We are Vedran and Matea, two academics who are enthusiastic about scientific research, especially in the field of computational chemistry, and open-source software, especially Linux, Python, FreeBSD, LaTeX, and Markdown. We sometimes write about these and related topics on this blog.
It has been three months since we published our preprint, and it has now passed 500 downloads. That seems like a good occasion to write about how I ended up working on molecular simulations and scientific software after leaving a promising assistant professorship.
I have tried to write about that decision several times, but the drafts kept growing and none reached publication. There was always more to explain. This time, I will focus on what led to the decision and what came of it.
Attending the Agentic AI for Science workshop at MPCDF in Garching got me thinking about the platforms we are starting to build our scientific workflows around. The workshop programme brings together hands-on work with agents, examples from scientific software development and materials simulations, and perspectives from industry. For someone interested in free and open-source software, this raises some familiar questions: who controls those platforms, what can we change ourselves, and how much of our work can we take elsewhere?
Coding agents use language models to generate code and tools to execute it, while engineers define the problems, guide the work, and verify the results. What interests me here is the software around the model. An agentharness supplies context, exposes tools to the model, and manages their execution. The flagship GPT, Claude, and Gemini models are proprietary, but the software using them does not have to be (and often isn't, as we will discuss below). We can use the same model through different harnesses, with different licenses, tools, and restrictions. Choosing the model, deciding where it runs, and choosing the harness are separate decisions, even when a product offering bundles them together.
A few harnesses and development environments for programming, scientific workflows, and other tasks illustrate these differences. This is only part of the picture: agents embedded in enterprise applications bring their own integrations and dependencies.
This led me to a somewhat loose comparison: Claude Code reminds me of Windows, Codex CLI of BSD, OpenCode and the like of Linux, and Antigravity of the Macintosh. Each comparison emphasizes something different: compatibility, permissive reuse, independence from suppliers, or the appeal of an integrated experience.
If you have used Claude Code in a terminal, the first part probably sounds wrong already. How does a program that reads files, invokes shell tools, and fits into a Unix workflow resemble Windows? Bear with me. The resemblance is in how people build tools and workflows around the software, rather than how its interface looks.
The other day, I set up a machine using a fairly standard Fedora43Server installation with a WireGuard VPN tunnel to another machine running a fairly standard Arch Linux installation (of course).
When I tried to SSH from the Arch machine to the Fedora machine, the OpenSSH client would hang.
I would like to emphasize the two reasons that she mentioned why career guidance is so difficult in Croatian academia (translation mine):
Limited experience of the mentors: Many mentors rely on their own experience in a system they are part of and know very well, and outside of which they often cannot help us much (depending, of course, on the field and personal affinities).
The “academia-only” narrative: Continuing higher education and a doctorate seem to presuppose a postdoc position and a teaching career. In addition, a good number of academics I was (and still am) surrounded by are convinced that academia is the only place where you can be creative and flexible, which is not true and makes it difficult to explore other possibilities.
I thought quite a bit about both of these points, given that I am guiding Matea Turalija through her PhD journey and I am eager to help her with my acquired experience as much as I can. However, as Klara explained above, this is far from a trivial endeavor.
The fix for the issue also added several tests to avoid future regressions. I like developing free and open-source software in general and GROMACS in particular; it feels a lot like the postdoc years in Heidelberg again. However, that was not the only change I proposed that got merged for the bugfix release.
Over the last couple of weeks, there has been a lot of discussion about scientific research funding in the US. There are many changes, and it remains to be seen what will come out of it in the end. If some of the new policies stick around for a longer time, I have no doubt they will inspire discussions and considerations in the rest of the world, including Europe, as many of them in the US tend to do.
While science funding in the US is an important and interesting topic, I would like to say a couple of words regarding science funding that hits closer to home, namely in Croatia. Earlier this week, the Croatian Science Foundation (HrZZ) closed the project call for Installation Research ProjectsUIP-2025-02. This was the first project call of this type after 5 years, which is possibly the reason why the usual range of eligible scientists with a maximum of 7 years after obtaining the PhD degree was extended to a maximum of 10 years after it. The call still requires a minimum of 2 years of experience after the degree.
As many readers of this blog are already aware, we make great use of GitHub Pages for hosting this website and severalothers. In particular, after FIDIT's inf2 server was finally decomissioned, Pages was the obvious choice for replacing the remaining services it offered.
Since the number and variety of applications and services hosted on inf2 server grew and diminished organically over time, what remained afterward was a collection of complex, but unrelated link hierarchies that had to be redirected to new locations (remember that Cool URIs don't change).
