1,000 Agent Tasks in a Sandbox: What Breaks When LLMs Run Code — Kevin Orellana, AWS
Kevin Orellana, an engineer on Amazon Bedrock AgentCore, wrapped AgentCore's code sandbox and cloud browser in an MCP server and had a coding agent write and run 1,000 tests across 10 domains, logging results to SQLite.
Proof1 independent outlet
Key points
- We pointed a coding agent at our own sandbox overnight and told it to break things.
- NEXT UP: AI ENGINEER CODE SUMMIT
- • Using a coding agent plus MCP to stress-test your own platform
- • How sandbox limits decide whether an agent can finish its task
Sources (1)
- [1]1,000 Agent Tasks in a Sandbox: What Breaks When LLMs Run Code — Kevin Orellana, AWSAI Engineer (YouTube) · Oct 12, 05:36 AM
Kevin Orellana, an engineer on Amazon Bedrock AgentCore, wrapped AgentCore's code sandbox and cloud browser in an MCP server and had a coding agent write and run 1,000 tests across 10 domains, logging results to SQLite.
We pointed a coding agent at our own sandbox overnight and told it to break things.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 12, 2026Inside the AI Economy: What Stripe's Data Reveals — Nilofer Rajpurkar, Stripe
- Oct 12, 2026Autoregressive Decoding as a Martingale: Characterizing the Stochastic Dynamics of LLM Generation
- Oct 11, 2026SonarQube + OpenAI: Agentic Development — Killian Carlsen-Phelan, Sonar
- Oct 11, 2026Agent Speedrun — Elizabeth Fuentes Leone & Sandhya Subramani, AWS
- Oct 9, 2026Which Metadata Matters for LLM Tool Selection in MCP? Evidence from Controlled Factor-Level Interventions
- Oct 9, 2026Not All Correct Fixes Run Alike: An Empirical Study of Patch Performance in Repository-Level Issue Resolution
