Blog
Architecture, AI and Mind
Software architecture and code quality. AI — from real-world deployment to questions of consciousness, knowledge and values. And, at the edge, the philosophy of science and the biology of symbiosis. No filler, with a clear line between hypothesis and what we know.
AI as a Scientific Instrument: When a Model Discovers and When It Only Speeds Up the Search
We have sped up the search. We have sped up verification only where it could be handed to a machine — and that is the smaller part of science.
AI Agents and Technical Debt: What Happens When the Team Writing the Code Is Faster Than the Team That Understands It
Agents did not bring a new kind of technical debt. They removed the moment in which debt is taken on deliberately.
Brainrot: What We Know About Short Videos, What We Suspect and What We Make Up
Whoever defends a right conclusion with a bad argument hands the opponent ammunition. Right conclusions deserve arguments that can carry them.
Mechanistic Interpretability for Practitioners: What You Can Actually Find Out Inside a Model
Nanotechnology Without the Sci-Fi: What "Nano" Means and What Is Just Marketing
Forever Chemicals: Why the Problem Can't Be Fixed One Substance at a Time
When to Trust AI: A Practical Framework for Developers and Teams
You do not need to know whether the model can be trusted. You need to arrange the work so that it matters less.
Choosing an Open-Weight Model: By Use Case, Not by Leaderboard
The leaderboard answers a question you do not have. You are not looking for the best model — you are looking for the worst one that still suffices.
GaN and SiC: When Silicon Is Not Enough
"Silicon is not enough" is true, but only in one discipline: power electronics. GaN and SiC won it not through miracle physics but because their manufacturing grew up.
Chiplets: Why the Future of Chips Is Not One Big Piece of Silicon
AI Hardware: Why GPUs Won and Where the Memory Wall Hits
How a Chip Is Made: From Sand to EUV Lithography, No Miracles
You Downloaded a Model — What Did You Let In?
You download the weights, give the agent tools and let tens of gigabytes of unreadable logic into your system. No tool will tell you it is clean — so build so that it does not matter.
What Your Model Was Trained On — And the Risk You Inherit
You know the context window, the price per token and the benchmark score. What the model actually read you almost certainly do not — and that is a risk you take on with it, unseen on the invoice.
Graphene: The Wonder Material Still Waiting After Twenty Years
Graphene won a Nobel Prize and hundreds of "miracle" headlines. Twenty years on, it is in no chip. That contradiction is the best lesson in how to read the promises of wonder materials.
No AI Model Is Apolitical: How to Recognize and Test Model Bias
Why AI Model Benchmarks Are Not Enough: What to Measure When You Care About Real Work
The Chip War: Why the Modern World Hangs on a Few Factories
Codex vs Claude Code: What Changes When the Editor Stops Being Just an Editor
“AI in the editor” no longer means autocomplete but an agent that reads an issue and opens a PR. The tool choice is not the key decision — where you trust the agent and where you hold the line is.
Truth, Sources, and RAG: Why Citations Alone Are Not Enough
A citation looks like proof, but RAG only solves where the model got its information — not whether it holds. Four ways a citation lies, and how to measure truth, not just faithfulness to a source.
The End of Moore's Law? First Ask Which One
A vendor announces a "2-nanometer chip" and it sounds like engineering at the edge of the atom. But that number measures nothing — and "the end of Moore's law" depends on which of the three you mean.
Reasoning in Words or in Vectors? Latent Reasoning and the Window We Might Close
How Language Models Get Hacked: A Map of Attacks From Training to Deployment
Prompt Injection: The Security Hole You (Still) Can’t Patch
How to Track New AI Models Without the Hype: A Practical Radar for OpenAI, Claude, Gemini and Coding Agents
A tweet, a demo, a benchmark, a podcast — and the question that actually matters: did anything change that I should ship? Usually no, occasionally a lot. Only a disciplined radar tells them apart.
AI Agent Memory: Why a Long Context Is Not Memory
You give an agent a million-token context and expect it to stop “forgetting”. But memory is an umbrella word for several different mechanisms — and a long context solves only one of them.
Fine-Tuning Is Not Teaching: What You Risk When You Touch the Weights
Fine-tuning looks like the natural step after prompting. But it forks the model and takes on infrastructure: the durable asset is the data pipeline, not the weights, and the real cost is operational.
Can AI Invent Something New, or Only Recombine?
Technical Debt and Architecture: Why You Will Not See the Most Expensive Debt in Code Review
When Does AI Deserve Moral Status? A Decision You Can’t Outsource to Philosophers
Can AI Be Conscious? A Serious Answer Without the Sci-Fi
One camp says “expensive autocomplete”, the other says “it’s alive”. Both skip the question of what would actually have to be true — and one scientific theory answers it.
Fungi, Mycelium, and the Illusion of Individuality: Where One Organism Ends
“One organism” sounds like an obvious unit. Fungi break that obviousness: mycelium, symbiosis and mycorrhizal networks show the boundary of an individual is often a habit, not a fact.
Bioelectricity, Form, and the Body’s Memory: How a Body Knows What Shape to Be
The genome is a parts list, not a build plan. So what holds the template of a body’s shape and guides regeneration? Michael Levin points to a measurable, editable layer: bioelectric signals.
How to Read Bold Claims: The Case of Morphic Resonance
AI and Values: Why Alignment Is Not Just a Safety Problem
Scientific Dogma and New Ideas: When Skepticism Protects and When It Blocks
AI and Identity Over Time: Is an Agent With Memories Still the Same Agent?
Your agent got memory — it remembers projects, preferences, past mistakes. A year later, is it still the same agent, or someone else? The question sounds philosophical; the consequences are practical.
Science Is Fragmenting: Why Almost Nobody Sees the Whole Picture Anymore
Science has never known more — and never has anyone understood less of it whole. Specialization is both a triumph and a trap. On what is lost when fields stop talking.
Does AI Have Intentions? Intentionality, Goals, and Simulated Reasons
“The agent decided to…” — we say it without blinking. But does the agent have intent, or just generate behavior that looks like intent? And does it matter in practice? It does.
AI Architecture for Companies: When a Prompt Is Enough, When You Need RAG, When You Need an Agent
Agency and Accountability in AI Agents: When Does a Tool Start to Act?
Why Code Review Won’t Save Your Architecture If You Only Ask About Lines
AI and the Problem of Authority: Who Gets to Decide What Is True?
The source of knowledge is shifting from experts and institutions to the model. That is neither good nor bad — it is a transfer of authority. Delegate it blindly and you lose the thread.
Architecture Is the Shape of the Team: Why Code Boundaries Mirror Company Boundaries
An architecture diagram and a company org chart tend to look suspiciously alike. That is not a coincidence — it is Conway’s law. And while you treat it as trivia, it works against you.
What Does It Mean for a Model to Understand? Between Tokens, the World, and Useful Behavior
The “understands vs. parrots” debate is framed wrong. The useful question is what kind of understanding a model has, where it ends, and where it fails you.
Architecture Decision Records: The Team Memory That Saves Future Refactoring
Local Models vs Frontier Models: Where the Practical Limit Really Lies
Maintainable Code Is an Economic Property, Not an Aesthetic Preference
How to Read AI Papers Without Spending Your Life on arXiv
You cannot read everything, and that is not the goal. A workflow to triage papers, read them in layers, and tell when results are hype and when they are something you would ship.
Technical Debt Is Not Messy Code: How to Spot the Debt That Actually Costs Money
Messy code and technical debt are not the same thing. Real debt is a deliberate or structural gap between your code and what you now know — and to manage it you first have to classify it.