My thoughts on agentic coding | vucenovic

aleksandar vucenovic<br>software engineering +

blog<br>portfolio<br>github

20 August 2026

My thoughts on agentic coding -

There are likely thousands of these kinds of blog posts out there by now. I read at least a dozen of them, and they were more often than not written by people who have either more prior “traditional” software engineering experience (over 7 years) than I have, or people who who have been very passionate about the craft of writing and reviewing code.

Usually there are two positions represented in those samples:

AI evangelists talking about a point-of-no-return - the paradigm shift happened, they are Yx more productive now, shipping code by hand is seen as insanity or cope.

Engineers who saw programming as a craft, experiencing the push of agentic coding in workspaces as deprecation of their skill-set and ultimately as some form of dehumanization.

I always saw myself as a generalist, one who builds knowledge through deep-diving various topics periodically for shorter time frames. Even though my CV consists mainly of backend and data engineering roles, this description is applicable to a vast number of areas, including design, product development, branding; but also areas which are completely unrelated to the corporate business world, such as philosophy, music and art. For a profile such as mine, the shift to agentic coding therefore removes a bottleneck which was very real in the job market before Claude took over: The age of a required specialization is over. There is not as much need anymore to study a web framework for years to become really good at it or remember how a database works internally to boost its performance when your agentic assistant can answer most questions about them in a matter of seconds. On the other hand, having a good intuition for the bigger picture became even more valuable - the people getting most out of agentic coding tools are the ones, who can combine their expertise in software architecture with a feeling for UI/UX, documentation, business requirements, etc.

I went down this path myself by using my skillset to work on ideas which were lingering in my head for quite a while already, one of them being a creative project where I worked on a real-time GPU-accelerated 3D Physarum transport network simulation in Rust/nannou, where hundreds of thousands of agents self-organize into luminous web-like structures through nothing but local chemical sensing,rendered with additive accumulation, bloom, and a density-mapped gold-to-ember color palette. Another project of mine included working as the sole engineer for a SaaS startup. Both of these projects were kind of unimaginable for me to build prior to the agentic revolution. They did require strong fundamentals and certain confidence in what I am doing, but I felt much more like a product developer/designer than a “traditional” software engineer, who sometimes struggled with handling async event loops and debugging a race condition for days or even weeks.

While the possibilities that opened for me seemed like a gold mine, the feeling that I had in reality when finishing my simulation project was humbling. The inspiration for it was the work of a 3D artist I used to admire a lot when I was still a teenager and started my computer science studies. I was doing my fair share of 3D art myself, and his work - which was obviously inspired by complex biological processes, such as the growing of neural or bacteria networks - seemed unobtainable for me, a pure craft, which was truly one of a kind (the artist was chosen to work with NASA for one project at that time). It took me about ten minutes to research the biological processes he was inspired by using screenshots of his work, and about another one or two hours to implement simulations which were scaringly similar to his in a programming languages I had hardly used in years, with a framework I was sometimes interested in as student but never had the time enough to try in the end.

Of course there are examples where I let my own creativity play a bigger part in the outcome of my agentic endeavours. But a part of this feeling still sticks - are these accomplishments really mine? Did the agent’s feedback during our conversation lead me down a certain path, that many others are taking by using the same agent, trained on the same data? I recently watched a video of a talk by Iain McGilchrist1, who argued that AI replicates the left-hemisphere (rational thinking) functions at an immense speed, but does not possess the encounters made by the right-hemisphere, such as creativity, humour, narrative, the unknown and the sacred - things many of us connect to the essence of human being. The statement does light a spark of recognition for me - in the context of how I engaged with my ideas in the process of making them reality as stated in the aforementioned paragraphs. The collaboration was reduced to me talking to a machine only, as there...