Matthew Berman demonstrates a hands-on Codex workflow including model selection, threads, agents.md review, browser use, skills, loops, remote, hooks, and access, with reusable prompts and a loop library.

Using Codex with Zapier MCP and custom hooks allows building agent systems with browser use, skills, loops, and remote access.

Agent loops can be categorized into four types (turn-based, goal-based, time-based, proactive) each suited to different task characteristics.

Agent loops can be categorized into four types (turn-based, goal-based, time-based, proactive) each suited to different task characteristics.

Using goal, loop, and schedule commands, AI agents can run for hours autonomously with three useful loops.

Agent loops can be categorized into turn-based, goal-based, time-based, and proactive, each suited for different task types.

Matthew Berman demonstrates using Codex with Zapier MCP to build automated AI workflows, including model selection, threads, agents, and loops.

Agent loops can be classified into four types (turn-based, goal-based, time-based, proactive) each suited for different task characteristics.

Agent loops have five hidden decisions per turn; bugs live in the first four, not the final output.

Agent loops can be categorized into four types—turn-based, goal-based, time-based, proactive—each suited to different task characteristics.

Agent loops can be categorized into turn-based, goal-based, time-based, and proactive, each suited for different task types.

Agent loops can be categorized into four types—turn-based, goal-based, time-based, proactive—each suited to different task characteristics.