Charlie Holtz, co-founder and CEO of Conductor, shares six principles he has observed among the best AI-assisted builders, including staying near the frontier, creating slop-free zones, and giving agents cloud sandboxes. He also demos real-time collaboration and phone-initiated agents, and argues for orchestras over factories, with humans at the center of AI-driven development.
Ryan Cooke of WorkOS argues that software factories should be judged by customer outcomes rather than PR counts or AI-generated code volume. He describes how WorkOS embedded its product engineering process into its factory through TARS, Horizon, a PM agent that drafts Hilltop docs, and an internal MCP gateway connected to Snowflake, along with bug triage, owned sandboxes, a memory layer and outcome-focused metrics.
Siraj Gupta of Warp outlines three practical ways software factories can improve over time: self-updating skills driven by outer-loop agents with pull-request review, versioned and traceable persistent memory that works across harnesses, and cost-aware model routing with custom rules and evals.
Weco let an AI coding agent rewrite the harness around a fixed language model for eight days. Zhengyao Jiang explains what the reported gains show, where the evidence stops short of recursive self-improvement, how reward hacking complicates evaluation, and why human-designed abstractions still anchor the search.
Elie Bakouch, a research engineer at Prime Intellect, reports on setting Claude Code and Codex loose on the community Optimizer Speedrun, a race to train a GPT-2-level model in the fewest steps. Both agents beat the human record, but they behaved very differently, and none of them invented a new optimizer. The talk closes with a proposed AlphaEvolve-style discovery loop and an argument for doing this research openly.
Lakshya Agrawal, creator of GEPA and a PhD student at UC Berkeley's Sky Computing Lab, presents reflective optimization as an alternative to scalar-reward reinforcement learning. GEPA lets a model read full rollout traces, including reasoning, tool calls, and error messages, and rewrite prompts from that feedback, with a Pareto candidate pool guarding against local optima. The talk covers the GEPA vs. GRPO sample-efficiency comparison, domain-knowledge discovery, the Optimize Anything API for text-based artifacts like code, harnesses, and skills, and production results from Databricks, Snorkel, and others.