Dev Interrupted It’s Tuesday and your tech stack is obsolete (again). Now what? | Theory Venture’s Bryan Bischof
23 snips
May 12, 2026 
Bryan Bischof, Head of AI at Theory Ventures and applied AI educator with a PhD in math. He unpacks inference levers beyond model size, why prompt optimization often hits diminishing returns, and how to design high-performing data agents and investor-facing AI systems. They also riff on hype cycles, rip-grep satire, and a playful data-agent game show coming soon.
AI Snips
Chapters
Transcript
Episode notes
Inference Has Many Hidden Optimization Levers
- The inference layer has many underappreciated levers beyond just bigger models or GPUs.
- Bryan maps levers like quantization, KV caches, optimizer choices, and serving vs training optimizations onto Charles Fry's cost-of-intelligence chart.
Run Learning Like A Research Project
- Design conferences and learning like a research project: identify gaps, ask experts, then hunt missing threads.
- Bryan curated his AI Council track by mapping what smart people didn't fully understand and recruiting specialists to fill those holes.
Agent Ergonomics Unlock Agent-Only Superpowers
- Designing for agents differs from designing for humans and can unlock agent-only capabilities.
- Bryan calls this 'agent ergonomics' and gives search and task-harnesses as examples where agents can outperform humans with different interfaces.