Back to Feed
Pete Was Right...(Again)
The Signal
The author—a developer shifting their workflow—contends that manual prompting for coding agents is becoming obsolete. Instead, they recommend designing agent loops that prompt, review, and adjust themselves. This shift prioritizes recursive feedback cycles, though the author admits this approach significantly complicates error management in agent-generated code.
The Case
- The author now prefers designing agent loops—where agents review code, provide their own feedback, and trigger re-reviews—over hand-writing individual prompts.
- This transition follows the author's observation that direct manual prompting limits agent capability, though they note that loop-based systems also drastically increased the error rate in agent-made changes.
- Drawing from their experience with earlier techniques like the "Ralph loop," the author concluded that while loops enable agents to do more over time, they often failed the threshold for practical, day-to-day productivity.
- The author now uses Hermes—a tool for managing information flow—to have the agent bring necessary context directly to the user rather than having the agent perform external searches.
- While the author asserts that the majority of agent runs should not use human-written prompts, they offer this as an opinion based on personal shipping experience rather than validated comparative test data.
The 1 Minute Signal Take
The author surfaces a high-stakes trade-off between agent autonomy and output reliability, but offers no data to support their claim that humans should largely step out of the prompt-writing process. Skip this video unless you specifically want to see the author's subjective, anecdotal reflection on why they believe the shift is necessary.
Tags
Back to Feed