Artificial Intelligence
The Human Who Wages War Is the Same One Who Calls AI a Threat
Does the real risk lie in artificial intelligence, or in the human intentions guiding its use? A reflection grounded in hands-on experience with AI agents.
Author
Maycon Menezes
Software Engineer & Architect · Founder of MnzAI Labs
Maycon works across AI engineering, software architecture and developer tools. MnzAI Labs is where he publishes practical experiments, product work and technical analyses.
Published articles
Cybersecurity
Binance and Mercado Bitcoin attacks: what happened and the human factor
A fake interview, compromised credentials and billions on the move. What do the Binance and Mercado Bitcoin attacks reveal about the engineering of trust?
AI Engineering
Choosing GitHub Copilot models without wasting credits
A cheaper call only helps when rework does not consume the savings. A practical view of where I would use Luna, Haiku, Terra, Sol, Sonnet and Opus.
AI & Society
Stop fearing AI. Start using it.
There is more risk in staying out through passivity and comfort than in becoming useless. What do we do when AI expands a team's capacity?
AI & Society
AI and Privacy: what ChatGPT, Gemini, Claude, and Grok really do with your data
Not everything you tell an AI becomes training data, but there is no zero-risk cloud service. Here is what the official policies actually say.
AI Engineering
AST and local parsing: when an agent stops reading code like a novel
Code is more than text with semicolons. See how local parsing turns files into navigable structure — and where this approach still falls short.
AI Engineering
Graph Engineering: when multiple agents need a map, not a group chat
Adding agents does not automatically create a team. Graph Engineering treats nodes, relationships, state, and transition criteria as executable architecture.
AI Engineering
Harness Engineering: the model is the engine, but someone must build the car
Changing the model does not always fix an unstable agent. What may be missing is the harness: an environment that makes work legible, bounded, and verifiable.
AI Engineering
Static prompt caching: stop paying the AI to reread the same manual
Does your agent receive the same instructions and tools on every call? A reusable prompt prefix can save time — without caching responses.
AI Engineering
Claude Code: why can a simple task use so many tokens?
A small diff, but a costly session? See what Claude Code carries into each request and how to manage context, caching, tests, and subagents.