Uploaded March 2026 | Updated September 2026, 22 hours ago
At Callstack x Vercel AI SDK Meetup Szymon Chmal explains how to design AI systems that are safe and reliable in production using Vercel AI SDK, Workflows, and Sandbox.
This talk focuses on three engineering problems teams hit early when building AI-powered products:
- avoiding model vendor lock-in
- handling long-running AI workflows
- executing generated code safely
Using a practical file-processing example, Szymon shows how the Vercel AI SDK, Workflows, and Sandbox can help teams build systems that are flexible, durable, and safer to run in production.
You’ll hear about:
- switching between model providers without rewriting your app
- structuring durable workflows with retries and human approval steps
- isolating generated code in a sandboxed runtime
- why system prompts are not a real security boundary
- how to think about production AI architecture beyond demos
Follow Callstack for more talks on React Native, AI, and production engineering: https://x.com/callstackio
At Callstack x Vercel AI SDK Meetup Szymon Chmal explains how to design AI systems that are safe and reliable in production using Vercel AI SDK, Workflows, and Sandbox.
This talk focuses on three engineering problems teams hit early when building AI-powered products:
- avoiding model vendor lock-in
- handling long-running AI workflows
- executing generated code safely
Using a practical file-processing example, Szymon shows how the Vercel AI SDK, Workflows, and Sandbox can help teams build systems that are flexible, durable, and safer to run in production.
You’ll hear about:
- switching between model providers without rewriting your app
- structuring durable workflows with retries and human approval steps
- isolating generated code in a sandboxed runtime
- why system prompts are not a real security boundary
- how to think about production AI architecture beyond demos
Follow Callstack for more talks on React Native, AI, and production engineering: https://x.com/callstackio










