research, tools, essays, and long-form writing from anton morgunov on ischemist.com.
AI has collapsed the cost of sophisticated cyberattacks. A breakdown of the new threat landscape, the tactical defenses you need to adopt, and a look at the existential stakes this shift entails for society. This fugue contains 3 voices: Attacks that read like science fiction; Why lockfiles aren't enough: your new defense plan; The inevitable dual-use problem of frontier AI.
Three essays on why even simple for-loop software has absurd leverage, why AI will not reduce demand for the value of SWE-style thinking, and why progress is often a latency problem. This fugue contains 3 voices: A butterfly effect from a nested for loop; AI may write the code. We still need SWE minds; Progress is constrained by latency.
A full translation of an interview with Grigori Perelman's math teacher. He explains Perelman's rejection of the Fields Medal as a protest against a 'dishonorable' math community that treats theorems as a commodity to be stolen. Also features a brutal, unapologetic defense of Soviet-era educational philosophy.
A comprehensive review arguing that synthesis planning is the chemical equivalent of next-token prediction — the foundational objective for artificial chemical intelligence. Venue: Chemical Reviews.
Deriving the necessity of eternal punishment from the Prisoner's Dilemma. How infinite repeated games, discount factors, and the Folk Theorem explain the structural utility of Hell in fostering human cooperation.
A complete walkthrough of modern python tooling, from understanding virtual environments to mastering lockfiles and dependency groups. Why I switched from Conda to uv, and why you should too.
How to benchmark conflicting academic codebases without losing your mind. Moving from venv to uv's conflicting groups, and eventually to independent project isolation.
A dive into the gregorian calendar as a 400-year software patch for the Julian's memory leak. A deployment full of hilarious bugs, like Sweden inventing a February 30th, with a surprising connection to our modern holiday traditions.
Cursor is dying because cost-optimization forces models into tunnel vision. RAG agents fail because they only see what they search for. The superior workflow for 2026 is massive context windows (gemini 2.5 pro) and manual control. Stop letting agents hide code from the model.
a deep dive into how the metrics for ai chemical synthesis are broken, rewarding models for "solving" routes with impossible, hallucinatory chemistry. we present the data and the open-source tools to fix it.
science is infrastructure-constrained, not ideas-constrained. an argument on why better tooling, not smarter models, is the missing link in the AI for Science revolution.
we built an ai steering system for molecular design that efficiently guides a generative model to invent potent, protein-specific molecules, even rediscovering existing drugs from scratch. Venue: J. Chem. Inf. Model.
we developed a computational shortcut that predicts quantum properties with gold-standard accuracy at a fraction of the cost, making high-precision modeling accessible for larger molecules. Venue: J. Phys. Chem. A.
We will see why chemistry could not exist as a science without Born Oppenheimer approximation and what career paths that approximation opens.