@farmostwood > For an AI agent
so that's why many agentic models prefer the pattern of parallel tool calls: await all results in a while-loop then join all files/results/prompts into a single request
the second workflow is smoother for my daily tasks. I can't tell if it's a pattern from model or a context issue, because it’s totally ignoring the specific rule in my agents dot md
idk why but gemini in android always triggers a full gradle build (with a wrong task that takes minutes 😅) after edits, while gpt or opus in cursor simply read the lints and leave the changes to me for verifying