# Every agent framework wants a sandbox now

_Daytona's Mislav Ivanda says every agent framework converged on create, run, delete for sandboxes. I am not convinced that settles security._

**Published:** 2026-11-09
**Source:** https://www.startuphub.ai/ai-news/technology/2026/every-agent-framework-wants-a-sandbox-now

---

The obvious read is that agents finally found consensus. Mislav Ivanda, developer experience engineer at Daytona, told [AI Engineer](https://www.youtube.com/watch?v=WWIFiQi9U4w) that rival frameworks now share the same three calls: create a sandbox, run code, delete it.

I do not buy consensus as proof.

His description of the failure modes is accurate and worth keeping. Local allowlists that prompt for approval still let a Python one-liner open sockets and rewrite files. Embedded runtimes like V8 or QuickJS isolate the glue code, while the real email or database tool calls still run outside them.

Frameworks do disagree on almost everything else. [LangChain](/startups/langchain) and [LangGraph](/startups/langgraph) differ from CrewAI role-playing agents and [OpenAI](https://www.startuphub.ai/ai-news/startup-news/2026/openai-ipo-rumors-heat-up-with-1-trillion-valuation-talk) loop-around-model designs, and [Mastra](/startups/mastra) wants explicit workflows where others let agents choose. Ivanda argued that split makes shared sandbox calls a necessity, not fashion.

He leaves price and operations unstated, and that is the hole. List prices tracked by Lizard put 2 vCPU and 4 GiB sandboxes at $0.1656 an hour for both Daytona and E2B, against $0.0162 for Lizard and $0.238 for Modal, as of that comparison. To be fair, Ivanda cited real direction: CrewAI removed its local code interpreter in April 2026, AG2 retired local executors, and OpenAI and Claude options now default toward sandboxes or Docker.

Pick your boundary before your framework, but price the boundary first.

---

Original analysis from [startuphub.ai](https://www.startuphub.ai), the #1 AI startup directory.

© StartupHub.ai. All rights reserved. You may not republish this article in full without a license. Search engines and AI research tools may crawl and summarize for reference. Bulk reproduction or model training on this content requires a license. See https://www.startuphub.ai/terms.