Building an AI Platform
summary
Builder describes a six‑year journey that launched dozens of AI products before generative AI, starting from half a dozen disparate projects and an internal assistant called Ava VP of culture. Early platform choices — a distributed event‑driven microservices stack with MongoDB, Python, Java and later Go plus Kafka and Kubernetes — create maintenance and performance chaos, leading to repeated refactors. Platform 6.0 stable ships in 2024 after 6–7 months beta, exposes a clear API and reusable agents, and deliberately avoids LLMs for orchestration using them only for messaging and parsing.
AI-written summary · about
chapters (17)
- 0:00Building an AI platform
- 1:09Product experimentation early days
- 2:33Data value and guiding principles
- 5:01Proactive AI with user knowledge
- 7:57Pivoting to successful AI projects
- 9:00Mistake building Mindsmith 1.0
- 11:57Over-engineering with microservices
- 13:22Multi-language tech debt
- 15:09Performance issues and GoLang
- 16:33Platform 1.5 enterprise
- 17:19Human-in-the-loop prototyping
- 18:25Automating Docker and Kubernetes
- 19:09Scaling challenges and business
- 20:14Developer feedback on platform
- 21:40Refactoring and platform 1.5
- 22:49No LLMs for orchestration
- 23:35Platform 6.0 stable release
related talks
28mTransforming Internal Developer Platforms with AI: Accelerating Innovation
34mUnlock the full potential of Generative AI with Agent Platforms
17mAI-Driven Platform Engineering
25mAI for Platform Engineering
27mTransforming Developer Platforms with Generative AI
40mOpening Keynote: AI & Platform Teams
28mSoftware 3.0 and The Emerging AI Developer Landscape
24mPlatform Engineering and AI - Two Buzzwords Finally Meet!