01
Deep Agents Pattern: Planner, Files, Subagents 2026
The deep agents pattern behind Claude Code and Deep Research: a planner, filesystem, isolated subagents, and memory. Plus the 15x token cost to plan for.
Sebastian Mondragon
CEO & Founder
AI System ArchitectureLarge Language ModelsRAG SystemsAI Agent DevelopmentEnterprise AI Strategy
CEO & Founder
AI systems architect and founder of Particula Tech, building production-ready AI solutions for enterprises worldwide.
AI System ArchitectureLarge Language ModelsRAG SystemsAI Agent DevelopmentEnterprise AI Strategy
About
Sebastian Mondragon is the founder and CEO of Particula Tech, an AI consulting and development company that has delivered over 600 AI solutions to 450+ clients worldwide since 2023.
With deep expertise in building production-ready AI systems, Sebastian specializes in architecting solutions that handle real-world scale and complexity. His approach combines rigorous engineering discipline with practical business understanding, focusing on AI implementations that actually ship to production rather than remaining proof-of-concepts.
Before founding Particula Tech, Sebastian worked extensively on enterprise software systems, developing a strong foundation in building reliable, scalable applications. This background shapes his philosophy that AI systems need the same engineering rigor as any other production software: proper architecture, comprehensive testing, monitoring, and maintenance plans.
Sebastian writes regularly about AI development best practices, from technical deep-dives on RAG systems and AI agents to strategic guidance on when and how businesses should adopt AI. His writing reflects hands-on experience solving real problems for clients across manufacturing, professional services, healthcare, and retail industries.
Articles by Sebastian (184)
02
Self-Hosted vs API Embeddings: Qwen3, Voyage 2026
Qwen3-Embedding-8B tops MTEB multilingual at 70.58 and self-hosts at ~1/20th API cost. A decision framework for self-hosted vs API embeddings in 2026.
03
DSPy GEPA vs MIPROv2: Auto Prompt Optimization 2026
GEPA beats MIPROv2 by over 10% and outperforms GRPO with up to 35x fewer rollouts. Here's how DSPy's reflective prompt optimizer actually works in 2026.
04
AI Agent Pricing: Per-Seat vs Outcome-Based 2026
Seat-based AI pricing fell from 21% to 15% in a year. The 2026 buyer framework for evaluating per-seat, per-resolution, and outcome-based agent pricing on real TCO.
05
Slopsquatting: AI Package Hallucination Rates 2026
Open models hallucinate package names 21.7% of the time vs 5.2% for commercial. The data, the react-codeshift incident, and how to gate AI deps.
06
Spec-Driven Dev Tools: Spec Kit vs Kiro vs Tessl
Spec Kit hit 111k stars and 30+ agents, but Kiro ships 3 files and Tessl makes the spec the source. A 2026 framework for picking the right SDD tool.