← back
Designing AI-Intensive Applications - swyx
Takeaway
AI engineering still lacks a canonical architectural pattern; evals, orchestration, and security are where production work lives beyond commoditized LLM calls.
Summary
- swyx's conference opener proposes finding the 'standard model' for AI engineering — analogous to ETL, MVC, MapReduce — beyond just RAG
- Candidate models discussed: Karpathy's LLM-OS (updated with MCP and multimodality), the LLM-SDLC where evals/security/orchestration are where customers actually pay
- References Anthropic's 'building effective agents', OpenAI's swarm-style agents SDK, and his own descriptive top-down agent taxonomy
- Argues AI News (a workflow, not an agent per Soumith) still delivers value, so the field should optimize for value over arguable terminology
ai-engineeringsdlcagents
Original description
Whether you call it a workflow or an agent, AI engineered applications are seeing user-input:LLM-call ratios go from 1:1 (ChatGPT) to 1:100 (Deep Research, Codex) and even 0:n (Ambient/Proactive agents). How does AI Engineering change as you build increasingly AI intensive applications? Recorded at the AI Engineer World's Fair in San Francisco. Stay up to date on our upcoming events and content by joining our newsletter here: https://www.ai.engineer/newsletter Timestamps: 00:00 Conference Welcome and Overview 00:42 Conference Logistics and Growth 01:47 Audience Preferences and Survey 02:22 Innovations in AI Engineering (MCP and Chatbots) 02:58 Evolution of AI Engineering (Past Talks) 03:50 Simplicity in AI Engineering 04:17 AI Engineering as a Developing Field 05:23 Seeking the "Standard Model" in AI Engineering 06:02 Candidate Standard Models in AI Engineering 09:26 Human Input vs. AI Output (AI News Example) 11:05 SPADE Model for AI-Intensive Applications 12:29 Call to Action for Conference Attendees