Companies preparing to give AI agents real access to their systems should invest in flexibility and settle who is accountable for the agents' actions, according to Alexey Tulia, Executive Leader at Coinspaid Dev.
According to The Next Web, Tulia made these points while speaking on the AI Impact in Engineering panel at Tech Race Summit 2026 in Warsaw. The conversation centred on what happens as AI moves beyond helping with drafts and analysis and begins operating through company infrastructure, with reach into sensitive data and deployment pipelines.
Much of Tulia's advice concerned how technology leaders spend their budgets. He urged CTOs to link AI spending to a clearly identified organisational need, and named the foundations he considers most important: well-built APIs, dependable data, automated testing, observability, security and an architecture that can be adapted. These, he said, give a business the room it needs to adopt new technology safely. Work on architecture and on avoiding vendor lock-in may produce little revenue in the short run, yet it lets a company replace a supplier or rework a system when the original assumptions no longer apply. "I don't need to predict the future perfectly. I need to make being wrong cheap," Tulia said. He also cautioned that a roadmap which absorbs all available resources leaves teams unable to trial new tools or adjust to changing priorities.
The same thinking shapes his view of AI autonomy. Tulia asked the audience to imagine an agent that can put together a change and release it to production. Should it do that without a person signing off, and who answers for it if the release goes wrong? Before an agent receives such access, he said, the company needs controls on its permissions, audit logs of its activity, a way to stop it and a plan for recovering from a failed deployment. The broader lesson, in his words: "The more authority we give machines, the more important accountability becomes." More independence for AI in production, he argued, calls for precisely defined authority and a person who remains responsible.
His remarks arrived during a wider industry discussion about whether AI capabilities are outpacing safety practices. In September, Anthropic CEO Dario Amodei called for capability development to slow so that safety work could keep up. Tulia's focus was narrower and more immediate: the boundaries a single company should draw once its AI agents can touch production systems.
He also offered a forecast. By 2029, Tulia expects leaner engineering teams to take charge of broader areas of responsibility, while AI writes most production code. That would place greater weight on verification and technical judgment. The CTO, he said, will still need deep technical knowledge and a firm grasp of the business, especially as easier development brings more vendors and AI-built systems into each organisation. "I think technical judgment becomes even more important," he said.
For individual engineers, Tulia sees the faster pace of coding and prototyping as a chance to engage more closely with the business problem and to stay involved as their work runs in production. Leaders can encourage this by sharing the business context and stating the expected outcome, then evaluating teams on correctness, maintainability, security and operational performance instead of the amount of code they produce.
Coinspaid Dev is an independently owned software engineering company specialising in blockchain infrastructure, with more than 120 engineers, over 11 years of experience and teams working across more than 20 blockchain networks. For engineering leaders, Tulia's conclusion was that safeguards and clear ownership must be established before AI agents are given access to critical production systems.
