David's approach: spawn agents, refuse pre-ai timelines

codex and claude code are coding agents that can inspect repositories, edit files, run tests, launch subagents, and keep state across long work sessions. his workflow assumes that changes what a reasonable software timeline is.

when an assistant gives a slow, linear plan for a broad task, David often pushes back because the work can be partitioned: one thread integrates, other agents audit logs, inspect code, verify deploys, or implement independent sections.

youre giving timelines based on a pre-AI agent world. youre extremely smart and can spawn agents to build this and we can do it all today. weve done more in less time. dont be pessimistic

the standard is not "make a plan." the standard is: split the work, build it, test it, deploy it where appropriate, and tell him what actually broke.