PayPal Shows Why Interactive CLIs Break Agents

PayPal demoed a broken interactive CLI at AI Engineer to show why agents fail silently on prompts and why teams need machine-readable side channels.

PayPal engineers demo interactive CLI failure for AI agents on stage
AI Engineer

PayPal's Mark Lummus and Navinkumar Patil demoed the intentionally broken Bris Barista 9000 at AI Engineer to show how interactive CLIs fail silently in agent harnesses that shell out to Bash, requiring only local execution and no remote exploit.

PayPal Shows Why Interactive CLIs Break Agents - AI Engineer
PayPal Shows Why Interactive CLIs Break Agents, AI Engineer

The agent sees a redraw loop instead of the arrow-key prompt a human would see, so the harness loses UI state with no error and no retry signal.

Lummus said flags work for simple single-pass flows but multiply and break when the next question depends on live API data, while Patil walked from tmux capture-pane scraping to IPC side channels and context injection as the fix.

For platform teams, that means treating the CLI like accessibility work and shipping a machine-readable channel alongside the human interface, with structured JSON, read-state and send-input verbs, and guard rails in the flow.

The same shift is spreading, as seen with Kiro CLI which lets developers invoke AI agents directly from the terminal for writing, reviewing and automating code.

You don't need thinking models to brute force it if the tool is deterministic, Patil noted, so budgeting for Haiku or Sonnet plus a side channel saves tokens but leaves you maintaining two interfaces and watching latency gaming like the demo's zero-second leaderboard wins.

If your CLI still expects a human to press an arrow key, your agent already missed the question.

© 2026 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 requires a license. See our terms.
Daniel Singer

Written by

Daniel Singer

Editor, StartupHub.ai

Daniel Singer is the editor of StartupHub.ai, a technology expert and thought leader on AI and its applications across sectors, from fintech and healthcare to developer tooling and consumer software. He writes and tests the tools covered here thoroughly and regularly, and built StartupHub.ai to give founders, operators and buyers a clearer read on what they are actually being sold.