The global conversation about AI safety tends to focus on the companies that build the most powerful models. Alexey Tulia, Executive Leader at Coinspaid Dev, is drawing attention to a more immediate concern: what happens when those models begin working inside ordinary businesses.
As reported by TechBullion, Tulia raised the issue during the AI Impact in Engineering panel at Tech Race Summit 2026 in Warsaw. His remarks came at a time of broad debate over whether AI capabilities are developing faster than the practices designed to keep them safe. The publication notes that in September, Anthropic CEO Dario Amodei called for slowing capability development so that safety work could catch up. Tulia brought that discussion down to the scale of a single organisation, asking where companies should set boundaries once AI agents can operate within their production systems.
Today, most businesses use AI as a helper for drafting documents and analysing information, with people still carrying out the final actions. The next phase, as Tulia described it, will link agents to live infrastructure, from sensitive data to deployment pipelines, giving them the ability to act on their own. That access immediately raises questions about what an agent may do and who carries responsibility for each decision it makes. Tulia summed up his stance by saying that accountability becomes more important with every increase in the authority people hand to machines. To show what this means in practice, he pointed to an agent capable of preparing a change and pushing it into production. Should it be allowed to complete that step without a human approving it? And if the deployment fails, who answers for the consequences?
According to Tulia, an organisation should settle these questions before it grants such powers. It needs permission controls that restrict the agent to clearly defined tasks, as well as audit logs that make every action traceable. It must also be able to stop the agent whenever necessary and restore normal operation after a failed deployment. Taken together, these conditions express the core of his argument. An agent can be given greater autonomy in production only when its authority is precisely defined and a named person remains responsible for the results. For engineering leaders, this makes safeguards and ownership the first item on the agenda, well ahead of any decision to connect agents to critical systems.
The same shift is reshaping what companies expect from their engineers. AI already accelerates coding and prototyping, and Tulia believes the time saved should go into understanding the business problem and following each piece of work into production. Managers have a role to play here. When they explain the business context and describe the outcome a team is expected to achieve, engineers can take real ownership of the result. Performance can then be judged by correctness, maintainability, security and operational stability, while the amount of code produced stops being a meaningful measure.
Tulia also addressed how CTOs should approach AI budgets. His advice is to connect spending to a specific need within the organisation. The investments he prioritised are those that give a company room to introduce new technology safely: strong APIs and reliable data, supported by automated testing, observability, security and a flexible architecture. Teams also need spare capacity, since a roadmap that absorbs all available resources leaves no space to test a promising tool or adjust when priorities change. Architecture work and efforts to reduce vendor lock-in may bring in little revenue at first, yet they make it much easier to replace a provider or rework a system when earlier assumptions prove wrong. Tulia said he has no expectation of predicting the future perfectly, and that his real goal is to make being wrong cheap.
His outlook for the coming years points in the same direction. By 2029, Tulia expects smaller engineering teams to handle larger areas of responsibility and AI to produce most production code, which will raise the importance of verification and technical judgment. The CTO role will continue to require deep technical expertise combined with business understanding, particularly as easier software creation brings more vendors and AI-built systems into companies. In his opinion, technical judgment will become even more valuable than it is now. Tulia represents Coinspaid Dev, an independently owned and operated software engineering company specialising in blockchain infrastructure development. With more than 120 engineers and over 11 years of industry experience, the company combines software engineering, infrastructure, security and R&D teams that build distributed systems and blockchain infrastructure across more than 20 blockchain networks
Facebook
Twitter
Instagram
LinkedIn
RSS