Businesses should put clear safeguards and lines of responsibility in place before letting AI agents operate inside their core systems, a software engineering executive said at a technology conference in Poland.
As covered by Tech Times, Alexey Tulia, Executive Leader at Coinspaid Dev, raised the issue on the AI Impact in Engineering panel at Tech Race Summit 2026 in Warsaw. He spoke about how the work of engineers and chief technology officers is shifting as AI moves from assisting with writing and analysis to acting directly within company infrastructure, with access that can extend to sensitive data and the pipelines used to release software.
The discussion comes as the technology industry debates whether AI is advancing faster than the practices meant to keep it safe. In September, Anthropic CEO Dario Amodei called for capability development to slow so that safety work could catch up. Tulia focused on the same concern at the level of a single company, asking what limits an organisation should set once its AI agents can reach production systems.
He offered an example. If an AI agent can prepare a software change and release it to customers on its own, should it be allowed to skip human approval, and who is responsible if the release fails? Tulia said companies need permission controls and activity logs before granting such access, along with the ability to shut the agent down and recover from a failed deployment. “The more authority we give machines, the more important accountability becomes,” he said.
On spending, Tulia told technology leaders to connect AI budgets to a specific business need. He named strong APIs, reliable data, automated testing, system monitoring, security and adaptable architecture as the priorities that allow a company to adopt new tools safely. Some of that work, including efforts to avoid dependence on a single vendor, brings little immediate revenue, but it makes switching providers or redesigning systems easier when plans change. “I don’t need to predict the future perfectly. I need to make being wrong cheap,” Tulia said. He added that teams need spare capacity to test new tools and adjust to shifting priorities.
Tulia also said faster AI-assisted coding should free engineers to understand the business problems behind their work and follow it after launch. He encouraged managers to share business context and clear goals, and to evaluate teams on the correctness, maintainability, security and reliability of their work instead of how much code they write.
Looking ahead, Tulia predicted that by 2029 smaller engineering teams will handle broader responsibilities while AI produces most production code, making verification and technical judgment more critical. He said the chief technology officer will still need deep technical expertise and business understanding as more vendors and AI-built systems enter companies. “I think technical judgment becomes even more important,” he said.
Coinspaid Dev is an independent software engineering company specialising in blockchain infrastructure. It has more than 120 engineers and over 11 years of experience in the industry.