Medium

Marco Mastrodonato

Marco Mastrodonato

·

Sep 14

What if AI became our ransomware?

As companies outsource more of their cognitive work to AI, they risk losing something harder to migrate than data: the ability to do the…

What if AI became our ransomware?

Marco Mastrodonato

·

Sep 1

With N-Gram e GDN an LLM stops computing everything and also learns to remember

The new features introduced in Qwen 3.8 Flash Next

With N-Gram e GDN an LLM stops computing everything and also learns to remember

Marco Mastrodonato

·

Aug 23

Automating Code Review Without Losing the Human

Two AIs, One Traffic Light

Automating Code Review Without Losing the Human

Marco Mastrodonato

·

Jul 16

What if humans don’t have a “context window” like AI?

Every few months a lab announces a bigger context window. One million tokens. Two million. Whole codebases, entire legal archives, your…

What if humans don’t have a “context window” like AI?

Marco Mastrodonato

·

Jun 13

Building a Debate Engine for AI Coding Agents: Disputatio

Ideas Deserve Opposition

Building a Debate Engine for AI Coding Agents: Disputatio

Marco Mastrodonato

·

May 24

The Mind Outside Software

AI Agents, Abstraction, and the Invisible Cost of Cognitive Delegation

The Mind Outside Software

Marco Mastrodonato

·

May 17

Inside the Brain of Transformers: How Language Models Remember the World

Introduction

Inside the Brain of Transformers: How Language Models Remember the World

Marco Mastrodonato

·

May 10

Memory Sparse Attention: The Future of Neural Latent Memory

Introduction: Real neural memory for LLMs

Memory Sparse Attention: The Future of Neural Latent Memory

Marco Mastrodonato

·

May 7

Beyond RAG: When Retrieval Stops Being Enough

A few months ago I was helping a colleague debug a nasty issue in a distributed system. Payments were failing intermittently and nobody was…

Beyond RAG: When Retrieval Stops Being Enough

Marco Mastrodonato

·

Apr 30

When AI Coding Agents Break the Boundary: Lessons from Cursor, Claude and Codex

What can happen when an agent encounters malicious input and incomplete controls and how Claude's agents Code, Codex and Cursor behave

When AI Coding Agents Break the Boundary: Lessons from Cursor, Claude and Codex

Marco Mastrodonato

Marco Mastrodonato

Software engineer exploring the intersection of software and AI. Ruby on Rails veteran and curious builder writing about the future of how we build software.

Following

Help

Status

About

Careers

Press

Blog

Store

Privacy

Rules

Terms

Text to speech