At a Warsaw tech conference, Alexey Tulia, Executive Leader at blockchain infrastructure developer Coinspaid Dev, said that AI agents gaining access to live company systems change what engineering leaders need to plan for.
As reported by The Tech Portal, Tulia made his case during the AI Impact in Engineering panel at Tech Race Summit 2026. So far, most businesses use AI to draft text and run analysis. In the next phase, agents will get real permissions over data, infrastructure and deployment pipelines. Even when software acts on its own, a person still has to answer for the result.
Consider a practical case. An agent writes a code change and pushes it to production. Did a human approve it? If the release breaks something, who deals with the fallout? Tulia believes companies should answer these questions before they grant access. The groundwork is concrete: permission controls, audit logs, a reliable way to stop the agent, and a recovery path when a deployment fails. “The more authority we give machines, the more important accountability becomes,” he said.
His advice on spending follows the same logic. He thinks AI budgets should serve a specific organisational need. The investments he ranked highest are mostly invisible to customers:
- well-designed APIs and trustworthy data
- automated testing, observability and security
- a flexible architecture that lets teams replace components or providers
Reducing dependence on a single vendor rarely shows up in quarterly revenue. It does let a team change direction cheaply when an assumption proves wrong. “I don’t need to predict the future perfectly. I need to make being wrong cheap,” Tulia said. He also pointed to spare capacity: a roadmap that absorbs every hour of engineering time leaves nothing for testing a new tool or reacting to shifting priorities.
For individual engineers, faster coding and prototyping move the job closer to the business problem itself and to how the result performs once it ships. Leaders can support this by sharing context and a clear expected outcome, then judging productivity by correctness, maintainability, security and operational stability. The amount of code written says very little on its own.
Looking ahead to 2029, Tulia expects smaller teams to own broader areas and AI to write most production code, which makes verification far more valuable. CTOs will still need deep technical knowledge combined with business sense, especially as cheaper development brings more vendors and machine-built systems into the organisation. “I think technical judgment becomes even more important,” he said. His practical message to engineering managers is to set safeguards and assign ownership now, before agents reach critical production systems.
