Page 104
12 articlesCompound AI Systems: Why Your Best Architecture Uses Three Models, Not One
Production AI systems that compose a classifier, generator, and verifier consistently outperform single frontier models — delivering higher accuracy at lower cost, as long as coordination overhead stays below the 40% latency threshold.
Database-Native AI: When Your Postgres Learns to Embed
PostgreSQL extensions like pgvector and pgai now handle embedding generation, vector search, and LLM calls inside the database — eliminating the sync pipeline most RAG architectures carry and keeping vectors transactionally consistent with source data.
The Death of the Glue Engineer: AI Is Absorbing the Work That Holds Systems Together
AI agents are rapidly automating the integration work — ETL pipelines, API adapters, webhook handlers — that glue engineers built careers on. Here's what falls first, what remains human-essential, and how to move up the stack before the implementation layer disappears.
Debug Your AI Agent Like a Distributed System, Not a Program
Print statements and flat logs fail for multi-step AI agents. Structured tracing, deterministic replay, and the replay-diverge-compare methodology bring distributed systems debugging to agent workflows.
Edge LLM Inference: When Latency, Privacy, or Cost Force You Off the Cloud
A fine-tuned 7B model on one GPU can beat GPT-4 in narrow domains at zero marginal token cost. A practical guide to hardware sizing, quantization formats, hybrid local-cloud routing, and the deployment frameworks that make edge LLM inference production-ready.
The Inference Gateway Pattern: Why Every Production AI Team Builds the Same Middleware
The inference gateway is an emergent architectural pattern — a middleware layer between applications and LLM providers that consolidates rate limiting, failover, cost tracking, and routing. A practical guide to why every production AI team converges on this pattern and how to build or buy one.
Internal AI Tools vs. External AI Products: Why Most Teams Get the Safety Bar Backwards
Internal AI tools often need more safety engineering than customer-facing products — but a completely different kind. How ambient authority, silent failures, and data synthesis across classification boundaries make internal deployments the higher-risk bet.
Knowledge Graphs Are Back: Why RAG Teams Are Adding Structure to Their Retrieval
Baseline RAG captures only 22-32% of multi-hop answers while GraphRAG achieves 72-83%. A practical guide to adding knowledge graph structure to your retrieval pipeline — construction patterns, routing strategies, and when the schema overhead isn't worth it.
LLM Provider Lock-in: The Portability Patterns That Actually Work
Most LLM lock-in advice stops at API wrappers, but the real lock-in hides in prompts, tool-calling assumptions, and behavioral quirks. Portability patterns that address what abstraction layers cannot.
The MCP Composability Trap: When 'Just Add Another Server' Becomes Dependency Hell
The MCP ecosystem hit 10,000+ servers and 30 CVEs in sixty days. How dependency sprawl, supply chain attacks, and tool conflicts turn composability into a liability — and the operational patterns that prevent it.
Open-Weight Models in Production: When Self-Hosting Actually Beats the API
A practical decision framework for self-hosting open-weight models like Llama, Mistral, and Qwen versus using frontier APIs — covering real cost breakdowns, compliance triggers, operational burdens, and the hybrid architecture most production teams actually need.
The Post-Framework Era: Build Agents with an API Client and a While Loop
Why 80% of production AI agents need nothing more than a prompt, a tool list, and a while loop — and how framework complexity becomes the bottleneck it promised to eliminate.
About Tian Pan
I'm Tian Pan, an engineer-founder focused on agentic engineering — building autonomous AI systems and scaling engineering teams. I write practical guides on system design, technical leadership, and shipping with AI agents. Previously an early engineer at Uber, Brex, and IoTeX.
