Emulated: Training Fully Autonomous Software Engineers in Full-Company Simulations
AI Engineer · youtube · 2026-08-01
Joseph Wang argues that current post-training environments fail to capture the complexity of real production. A genuine engineering task isn't a simple code diff; it involves 50 to 100 turns of complex troubleshooting amid sudden node failures and config conflicts.
To solve this, his team built Emulated, which simulates whole companies in high fidelity. It lets agents act as cloud engineers, provisioning resources across VPCs and security groups, managing costs, and keeping services online. The core thesis is that combining deep domain expertise with faithful simulation is the only way for agents to learn the messy, end-to-end reality of infrastructure work.
More from coding & agent
- Beyond One-Shot Prompts: The Real Value of AI Agents Lies in Underlying Context — thisiskp_ · 2026-08-01
- Developer Builds Automated Music Discovery Tool Using ChatGPT Sites Plugin — simpsoka · 2026-08-01
- The Hidden Cost of Coding Agents: Developers Are Losing Grip on Their Code — MarcJSchmidt · 2026-08-01
- Using Codex to Improve Voice-to-Text: Auto-Extracting Custom Vocabularies — jdjohnson · 2026-08-01
- Replace Middle Management with AI Agents to Automate Tactical Tracking — ycombinator · 2026-08-01
- Benchmarking Agent Harnesses: Kimi K3 Shines, Claude Code Costs 4x More — omarsar0 · 2026-08-01