Claude Code , Codex, Sandboxes & The Future of AI Coding | Sanket Agarwal X Dylan Mikus (YC25)

AI coding agents are getting powerful. But where should they actually run? In this episode of AI with Cheff, I sit down with Dylan Mikus, Founder & CEO of Amika (YC F25), to talk about the infrastructure behind the next generation of AI agents. We go deep into Claude Code, Codex, AI sandboxes, cloud coding agents, software factories, multi-agent workflows and what happens when developers start running dozens of autonomous coding agents in parallel. Dylan explains why running everything locally eventually breaks down, why agents need their own computers, how sandboxes make autonomous coding safer and more scalable, and why he doesn't believe one cloud will win the AI agent infrastructure market. We also discuss: • Why AI agents need sandboxes • Claude Code vs cloud coding agents • Running multiple coding agents in parallel • What a software factory actually is • How agents can debug production issues automatically • Giving AI agents access to Sentry, Grafana, AWS and developer tools • Managing and orchestrating multiple agents • Why CI becomes the source of truth when agents write code • Why SSH still matters for AI infrastructure • The future of cloud infrastructure for AI agents • MCP and agent workflows • Why “context engineering” can become hype • Whether dark software factories are actually possible today • The future of AI coding • Devin, OpenAI, Anthropic and the coding-agent race • Why many AI frameworks may become unnecessary • What AI products are real vs overhyped Dylan is a two-time YC founder and is currently building Amika, infrastructure that lets teams run sandboxed coding agents across computers and clouds and build their own software factories. If you're a developer, AI engineer, founder or builder trying to understand where coding agents are heading, this conversation goes much deeper than another AI tools demo. Subscribe to AI with Cheff for conversations with the founders and engineers building the AI ecosystem, along with AI tutorials, demos, coding agents, Claude Code, Codex and the latest developments in AI. #AIAgents #ClaudeCode #AICoding 00:00 Intro 01:17 How Amika Started 02:19 Giving AI Agents Their Own Computers 04:05 What Is an AI Sandbox? 04:24 Codex vs Claude Code: Local vs Cloud 06:22 Why Different AI Workloads Need Different Computers 08:31 What Is a Software Factory? 11:09 Why Companies Will Build Their Own Internal Agents 13:15 How to Choose an AI Sandbox 16:18 Amika Demo 20:23 Running Multiple Coding Agents in Parallel 22:36 Managing Agents From Slack 23:53 Do We Need an Agent Orchestrator? 26:25 How Dylan Manages Multiple AI Agents 30:02 How Do You Stop AI Agents From Lying? 31:25 Reviewing AI-Generated Work 35:30 The Problems With Current Sandbox Providers 37:43 Why SSH Still Matters 40:12 Can Sandbox Startups Become Huge Companies? 43:38 Why AI Agent Infrastructure Is Changing So Fast 46:51 Loop Engineering, Graph Engineering & Context Engineering 48:45 Are Software Factories Actually Real? 51:28 What's Actually Working in AI? 52:29 Why Consumer AI Hasn't Happened Yet 54:08 Rapid Fire: AI Companies & Predictions 55:51 Are AI Frameworks Becoming Useless? 56:31 Will Most Coding-Agent Startups Die? 57:40 What Is the Cloud Getting Wrong? 58:40 Final Thoughts AI agents AI coding agents coding agents Claude Code OpenAI Codex Codex Anthropic Claude AI sandbox AI sandboxes cloud coding agents AI agent infrastructure agentic AI software factory AI software factory autonomous coding AI software engineering AI developer tools AI engineering AI infrastructure multi agent systems agent orchestration Claude Code tutorial Codex cloud AI coding vibe coding MCP Model Context Protocol AI workflows AI automation future of coding future of software engineering Y Combinator YC founder AI startup AI founder podcast AI podcast Dylan Mikus Amika