Since February 2026 I no longer work as an employee. Just like that :)
What pushed me to this decision was a growing sense of professional stagnation and a lack of prospects at my last job — but that is a topic for a separate post. I decided to finally devote the freed-up time to my own projects, many of which had been getting leftovers of my attention for years.
Around the same time, my approach to using AI changed dramatically. Before, it mostly helped write a single function, debug an error, or draft data and texts for checking an interface. By the end of 2025, models with huge context windows (1M tokens) started to appear, and by early 2026 they became the standard thanks to Chinese companies. The capabilities of agents themselves also grew a lot: skills appeared, MCP, and models got noticeably better at working with tools. The agent stopped being just a talking reference book and an apprentice for simple routine — it gained the ability to cover entire small projects with documentation in its context, reliably work with external systems, use sets of different tools, plan its work and carry out complex tasks.
At some point I caught myself thinking something quite strange: for certain tasks I no longer double-check the agent’s work — I just read its completion report and look at the result. Of course, you cannot always do that, but half a year ago such an approach to AI would have seemed wild.
In effect, by leaving employment my “team” got dramatically stronger. I still own the products entirely: what we build, for whom, why, in what order, and what the result should look like. A significant share of the work can be delegated — from simple research and analysis to implementation, testing and refining full-fledged tasks.
This moved a lot of projects forward that had long lacked attention. In this short time LumaSteward appeared and even went through several releases, ProVerka reached a stable state and finally became a public project, Stoballnik started growing rapidly, and the new “V kurse” project is already in pre-release testing. The excitement of ideas finally no longer being limited solely by the throughput of my hands and time was so strong that sometimes I could literally work around the clock without sleep, switching between projects and bringing to life what I had long wanted to do.
But the most unexpected effect was not even the acceleration of my work. Working with agents is very much like working with a young team of very well-read, productive and diligent employees who, at the same time, seriously lack real business experience and reflection. They can get a lot of work done very quickly — including work that is completely not what the project actually needed :)
That is why working with agents unexpectedly strengthens my management skills even more. You need to formulate the goal properly, pass on all the necessary context, define the stages of implementation, decompose tasks and describe the acceptance criteria, check the direction of “thinking” in time, and understand what should not be done at all.
The better agents become, the less sense it makes to control individual actions, and the more important it is to set up the process itself correctly. In that sense, little has changed compared to a regular team — it is just that the team has become noticeably more powerful and cheaper. And for the first time in a long while, I have the feeling that the pace of my projects’ development is no longer limited by the number of free hands and the team’s experience, but by how well I understand where and how to lead the development.