Episodes (Page 5)
✨
Ryan Cooke questions the 'software factory' model if outcomes aren't improved.
✨
Tereza Tížková defines a software factory as the autonomous, full software lifecycle.
Tereza Tizkova
✨
Suraj Gupta details three ways software factories can improve themselves.
✨
Google's Antigravity built an OS kernel running Doom using 93 subagents in 12 hours.
✨
Software engineering is shifting towards 'factory engineering' principles.
Zach Lloyd
✨
GLM-5.2, an open-weights model, rivals top proprietary models on coding benchmarks.
✨
Charlie Holtz prefers 'orchestras' over 'factories' for fast software building.
✨
Andrew Orobator argues judgment, not the model, is the bottleneck in AI development.
Andrew Orobator
✨
Ido Salomon likens managing coding agents to a strategy game like Warcraft.
Ido Salomon
✨
The agent loop itself is the product, not just the agent.
Roland Gavrilescu
✨
Claude Code and Codex raced to train a GPT-2 model, beating human records.
Elie Bakouch
✨
GEPA uses reflection on full traces, not just scores, to improve RL.
Lakshya A. Agrawal
✨
AI agents in games struggle with long-term memory and misinformation.
Erina Karati
✨
Weights & Biases' ARIA agent now helps build itself.
Zubin Aysola
✨
Morph used autoresearch to make models 3x faster using custom GPU kernels.
Tejas Bhakta
✨
Weights & Biases' ARIA agent runs autoresearch experiments live.
✨
DoorDash uses LLMs for semantic understanding in search.
✨
Spotify's Large Taste Model uses LLMs for generative personalization.
✨
Content feeds are cheaper to run than AI chat apps.
✨
Google DeepMind, ElevenLabs, Hugging Face, and Stripe presented technical talks.