applies to control. Businesses already rely on AI to help draft content and analyze information, but the next step connects AI agents to live systems, including sensitive data and deployment pipelines, and gives them the power to act. “The more authority we give machines, the more important accountability becomes,” Tulia said. He illustrated the point with an agent able to prepare a change and release it to production. Leadership has to decide whether that release can happen without a person approving it, and it has to know in advance who will answer for the outcome if the deployment fails. Before an agent receives this kind of access, Tulia said, the organization needs permission controls, audit logs of the agent’s actions, the ability to stop the agent and a dependable process for recovering from a failed release. His broader argument is that the more autonomy an AI system has in production, the more precisely its authority must be defined and the more clearly human responsibility must be assigned. For founders, this is a governance decision as much as a technical one, because an unchecked agent working in production can affect customers, revenue and reputation.
Tulia also sees AI changing what companies should expect from their engineers. Faster coding and prototyping free up time that engineers can spend understanding the business problem and staying with their work until it performs reliably in production. That shift only happens if leaders make it possible. Teams need the business context behind a project and a clear definition of the outcome it is supposed to achieve. With that information in hand, managers can evaluate productivity by the qualities that matter to a company: whether the software is correct, whether it can be maintained, whether it is secure and how well it performs in operation. The amount of code a team produces becomes a weak indicator when much of it can be generated automatically, and rewarding volume can quietly encourage the wrong behavior.
His outlook for 2029 raises the stakes further. Tulia expects smaller engineering teams to take charge of broader areas of responsibility and AI to produce most of the code that reaches production. In that environment, verification and technical judgment become central skills, since someone must confirm that generated code does what it should. The CTO role, he said, will continue to require deep technical expertise together with a strong understanding of the business. As creating technology gets easier, organizations will take on more vendors and more AI-generated systems, and each of them will need to be assessed by someone qualified to judge it. “I think technical judgment becomes even more important,” Tulia said. His practical advice for leaders is to act before the pressure arrives: define the safeguards and decide who owns each decision before AI agents gain access to critical production systems. Companies that settle those questions early will be able to expand the role of AI with confidence, while those that postpone them may end up answering hard questions after something has already gone wrong.
