Running production agents on sandboxed CLIs: lessons from iterating on agent-facing interfaces
nguyenfamjj · reddit · 2026-09-05
A developer building production-grade agent workflows shares how the team moved execution to CLIs running inside sandboxes, remapping their API gateway into an agent-friendly command-line interface. They argue MCP remains useful for bounded environments, but sandbox + CLI offers broader capability and better product-surface efficiency.
Iteration lessons that made the CLI agent-friendly:
- Task-oriented commands, not thin endpoint wrappers;
- Inspect and dry-run modes before changes;
- -- for reliable parsing, readable defaults for humans;
- Errors that explain recovery and suggest the next action;
- A simple feedback path when the interface blocks an agent.
The author invites discussion on when to use MCP vs. a sandboxed CLI.
More from coding & agent
- GLM-5.3 launches with vision on Baseten, Terminal-Bench 3.0 jumps 4.6% to 28.3% — baseten · 2026-09-05
- Swarms v15 rebuilds multi-agent execution with 10+ MCP tutorials — KyeGomezB · 2026-09-05
- Google GenAI SDK for Kotlin hits 1.0: idiomatic multiplatform access to Gemini — rseroter · 2026-09-05
- Cross-Model Code Review: Having Claude and Copilot CLI Battle Over Refactoring — DanWahlin · 2026-09-05
- Developer Uses Claude Code to Ship a Working F-Zero X Port to 3DS at Near 60fps — killermike523 · 2026-09-05
- Coinbase's x402 protocol replaces 700+ API keys with a single wallet signature for AI agents — kleffew94 · 2026-09-05