Beyond Bolt-On AI: A New Way to Build Software
summary
Argues that the legacy OLTP/OLAP split and SaaS AI APIs force moving data to algorithms and cap automation at 5–10% automation; cites chips $626B, servers/cloud $930B, foundational models $25B and trillion‑scale GVA potential. Describes a document‑processing system that handles millions of docs with a human‑in‑the‑loop bounding‑box workflow to achieve business‑process accuracy and capture training data, notes >50% annotation time for CV work and 1.66M Hugging Face models, and demonstrates knowledge graph + LLM steering to avoid costly fine‑tuning while converting natural language to SQL/SPARQL.
AI-written summary · about
chapters (12)
- 0:00AI evolution overview
- 3:40AI software challenges
- 5:46OLTP vs OLAP problem
- 9:02SAS AI limitations
- 11:11Human-in-the-loop example
- 13:40AI maturity models
- 15:49Smart software architecture
- 19:23Beyond RAG to planning
- 22:35Integrating knowledge graphs with language models
- 24:44NLP interfaces & agents
- 27:53Evolvable AI system
- 30:02Software evolution to AI
related talks
30mAI and API, made for each other
27mAI as a Service Scaling AI with Ease and Innovation
16mHow can leaders turn operational struggles into strategic value through the integration of AI?
11mSuccessful AI adoption
13mAI as a Strategic Asset: Building Competitive Advantage
09mAutomating cognitive tasks to increase business value
20mGetting business value from AI: Everyday AI, extraordinary people
13mDesigning AI-Intensive Applications