Instead of an introduction

Tasks get automated faster than professions. An Ansible role, a Terraform module, a GitLab pipeline, a Helm chart, log analysis — AI already handles these reasonably well. In a couple of years this kind of work will boil down to "AI writes, a human reviews". If an engineer's value lies in writing YAML, the pressure on them will be strong, and soon.

Responsibility for production is going away more slowly. Someone decides what to roll out, brings the system back up at three in the morning, negotiates with the security team and the regulator, knows their infrastructure with all its history and workarounds, and is legally accountable for the outcome. In Russia, with its on-prem setups, air-gapped networks, FSTEC requirements and import substitution, autonomous AI will not be given access to production within a year or two. It will arrive here more slowly than in Silicon Valley.

Over the next 2–4 years teams will shrink, and a strong engineer with AI tools will do the work of three. Juniors and those closing routine tickets will be hit first. Beyond that it is hard to predict: there are serious people expecting much faster change, and that scenario is quite possible. Infrastructure with real accountability, security and hardware look more resilient than the rest.

So my plan is:

• Bring AI agents into operations and keep them under control: access, restrictions, auditing of agent actions in infrastructure. The role is new, and there will be demand for it.
• Shift from execution to architecture, security and decisions — where a human signature is required.
• Use AI to the fullest right now, so that I am that "one doing the work of three" myself.
• Keep a financial cushion. Not panic, just ordinary insurance for a period of uncertainty.

The closer you are to accountability, context and security, the better your chances of surviving.