JNKINDILOGS

AI & Development

A Skill Can Hijack Your Agent Without Executing a Single Line of Code

📅 May 14, 2026 ⏱ 2 min read ✍️ Jacques Nyilinkindi

A SKILL.md is plain text injected into your agent's context window. Here's how a malicious one can hijack your agent—and what we built to stop it.

AI Security Context Steward Agentic AI

Read Full Article →

AI & Development

Prompt Injection Is the SQL Injection of LLMs—And Your CLAUDE.md Is the Attack Vector

📅 March 19, 2026 ⏱ 8 min read ✍️ Jacques Nyilinkindi

Any untrusted text the model reads can hijack it. CLAUDE.md files pushed in a PR are the new attack surface. There's no architectural fix yet—only mitigations.

Prompt Injection AI Security Claude Code

Read Full Article →

AI & Development

When 'We Own the Model' Beats the API: A Self-Hosting Playbook

📅 March 12, 2026 ⏱ 8 min read ✍️ Jacques Nyilinkindi

Llama, Qwen, DeepSeek, Mistral on your own infrastructure—when it pays off, when it doesn't, and how to wire the result into your existing OpenAI-shaped code with minimal lift.

Open Weight Llama Qwen

Read Full Article →

AI & Development

Small Models, Close to the Metal: SLMs Are the Default You're Missing

📅 February 21, 2026 ⏱ 8 min read ✍️ Jacques Nyilinkindi

Most production AI features don't need GPT-4. They need a 3B model running on the user's laptop. The economics of SLMs at the edge are unreasonable—if you actually deploy them.

SLM Edge Inference Ollama

Read Full Article →

AI & Development

Stop Defaulting to One Model: Task-by-Task Routing Across the Big Four

📅 February 2, 2026 ⏱ 8 min read ✍️ Jacques Nyilinkindi

Most production LLM stacks waste 60% of spend by sending every task to the biggest model. A routing strategy that picks the right model per task fixes that—without sacrificing quality.

AI Engineering Model Routing LLM Cost

Read Full Article →

AI & Development

MCP Servers: Extending AI Capabilities Through Standardized Context

📅 January 20, 2026 ⏱ 9 min read ✍️ Jacques Nyilinkindi

Understanding Model Context Protocol (MCP) servers and how they're transforming AI integration by providing standardized interfaces for extending AI capabilities with external tools and data sources.

MCP AI Claude

Read Full Article →

AI & Development

AI Development Tools 2025: Claude vs. Gemini for Engineering Teams

📅 December 15, 2025 ⏱ 9 min read ✍️ Jacques Nyilinkindi

Comprehensive comparison of Claude and Gemini AI tools from an engineering management perspective—evaluating code generation, reasoning capabilities, cost, and production readiness.

AI Claude Gemini

Read Full Article →

AI & Development

Hallucinations Are Architectural, Not Bugs: Manage Them, Don't Fix Them

📅 December 15, 2025 ⏱ 8 min read ✍️ Jacques Nyilinkindi

Next-token prediction has no 'I don't know' head. You manage hallucinations with retrieval, verification, and refusal—not by waiting for a bigger model. The fix is architecture.

Hallucinations RAG LLM Architecture

Read Full Article →

Software Architecture

Clean Architecture: Building Maintainable Software Systems

📅 December 10, 2025 ⏱ 6 min read ✍️ Jacques Nyilinkindi

Exploring the principles and practices of Clean Architecture to build scalable, testable, and maintainable software systems that stand the test of time.

Clean Architecture Software Design Best Practices

Read Full Article →

Software Architecture

Migrating to Microservices: A Strategic Engineering Approach

📅 September 22, 2025 ⏱ 8 min read ✍️ Jacques Nyilinkindi

A comprehensive guide to successfully migrating from monolithic architectures to microservices, covering strategy, patterns, and common pitfalls from real-world implementations.

Microservices System Design DevOps

Read Full Article →