Three Lessons for Engineering Leaders From Alexey Tulia’s Talk on AI Agents

According to TechBullion, Tulia presented his ideas on the AI Impact in Engineering panel at Tech Race Summit 2026 in Warsaw, where speakers examined how the responsibilities of engineers and CTOs are evolving. The discussion took place against the background of a wider argument about whether AI is advancing faster than safety practices can keep up. The outlet points out that Anthropic CEO Dario Amodei called in September for capability development to slow so safety work could catch up. Tulia focused on the company-level version of that problem: what limits a business should place on AI agents once they can reach its production systems.

The first lesson concerns money and priorities. Tulia believes AI spending should always answer a specific organisational need. In his view, the most useful investments are often the least visible ones, such as well-built APIs and dependable data, together with automated testing, observability, security and an architecture that can evolve. These elements create the conditions for bringing in new technology without disrupting the business. He also recommended protecting some team capacity for experiments, because a fully booked roadmap makes it nearly impossible to trial a new tool or react to changing priorities. Reducing reliance on individual vendors works the same way. It may not boost revenue in the short term, but it keeps the cost of switching providers or redesigning a system manageable when circumstances change. As Tulia put it, the aim is to make being wrong cheap, since nobody can forecast the future with precision.

The second lesson is about the people doing the work. With AI accelerating coding and prototyping, engineers have the chance to spend more of their attention on the business problem itself and to stay with their work after it reaches production. Tulia said leaders make this possible by providing business context and a clear picture of the desired result. Once teams work that way, output should be assessed by how correct, maintainable, secure and operationally reliable it is. The quantity of code written becomes a poor indicator of value.

The third lesson, and the most pressing, deals with authority. Agents that are connected to live systems, including sensitive data and deployment pipelines, will be able to take real actions. Tulia argued that accountability has to grow as machines gain authority. He illustrated the point with an agent able to prepare and deploy a production change, which raises two questions: whether it should proceed without human sign-off, and who is responsible if the deployment breaks something. His answer is that a company must build the right protections before granting access:

• permission controls that set strict limits on what the agent can do;
• audit logs that capture a full record of its actions;
• the ability to stop the agent at any point;
• a recovery process that restores systems after a failed deployment.

With these in place, greater autonomy becomes possible, provided authority is clearly defined and humans stay responsible for the outcome.

Tulia also looked toward 2029. He expects smaller engineering teams to take on wider areas of responsibility and AI to write the majority of production code, which will make verification and technical judgment more critical. CTOs will still need strong technical expertise combined with business insight, especially as simpler software creation fills companies with more vendors and AI-generated systems. For now, his advice to engineering leaders is to define safeguards and ownership before any agent is trusted with critical production environments. Coinspaid Dev, where Tulia serves as Executive Leader, is an independently owned software engineering company focused on blockchain infrastructure, with more than 120 engineers, over 11 years of experience and systems operating across more than 20 blockchain networks.