PayPal's Mark Lummus and Navinkumar Patil led a workshop arguing that agents fail silently on interactive terminal prompts, and that adding CLI flags is the wrong fix. Participants adapted Barista 9000, a coffee-ordering CLI, for agent use, with three suggested techniques and a scoring competition.
A developer is using agent experience (AX) testing to improve a CLI tool's compatibility with different AI models. By drafting 30-50 use cases with Claude Opus and running them on Haiku, GPT-5.4 mini, and GPT-5.6 Luna, they identify model-specific quirks like invented flags or working directory issues that unit tests miss.
A developer is using agent experience (AX) testing to improve a CLI tool's compatibility with different AI models. By drafting 30-50 use cases with Claude Opus and running them on Haiku, GPT-5.4 mini, and GPT-5.6 Luna, they identify model-specific quirks like invented flags or working directory issues that unit tests miss.
A CLI tool designed to be used by both humans and AI agents, with considerations for agent-specific constraints like zero runtime dependencies, headless authentication, and predictable output. It solves the problem of building command-line interfaces that work reliably when executed by LLM-based agents via bash tools.
A CLI tool designed to be used by both humans and AI agents, with considerations for agent-specific constraints like zero runtime dependencies, headless authentication, and predictable output. It solves the problem of building command-line interfaces that work reliably when executed by LLM-based agents via bash tools.
Headless Terminal is a skill that enables AI agents to interact with interactive terminal user interface (TUI) and command-line interface (CLI) programs. It solves the problem of agents being unable to use tools that require interactive input, by providing a headless interface. This is useful for developers and AI enthusiasts who want to automate tasks involving such programs.
Headless Terminal is a skill that enables AI agents to interact with interactive terminal user interface (TUI) and command-line interface (CLI) programs. It solves the problem of agents being unable to use tools that require interactive input, by providing a headless interface. This is useful for developers and AI enthusiasts who want to automate tasks involving such programs.
The Limoni terminal UI engine for Go now exposes its accessibility-style semantic tree to AI agents via a Model Context Protocol server, letting agents target widgets by role and label instead of scraping character grids. The integration includes secret-field redaction and layered access controls, and doubles as a Playwright-style test API.
The Limoni terminal UI engine for Go now exposes its accessibility-style semantic tree to AI agents via a Model Context Protocol server, letting agents target widgets by role and label instead of scraping character grids. The integration includes secret-field redaction and layered access controls, and doubles as a Playwright-style test API.
At AI Engineer World's Fair 2026, Airbyte software engineer Pedro Lopez presented lessons from building two agent interfaces over its Context Store, a search-optimized index of tools like Zendesk, Stripe, and HubSpot. He argues MCP servers and CLIs each fit different use cases, with concrete design rules for both. The talk offers practitioners a practical framework for exposing company data to AI agents.
At AI Engineer World's Fair 2026, Airbyte software engineer Pedro Lopez presented lessons from building two agent interfaces over its Context Store, a search-optimized index of tools like Zendesk, Stripe, and HubSpot. He argues MCP servers and CLIs each fit different use cases, with concrete design rules for both. The talk offers practitioners a practical framework for exposing company data to AI agents.