The model is 20% of the result; the agent harness is 80%
Quote
it's like the model is like 20% of the issue. The agent harness is the other 80%. 80 percent, and most of that boils down to intelligent use of the context window.
Episode 260: Tennessee Trickshot (episode page, opens in a new tab) 00:38:41 (play audio from this timestamp, opens in a new tab)
Voice-matched
Summary
With coding agents, the model matters far less than the harness around it, and most of the harness comes down to intelligent use of the context window. Assume every prompt starts brand new: the model doesn't know you. Dave expects slow reasoning models to give way to surgical use of context handed to sub-agents (he compares it to Ender forgetting up and down in Ender's Game's battle room).
Also said
Episode 260: Tennessee Trickshot (episode page, opens in a new tab) 00:35:41 (play audio from this timestamp, opens in a new tab)
Well, reasoning models are incredibly slow. and instead of taking a dense model and putting layers or basically rounds of reasoning into it which slow everything down and don't actually give you a huge return
Voice-matched