By 2029, engineering teams may be smaller, while the systems they oversee grow larger and are increasingly built by AI. That forecast, shared by Coinspaid Dev Executive Leader Alexey Tulia, comes with a clear warning about responsibility.
According to BlockchainReporter, Tulia presented his outlook during the AI Impact in Engineering panel at Tech Race Summit 2026 in Warsaw. He expects AI to write the majority of production code within the next few years, while compact teams take ownership of wider areas of a business. In that environment, the ability to check what AI produces and to make sound technical decisions becomes one of the most valuable skills an engineer can have. He also stressed that easier software creation will not simplify the CTO’s job. Companies will bring in more vendors and more AI-generated systems, and someone with deep technical knowledge and a firm grasp of the business will need to decide which of them can be trusted.
The shift is already visible in everyday engineering work. AI tools now speed up writing code and building prototypes, freeing time that used to go into routine implementation. Tulia argues that engineers should invest that time in understanding the business problem behind each task and in following their work all the way into production. This changes what good performance looks like. Instead of counting lines of code or closed tickets, leaders can evaluate whether software works correctly, whether it is easy to maintain, how secure it is and how well it behaves under real operating conditions. For that approach to work, managers have to share the wider business context with their teams and define the outcome they expect, so engineers know which problem they are actually solving.
The bigger change concerns what AI is allowed to do. Most organisations currently use AI in a supporting role, asking it to draft text or analyse information while people make the final moves. Tulia sees the next stage as one in which agents are connected to live company systems, from sensitive data to deployment pipelines, and can take action directly. He offered a simple case to illustrate the stakes. Picture an agent that prepares a change to the codebase and can release it to production on its own. Should it be trusted to do this without a human approving the step? And if the release fails, who is accountable for the damage? In his view, these questions need answers long before the technology is switched on.
Tulia’s position is that accountability matters more with every bit of authority handed to machines. Before an agent receives production access, a company should have permission controls that define exactly what it can reach, along with audit logs that keep a record of each action it takes. The organisation must also be able to halt the agent at any moment and roll back a deployment that goes wrong. Only with these mechanisms in place does it make sense to expand autonomy, and even then responsibility for each decision should stay with a clearly identified person. For engineering leaders, he framed this as the most urgent practical task: settle safeguards and ownership first, then open access to critical systems.
Spending decisions follow the same logic. Tulia advised CTOs to connect every AI investment to a specific need within the organisation instead of buying tools because they are fashionable. The foundations he highlighted are the ones that make safe experimentation possible: solid APIs, reliable data, automated testing, observability, strong security and an architecture flexible enough to accept new components. He also warned against roadmaps that consume all of a team’s capacity. Without some unallocated time, engineers cannot try out an emerging tool or respond quickly when priorities change. Work on architecture and on reducing dependence on any single vendor seldom produces immediate revenue, yet it pays off when a provider has to be replaced or a system redesigned because earlier assumptions no longer hold. Since perfect prediction is impossible, Tulia’s own goal is to keep the price of a wrong guess about the future as low as possible.
These ideas apply well beyond the financial technology sector. Any company that builds software, from a small studio shipping updates to a large platform serving millions of users, will soon face the same choice about how much freedom to give AI agents inside its infrastructure. Tulia’s perspective is shaped by his work at Coinspaid Dev, an independently owned and operated software engineering company specialising in blockchain infrastructure. The firm employs more than 120 engineers, has over 11 years of industry experience and brings together software engineering, infrastructure, security and R&D teams that build distributed systems running across more than 20 blockchain networks. His message to the industry is practical: AI can take on far more of the work, provided that people remain clearly in charge of the decisions it makes.