"The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis Infinite Code Context: AI Coding at Enterprise Scale w/ Blitzy CEO Brian Elliott & CTO Sid Pardeshi
343 snips
Feb 5, 2026 
Sid Pardeshi, Blitzy CTO and systems engineer, and Brian Elliott, Blitzy CEO building autonomous code generation, explain their “infinite code context” approach. They discuss dynamic agent architectures and model selection, large-scale code ingestion and schematized knowledge graphs, parallel testing and runtime-backed validation, pricing and paths to near-complete autonomy, and why the last 20% still needs humans.
AI Snips
Chapters
Transcript
Episode notes
Mix Model Families For Robustness
- Use multiple model families (OpenAI, Google, Anthropic) and have dissimilar models review each other's work.
- Assign roles: some models for first-pass generation, others for structured review and long-horizon planning.
Memory Over Fine-Tuning
- Fine-tuning is a last-mile optimization and often loses generality as new baseline models improve.
- Blitzy is more bullish on memory systems (application-level, persistent preferences) than on fine-tuning models.
Store Enterprise Memory In The System
- Enterprise memory must live at the application/system layer because many preferences are locally specific.
- Store decisions, traces, and preferences in the enterprise instance instead of compressing them into model weights.